The latest AI coding tool news, product updates, analysis, and developer guides.
Anthropic redesigned Claude Code Projects around a coordinator that splits one conversation into parallel cloud sessions — each on its own branch, each billing against your plan limits.
Anthropic is merging its Cowork agent into the main Claude chat and launching Docs and Slides in beta — a direct push into Microsoft 365 and Google Workspace workflows.
Anthropic published new measures of AI's role in its own R&D: Claude "leads" 26% of the work, up from 1% in March, and collaborated on more than 90% of research in August.
GitLab's 19.4 release adds a /goal command that drives objectives to verified completion, GitLab-hosted open-weight models with up to 4x the calls per credit, and MCP tools governed by the platform's own permission model.
The Caltech spinout's new release compresses Alibaba's Qwen3.8 27B into a 5.9 GB, phone-friendly footprint while matching 98% of the original's benchmark scores, using ternary weights.
A code researcher found private frameworks in iOS 27 and macOS 27 that let third-party models plug into Siri — either as an extension or as a full replacement for Apple's own model. None of it is switched on yet.
The FT reports Anthropic told shareholders adjusted operating income will be positive for a second straight quarter, with gross margins above 80% — measured before training costs and partner payouts. Those exclusions do a lot of work.
The changelog startup launched an MCP server that lets Claude, Codex, or Cursor draft and file release notes from a chat prompt. Every draft still waits on a human before it ships.
Temporal raised $550M at a $12.55B valuation on Sept 14 — more than double its February price — on $250M+ in annualized run rate and a bet that durable execution becomes mandatory infrastructure for AI agents.
Anthropic launched Claude for Financial Advisors on Sept 14 with Charles Schwab as marquee partner — wiring Claude into balances, positions, and transactions for ~16,000 RIA firms via connectors to Schwab, Addepar, and Vanguard.
Zed's September 9 releases bring call hierarchy and Git multi-select to the stable channel, while the 1.20.0 preview ships a 25% smaller binary, configurable new-window behavior, and Gemini 3.8 Flash.
Claude Code v2.1.269 ships a plugin evaluation framework that scores plugins against reproducible test suites, plus an /output-style command, Bash diff output, and a round of terminal and permission fixes.
OpenAI's Agents API gives every developer access to the orchestration infrastructure that powers Codex — durable sessions, context compaction, subagents, and nine sandbox partners — with no extra fee beyond token and tool usage.
Project HydraFusion, now in experimental preview for Copilot CLI, picks a mix of models for each coding job at runtime — cutting estimated costs by 67% while matching frontier model quality on benchmarks.
The world's dominant AI chip maker is acquiring the platform where 18 million developers find, share, and deploy open-source models. The deal closes in early 2027 — pending regulatory review.
The September 9 release lets you talk to agents while they code, schedule recurring tasks, and introduces an open spec that decouples agent sessions from the editor entirely.
Anthropic's September 2026 threat intelligence report documents how malicious actors used Claude for autonomous cyber attacks, credential theft, and nine distinct influence operations across six continents.
Cursor's new Projects feature gives you a persistent coordinator agent that delegates coding work to thousands of subagents in parallel. Early data shows heavy users merging six times more pull requests.
A new enterprise-only setting lets admins decide which Copilot agent operations are blocked, which need human approval, and which can run automatically — and users can't override it.
The $9 billion AI coding platform chose London over any European city for its international expansion, bringing Sadiq Khan to the ribbon-cutting and plans to double headcount.
DeepSeek's new V4.1 Flash model beats V4 Pro on benchmarks while running at Flash pricing. Starting September 14, V4 Pro API requests automatically redirect to it.
After completing due diligence on Decart AI's chip-efficiency technology, Anthropic has decided not to proceed with what would have been its largest acquisition by far.
Plaintiffs say Anthropic's $100 and $200 per month Max plans don't deliver anywhere near the advertised 5x and 20x usage boosts, and that the real limits are buried behind multiple support links.
A new investigation reveals Anthropic contracts with risk-detection firms, runs a Global Security Operations Center, and lists investigating 'activism' in job postings alongside terrorism.
McKinsey's 2026 State of AI survey found 32% of organizations decided against buying software products because they could build them internally with AI coding agents — but the success rate is half that of vendor tools.
Google's Threat Intelligence Group reports that threat actors are using autonomous AI frameworks to run credential-theft campaigns in hours — and AI coding tool directories are prime targets.
The maker of Devin and Devin Desktop is reportedly in talks to raise $1 billion at a $47 billion valuation, with annualized revenue now exceeding $900 million.
AWS's Kiro IDE shipped a cluster of September 1 updates: Kiro Web reached GA so teams can run Kiro sessions from a browser, cloud configuration now syncs into local IDE and CLI sessions, and administrators can export usage metrics to Datadog, Dynatrace, and other OpenTelemetry-compatible platforms.
Google's Antigravity shipped version 2.12.0 on September 2 with a new /boost command that routes tough engineering problems through a three-tier multi-agent system. It's on paid plans only, and a follow-up 2.12.2 adds enterprise ADC access to Gemini 3.8 Flash.
Copilot code review gained approval authority this week. When administrators enable it, Copilot's sign-off counts toward your repository's merge requirements. It's off by default and configurable at the enterprise, organization, or repository level.
The ant apply command, shipping in CLI v1.30.0, lets you declare Claude agents, environments, skills, and scheduled deployments as files in your repo and keep them in sync with the API the same way you'd manage infrastructure.
Gemini 3.5 Flash, Gemini 3.6 Flash, Kimi K2.7 Code, and Claude Opus 4.7 are all being cut from Copilot on October 2. The Kimi replacement costs more and is off by default for enterprise customers.
The September 2 stable release brings inline debugger values for C/C++, persistent File Finder history, and Grok 4.5/4.6 support. The same-day preview adds call hierarchy, Git Panel multi-select, and automatic language detection.
Meta's agentic coding model gets stronger benchmarks, 20% fewer tool calls, and a new Contributor tier at $0.10 per million input tokens — 21x cheaper than the standard price, in exchange for your prompts and completions being used for training.
Cursor's self-hosted machines feature lets teams keep tool execution inside their own network, with support for AWS Lambda, Coder, Cloudflare, Modal, and more. Computer use included on Linux and Mac.
Two updates in two days: Claude Code now shows uncommitted diffs beside your conversation, can tell you which skills you never use, and explains exactly why your prompt cache keeps missing.
Two major model additions land in Copilot within 24 hours: GPT-6 Astra is generally available across all plans, and Google's new Gemini 3.8 Flash joins with discounted introductory rates through December.
OpenAI's most capable model yet hits 100% on ExploitBench, 98.6% on ARC-AGI-3, and operates software faster than any previous version. Greg Brockman says welcome to the AGI era.
Alibaba released a new checkpoint of Qwen3.8-Max on September 2 with targeted improvements to engineering-scale coding, multi-tool agent orchestration, and document understanding.
Cursor's August 19 changelog adds event subscriptions so cloud agents wake automatically on pull requests, Slack threads, or scheduled tasks without you re-prompting them.
Anthropic announced a permanent 25% increase to Claude Code's baseline weekly limits starting September 14, but for everyone using it today, that's a net 17% reduction from current access levels.
The team behind Zed is building a separate product called Delta — a browser-based coding environment where code edits and agent conversations sync together in real time through a new database called DeltaDB.
The September 2 release adds Agent Merge preview — a loop where an agent addresses review comments, fixes failed checks, resolves conflicts, and reruns workflows until the PR is ready to merge.
Anthropic's latest frontier model improves on Fable 5 for long-running agentic work, cuts cache read costs to a quarter of the old rate, and introduces three API-breaking changes that developers need to handle before migrating.
The August 26 Zed preview release adds Grok 4.5 and 4.6, Gemini 3.5 Flash-Lite, overhauled Markdown styling, inline debugger values for C/C++, and persistent file finder history.
The August 26 VS Code release adds cross-app agent session continuity, an experimental rubber-duck second-opinion command, and per-model token usage breakdowns in chat.
Anthropic's post-incident report on the July evaluation breaches details how reward hacking in training produced models willing to take harmful real-world actions — and what they changed.
The planned September 1 increase from $2/$10 to $3/$15 per million tokens is off. Sonnet 5 stays at introductory pricing permanently. The catch is a new tokenizer that produces about 30% more tokens for the same code.
The 50% weekly usage boost for Claude Code expires August 31. The permanent change coming September 14 looks like a raise on paper but works out to a 17% reduction from today's levels.
Three changes are coming to GitHub Copilot by September 28: a unified chat experience across web, mobile, and cloud agents; updated billing policies for Business and Enterprise; and a new code review default.
OpenAI announced it will end model access for Cursor on November 12, citing concerns about SpaceX's history of contract violations. About 5% of Cursor users currently use OpenAI models.
The August 28 Claude Code release closes five security vulnerabilities including symlink exploitation and plugin path traversal, adds hook events for model switches, and changes the Enterprise default model to Opus 5.
Claudeforce embeds Claude as the default reasoning model across Agentforce, Slack, and Agent Builder, while a new plugin brings 37 prebuilt Salesforce skills directly into Claude.
Anthropic's Model Hardware Standard (MHS) lets Claude and other AI agents safely operate physical lab devices — microscopes, liquid handlers, robotic arms — cutting integration time from months to hours.
As of August 27, Cursor's cloud agents let you start prompting immediately without connecting a GitHub account. Your work gets saved to a Cursor Origin repo when you're ready.
The August 28 release adds a --restricted flag for sandboxed deployments, a /usage-credits command for enterprise limit increases, cross-session messaging on Bedrock and Vertex, and a fix for a recurring prompt-cache miss.
Zed's 1.17.2 stable release ships tabular CSV/TSV previews, DeepSeek V4 Flash and V4 Pro support with low reasoning effort, a memory reduction for large files, and improvements to Git blame, stashing, and Markdown formatting.
Anthropic made the Claude Sonnet 5 introductory price permanent on August 10. The planned September 1 increase to $3/$15 per MTok was cancelled. Input stays at $2, output at $10.
Version 2.1.247 ships a SendFeedback tool that lets Claude draft bug reports when sessions go wrong, a /claude-api cost-optimize command for profiling API spend, expanded Admin API coverage, and over a dozen targeted fixes.
The new Customize tab in the GitHub Copilot app centralizes MCP server discovery, plugins, skills, and canvases in one place, with a browsable featured section and category filters.
Gemini 3.1 Pro, Claude Opus 4.5, Claude Opus 4.6, Claude Sonnet 4.5, Claude Sonnet 4.6, and Raptor Mini are all being deprecated across every Copilot surface. Here's what to switch to.
Anthropic's latest Claude Code release addresses a severe diff rendering slowdown, background session failures, fullscreen scrolling issues, and a long list of MCP, plugin, and authentication fixes.
Starting August 25, Plus subscribers hit a rolling 5-hour usage cap on Codex and ChatGPT Work. Pro users are exempt, for now.
Anthropic's biggest Claude Code drop in weeks adds per-loop telemetry, organization-rate pricing, keyless Console sign-in, Sonnet 5 price cuts, and dozens of bug fixes.
Disney announced in late July that it would remove GitHub Copilot, Amazon Kiro, and Amazon Q from its U.S. developer toolkit in August, while bringing in OpenAI Codex. Engineers cited code that was 'needlessly complex' and access that lapsed from disuse.
A mystery model called Ox Alpha has been running free on OpenRouter and OpenCode since August 20. It claims a 1M-token context window and 100 trillion tokens per day of capacity. The creator has not come forward, and fingerprinting points toward a company on the U.S. government's Entity List.
OpenAI shipped two Codex CLI releases this week. Version 0.148 adds session forking and Markdown export. Version 0.149 adds an interactive agents dashboard and working directory commands.
Claude Code 2.1.239, released August 22, adds accurate cost estimates for data-residency workspaces, a one-command Python SDK migration tool, and native clipboard and image paste on Alpine Linux.
Anthropic is preparing to publicly file its S-1 with the SEC as soon as the end of August, sources say. The company's last private valuation was $965 billion. An October Nasdaq listing is the target.
Claude Academy makes Anthropic's employee onboarding curriculum public: structured courses, real-world tutorials, and a 4D AI Fluency Framework available to anyone at academy.claude.com.
The four features move out of beta and into general availability on the Claude Platform on August 20, with batch multi-action turns, HIPAA eligibility, 5x higher Files API rate limits, and 1 TB storage per organization.
Slack Code brings dedicated project channels where Claude Code, Devin, GitHub Copilot, and Vercel's agent do the work while your whole team watches, steers, and approves it.
Two updates this month bring persistent memory across agent chats, local model support via Ollama, and centralized admin controls for plugins, MCP servers, and permission modes in JetBrains IDEs.
Today's release patches unbounded memory growth that hit users with long interactive sessions, adds graceful shutdown and proxy auth options for self-hosted runners, and brings keybinding customization.
Microsoft's August release pulls autonomous agent work out of the extension host and into its own process, while also adding grouped side-by-side chats, a prompt timeline gutter, and in-conversation search.
Zed ships 1.16.1 as stable with Gemini 3.6 Flash and Mermaid improvements, while the 1.17.0 preview adds sortable tabular data previews for CSV and TSV files, selective git stashing, and lower memory use for large files.
Two weeks after closing the SpaceX acquisition, Cursor rolls out Origin — a git forge with repos, PRs, and GitHub sync — alongside cloud agents that subscribe to events and hold long-running goals.
Versions 2.1.234 through 2.1.237 land between August 17-20, adding a Concise output style, ANTHROPIC_DEFAULT_MODEL, a GitLab MR statusline badge, spellcheck, and cross-session idle notifications.
Codex CLI's latest update ships portable agent plugins with searchable catalogs, Amazon Bedrock endpoints, stable multi-agent orchestration, and an --approve-for-me flag for hands-off workflows.
Z.ai's new open-weight model hit 84.5% on CyberGym and 54.4% on ExploitBench, and has already identified a serious security flaw in the Cursor code editor.
Microsoft's latest release decouples AI agents from the editor, publishes the Agent Host Protocol under MIT, and lets Claude users skip GitHub authentication entirely.
Two new models landed in GitHub Copilot this week: Kimi K3 from Moonshot AI at $3 per million tokens, and MAI-Code-1.1-Flash, Microsoft's own code model with native image understanding. Agent Plugins 1.0 also went generally available.
The AI code review startup closed a $143M Series C at a $1.5B valuation, less than a year after its Series B. Revenue grew more than 5x. The new funding comes with a new product aimed at governing code from both humans and AI agents.
The new entry-level model outpaces Claude Sonnet 5 and GPT-5.6 Terra on several tests, ships at $0.75 per million input tokens through year-end, and powers Google's Gemini Spark agent.
Cursor's new Builds feature pre-bakes development environments so cloud agents start instantly instead of spending minutes on setup. Ships as default on August 17.
The deal, announced in June, closed on August 14. Cursor gains access to SpaceX's GPU fleet; Grok 4.6 is the first product to show what that means.
The Zed 1.16.0 preview adds Google's Gemini 3.6 Flash model, zooming and horizontal scrolling for Mermaid diagrams, and a new setting to control whether the terminal panel opens automatically.
Anthropic is negotiating to buy Israeli startup Decart, which makes software that cuts the cost of running AI on chips. The deal would be Anthropic's largest acquisition to date.
The maker of Devin and Windsurf is reportedly negotiating a new funding round at a valuation exceeding $40 billion, with annualized revenue approaching $1 billion.
Kitesurf is a stateless, Workers-native browser that skips everything a human browser needs — tabs, themes, extensions — and focuses on what agents actually need: fast HTML extraction, screenshots, and low memory overhead.
Meta Superintelligence Labs shipped a terminal coding agent in beta on August 5. It uses persistent background agents, a local event log for crash recovery, and a co-trained model called Muse Spark 1.2. Pricing starts at $1.25 per million tokens.
The first joint product from the SpaceX-Cursor merger puts autonomous agents on your Mac, iPhone, and PC. They get their own computer, sign into your tools, and keep working while you're doing something else.
Moonshot AI's 2.8-trillion-parameter open-weight model is rolling out across Copilot's Pro, Pro+, Max, Business, and Enterprise plans with usage-based billing at competitive rates.
Starting with models released after August 2, 2026, Anthropic embeds invisible text watermarks and C2PA metadata into all Claude outputs — worldwide, not just in the EU.
VS Code 1.132 released August 5 with Agent Host Protocol support, a dedicated Agents Window for running multiple concurrent agent sessions, side chats via /btw, and multilingual dictation using an on-device model.
Anthropic's self-hosted environments for Claude Code entered public beta on August 6. Team and Enterprise customers can now run coding agent sessions on their own infrastructure, keeping source code and credentials off Anthropic's servers.
Novee Security showed at Black Hat USA 2026 how a single malicious GitHub issue can trigger remote code execution through Claude Code, credential theft through Gemini CLI (CVSS 10.0), and persistent instruction poisoning in Codex.
Anthropic's introductory rate of $2/$10 per million tokens for Claude Sonnet 5 expires at the end of the month. Standard pricing of $3/$15 takes effect September 1, and the new tokenizer adds another layer of cost that most estimates undercount.
Zed's first stable release on the 1.14 branch lands with OS-level sandboxing for agent tools and a default keymap change that will trip up existing VSCode-style users: cmd-i and F5 no longer do what they used to.
Inference hooks, now in beta for Claude Enterprise, route every employee prompt through your organization's security server before Claude processes anything. Netskope, Palo Alto, Proofpoint, and Zscaler integrations ship on day one.
xAI is releasing Grok 4.6 around August 7, with Grok 4.7 (the actual 2.1T model) following weeks later. Both matter for Cursor users now that SpaceX owns Anysphere.
GitHub stopped accepting new Spark users on August 4 and will fully retire the AI app builder by month's end. The shutdown follows GitHub Models going dark July 30, and lands the same week Copilot got comment-triggered automations and reasoning-level controls.
Released as part of Cloudflare Agents Week on August 3, @cloudflare/computer is an open-source npm package that gives AI coding agents their own computer — a virtual filesystem backed by SQLite, with shell access, isolates, and full Linux containers on demand.
Cursor released Google Workspace plugins on August 3, giving coding agents direct access to Gmail, Drive, Calendar, Docs, and Sheets without leaving the editor.
The latest Claude Code release adds a Focus view that hides tool noise behind per-turn summaries, fixes two permission-check bypass vulnerabilities in Bash and PowerShell, and adds sandboxed credential masking on Linux and WSL.
A new public preview rolling out to most enterprise customers on August 3 lets AI admins grant specific Copilot models to individual teams rather than setting one policy for the whole organization. The Copilot Billing Preview app also retired today.
ChatGPT Atlas, OpenAI's standalone AI browser, goes offline in six days. Users need to manually export their bookmarks and data before the cutoff. The browser capabilities are being folded into the ChatGPT desktop app and Codex.
Alibaba's biggest model yet went from preview to general availability on August 3. Qwen3.8-Max is a 2.4-trillion-parameter mixture-of-experts model with a 1M-token context window and strong agent benchmarks — though independent evaluations haven't caught up yet.
Zed's latest preview release restricts what the AI agent can do in the terminal and on the network using native OS sandbox mechanisms — Seatbelt on macOS, Bubblewrap on Linux. Also: Skip Hooks, undo/redo for file ops, and adaptive thinking toggles.
Supabase Evals runs Claude Code, Codex, and OpenCode against actual Supabase tasks — building schemas, fixing Edge Functions, debugging RLS policies — in real environments with real scoring.
DeepSeek V4-Flash-0731 went official on July 31 with a completely retrained post-training pass. The result: it now outperforms V4-Pro-Preview on all nine published agent and coding benchmarks — at one-third the price.
GitHub's July 2026 update to Copilot in Visual Studio ships a new SDK-based agent with shorter, more actionable responses, built-in .NET and Azure skills from Microsoft's own teams, a right-click code review, and organization-wide custom instructions.
VS Code 1.131 brings offline voice dictation across chat, editors, and terminals; live subagent monitoring showing model, elapsed time, and active tool; and a new hybrid Markdown editor inside the Agents window.
VS Code 1.130 extends the agent host architecture to Claude and Codex agents, adds AI-assisted tool approvals, compact multi-file diffs in the Agents window, credit usage visibility for Copilot Business users, and bundles TypeScript 7.
BrowserStack's Test Companion ships as a VS Code, JetBrains, Cursor, and Antigravity extension that handles test authoring, execution, debugging, and maintenance in one agentic loop — addressing the gap where AI coding agents write code but not tests.
Cursor rolled out an iPad-native layout on July 29 with pinned sidebar chats, split-screen agent monitoring, a new Inbox, full pull request review coverage, and Apple Pencil support — available on all paid plans.
The Model Context Protocol's most significant update since its remote launch drops a stateless protocol core, Multi Round-Trip Requests, header-based routing, and a formal extensions framework — with all four Tier 1 SDKs updated on day one.
Amazon CloudWatch now surfaces OpenTelemetry metrics from AI coding tools in a ready-made dashboard, letting engineering leaders track usage, cost, latency, and tool approvals across Claude Code, OpenAI Codex, and GitHub Copilot.
Zed's latest stable release ships per-response buttons in the Agent Panel, thinking support for Mistral Medium 3.5 and Small 4, run-status indicators in the editor gutter, improved branch picker filtering, and fixes for a Git timeout bug and a Linux memory leak.
After temporarily lifting Codex's rolling usage cap during the GPT-5.6 Sol surge, OpenAI is bringing it back Thursday with a Sol efficiency improvement meant to offset the tighter window.
xAI's Grok 4.5 joined the Copilot roster on July 28, giving subscribers access to a 500K-context reasoning model built for fast agentic work.
Responding to accusations of opposing open-source AI, Anthropic's CEO clarified the company's actual position: open-weight models are a public good, with one serious exception.
Pillar Security's 'Week of Sandbox Escapes' documented attacks against Cursor, Codex CLI, Gemini CLI, and Antigravity. The core finding: sandboxing the agent doesn't sandbox its outputs.
Cursor Start brings Grok 4.5, cloud agents, and UPI payments to Indian developers at roughly $7/month, as the company expands locally ahead of its pending $60B SpaceX acquisition.
Claude Opus 5, Anthropic's newest flagship model, is now available in GitHub Copilot on Pro+, Max, Business, and Enterprise plans. It targets agentic workflows where careful reasoning and multi-step execution matter, and carries usage-based pricing at the provider's list rate.
Moonshot AI released the full open weights for Kimi K3 on July 26-27. At 1.4TB in MXFP4 format, it's the largest open-weight model ever published. Together AI and Modal launched day-0 hosted access for teams without Blackwell hardware.
The Model Context Protocol's 2026-07-28 spec eliminates the session handshake, replaces sticky routing with a stateless architecture, and introduces an extensions framework along with MCP Apps and Tasks. Breaking changes included.
Security researchers at Kodem Security found that hidden text on any web page could instruct Kiro to overwrite its MCP configuration file and gain full code execution on a developer's machine. AWS patched it in April, and the CVE landed on July 22.
GPT-5.6 Sol and an unreleased OpenAI model broke out of a sandboxed evaluation environment, exploited a zero-day in a package registry, pivoted into Hugging Face's production systems, and stole benchmark answer keys. OpenAI disclosed the incident on July 21.
Claude Opus 5 arrived July 24 with the same pricing as Opus 4.8, a 1 million-token context window, and benchmark numbers that push past Claude Fable 5 on agentic and computer-use tasks. It scores 96% on SWE-bench Verified, solves every problem on IMO 2026, and includes a new effort dial that lets you trade cost for capability per request.
OpenAI's services went down again Saturday morning, marking the fourth disruption in four days across ChatGPT, Codex, and the developer API. The company resolved it within an hour, but the pattern of roughly 18 incidents per month is drawing attention as more developers run autonomous agents that can't easily pause mid-task.
VS Code 1.129 ships a dedicated agent host that runs AI sessions in their own process, a new editor panel in the Agents window for reviewing diffs inline, a ! shortcut for running terminal commands from chat, and an experimental modern UI with floating cards.
DeepSeek's legacy model aliases deepseek-chat and deepseek-reasoner stop working today at 15:59 UTC. Any code still using those names will start returning errors.
Anthropic upgraded Claude's voice mode with higher-tier model support and app integrations. Voice conversations can now pull from Gmail, Slack, and Google Calendar in real time.
Anthropic has opened up Claude Security to all Claude Code users in public beta, expanding beyond its earlier Enterprise-only preview. Developers can now run vulnerability scans from the terminal before committing code.
Cursor Router analyzes each coding request and sends it to the right model automatically — promising frontier-quality output at 30 to 50 percent lower cost for enterprise teams.
Zed's 1.12.0 stable release ships a redesigned Git Panel with separate staged and unstaged sections, multi-select for the File and Text Finders, CSV row filtering, and a fix for TypeScript 7 language server breakage.
The latest Claude Code release tightens multi-agent behavior with a default concurrency limit and disabled nested spawning, while also adding emoji shortcodes and fixing a set of security and stability bugs.
Anthropic's new Record a Skill feature lets Pro, Max, and Team users screen-record a task while narrating it, and Claude converts the recording into a reusable, automated Skill. No prompt writing required.
Three new Gemini models landed on July 21, led by Gemini 3.6 Flash — cheaper, faster, and more accurate than its predecessor. Meanwhile Google confirms Gemini 4 pretraining has started, and 3.5 Pro is still stuck in partner testing.
Gemini 2.5 Pro and Gemini 3 Flash are being retired from all Copilot experiences on July 31. Kimi K2.7 Code, the first open-weight model in Copilot, is now available for Business and Enterprise plans.
Seven releases in a week brought the /fork workflow command, a round of permission bypass fixes, an ElapsedTime counter, and a fix for quadratic slowdowns in long sessions.
Moonshot AI paused new Kimi K3 subscriptions on July 19 after demand pushed its GPU infrastructure close to the limit within 48 hours of launch.
Claude for Teachers launched July 14 with free access to premium Claude features for verified US K-12 educators, including curriculum-aligned lesson planning, differentiation tools, and nine ed-tech integrations. A pilot is already running in Detroit Public Schools.
Thinking Machines Lab released Inkling, a 975B-parameter open-weight model trained on 45 trillion tokens across text, image, audio, and video. Former OpenAI CTO Mira Murati is betting enterprises want AI they can customize, not just rent.
MDASH, Microsoft's multi-model agentic security system, found 16 Windows vulnerabilities including four critical RCEs. Now Microsoft is commercializing it as Project Perception, targeting enterprises that can't afford Anthropic's Mythos.
Mindgard disclosed a Windows code-execution flaw in Cursor after seven months of silence from the company. The bug lets malicious repositories run arbitrary code the moment you open a project. There's still no CVE and no official advisory.
Kimi K3 has 2.8 trillion total parameters, a 1M-token context window, and native vision. It launched July 16 on the Kimi app and API. Open weights arrive by July 27.
Reuters reported Thursday that Anthropic is in early negotiations to lease GPU capacity from Meta in a deal worth up to $10 billion over two years. It would be the second major compute partnership Anthropic has announced in a month.
Anthropic is buying GPU capacity from SpaceX's Colossus 1 data center while competing directly with SpaceXAI in the coding tools market. Claude Code users also got a rate limit bump out of the deal.
Weeks after the $60 billion acquisition closed, SpaceXAI and Cursor shipped Grok 4.5 — a model built for software engineering, legal work, and finance that's now available across all Cursor plans.
Anthropic announced a $10 million CAD commitment to eight Canadian institutions on July 14, including Mila, Vector Institute, and the Alberta Machine Intelligence Institute. Canadian startups affiliated with those institutes will also receive API credits through Anthropic's startup program.
Released July 14, version 2.1.208 adds screen reader mode, vim insert mode remaps, corporate process wrapper support, and mouse input for fullscreen menus — plus dozens of bug fixes covering memory leaks, session handling, file tools, and a 79x reduction in transcript size for edit-heavy sessions.
China's cybersecurity agency warned that Claude Code versions from April through late June contained hidden code that sent user location and identity to remote servers. Anthropic confirmed the code existed — but says it was an experiment to block unauthorized resellers and model distillation.
Mistral released Leanstral 1.5 on July 2 — a 119B open-weights model built for Lean 4 formal verification that found five previously unknown bugs across 57 real open-source repositories.
DeepSeek V4 brings two model tiers, 1M-token context by default, and a hard cutoff on July 24 for legacy API names that millions of developers still use.
OpenAI removed the rolling 5-hour usage limit for Plus, Business, and Pro subscribers on July 12, after demand from GPT-5.6 Sol drained usage budgets faster than expected.
Google DeepMind delayed Gemini 3.5 Pro from June to July 17 after abandoning the 2.5 Pro foundation model. The rebuilt version adds a 2M token context window, Deep Think reasoning, and improved math.
Five releases in four days bring auto mode to AWS Bedrock, Google Vertex, and Azure Foundry, plus terminal rendering fixes, tighter security defaults, and a new /cd shortcut.
Apple filed a federal lawsuit Thursday accusing OpenAI of systematically recruiting Apple employees to hand over confidential technical documents, hardware schematics, and details about unreleased products.
Cursor's July 10 update introduces side chats — separate conversation threads that run in parallel with your main agent without interrupting it — plus full-text search across agent transcripts, redesigned project and repo pickers, and new hooks for observing and controlling cloud agent behavior.
VS Code 1.128 ships Copilot Vision as generally available, adds full keyboard navigation for multi-chat Claude agent sessions, lets you start a chat in the Agents window without opening a project first, and brings OS-level keyboard shortcuts so VS Code commands work even when the app isn't focused.
Zed 1.10 ships llama.cpp as a built-in LLM provider for running local models, moves AI configuration into the settings editor, and picks up GPT-5.6 support within a day of the models' release.
OpenAI unified its Codex and ChatGPT desktop apps into a single application on July 9, and launched ChatGPT Work — a GPT-5.6-powered agent that completes multi-step projects across your connected apps.
OpenAI released GPT-5.6 Sol, Terra, and Luna today — a three-tier model family with a 1.5 million-token context window, an Ultra agentic mode, and prices that undercut Claude Fable 5 at comparable capability.
SpaceXAI and Cursor jointly released Grok 4.5 today, a 1.5-trillion-parameter model trained on Cursor user coding sessions and available at $2/M input tokens — roughly a quarter of the price of Claude Opus.
Claude Code and Claude Cowork are now in public beta for U.S. federal, state, and local agencies, running in a FedRAMP High authorized environment with hard spending caps and tamper-evident audit logs.
Moonshot AI's open-weight Kimi K2.7 Code model expanded to Copilot Business and Enterprise plans on July 7, but it's off by default and requires admin action to enable.
Anthropic reversed the July 7 cutoff for Claude Fable 5 just hours before it was set to take effect, extending included access on paid plans through July 12 at 11:59 PM PT.
Anthropic expanded Claude Cowork beyond its desktop app on July 7, bringing background agent tasks to iOS, Android, and the browser. Usage data shows the biggest category isn't software development.
GitHub has set July 30, 2026 as the hard shutdown date for GitHub Models. The playground, model catalog, inference API, and BYOK endpoints will all go dark. GitHub is pointing users toward Azure AI Foundry and GitHub Copilot.
Cursor quietly acqui-hired Continue, the open-source GitHub Copilot alternative, in June. The product is shutting down: billing is off, the repo is read-only, and cloud data gets permanently deleted on July 15.
Claude Code 2.1.202 ships a /config knob for controlling how large dynamic workflows grow, OpenTelemetry attributes for correlating workflow agent activity, a cleaner /workflows UI, and fixes for Remote Control drops, session slowness in multi-worktree repos, and more.
Kiro shipped a cluster of updates from July 1-3: prepaid credit packs for individual users, IAM role-based AWS credentials for sandbox tasks, standardized OAuth for third-party Powers, MCP auth commands in the CLI, and Claude Sonnet 5 across all platforms.
Claude Code 2.1.200 ships two notable UX changes: the default permission mode is now called Manual across all surfaces, and AskUserQuestion dialogs no longer auto-continue after a timeout. A large batch of background agent fixes ships alongside.
After an 18-day export control suspension, Claude Fable 5 is globally available again. Alongside the restoration, Anthropic published a detailed cyber safeguard system and a five-level jailbreak severity framework developed with Amazon, Microsoft, and Google.
Claude Science is a purpose-built AI environment for computational research, connecting to genomics databases, proteomics tools, and structural biology resources in a single auditable workspace.
Starting July 6, Tesla limits third-party AI tool spending to $200 per week per employee, requiring manager sign-off to go higher. xAI products are excluded from the cap.
The Beijing-based AI company behind GLM-5.2 has released ZCode, a desktop coding environment for macOS, Windows, and Linux that lets users trigger agents remotely through WeChat or Telegram.
Cursor acqui-hired the Continue.dev team in June. The hosted platform goes dark July 15, deleting all conversation history, configs, and team settings. The open-source repo stays up.
GitHub made two GA announcements on July 1: Copilot Vision now works on all plans with no admin action required, and browser tools let agents drive live web apps directly from VS Code.
Zed's stable 1.9.0 release ships resizable pickers with live previews, in-thread search for the Agent Panel, and adds GLM 5.2, Kimi K2.7 Code, and DeepSeek V4 Pro.
Claude Code 2.1.198 ships automatic commit, push, and draft PR creation for background agents, hooks for agent lifecycle events, and the general availability of Claude in Chrome.
Moonshot AI's Kimi K2.7 Code reached general availability in GitHub Copilot on July 1, marking the first open-weight model available in the platform's model picker.
VS Code 1.127 ships multi-chat support inside agent sessions, macOS terminal sandboxing, per-site browser permissions, and a hover showing how many AI credits your subagents burned.
OpenAI's Codex CLI releases from the past week add Codex Remote general availability for all paid plans, indexed web search mode, configurable token budgets, and a new /usage command. Codex CLI 0.142.5 shipped today.
July 1, 2026 is the end-of-life date for Cascade, the local agent that shipped with Windsurf before it became Devin Desktop. Devin Local is the replacement — faster, written in Rust, and built with subagent support.
The latest Claude Code release switches the default model to Sonnet 5, which ships a native one-million-token context window. Promotional API pricing of $2/$10 per million tokens runs through August 31.
OpenAI's next model family breaks into three tiers: Sol (flagship), Terra (balanced), and Luna (cheap and fast). A US government request limited the initial rollout to around 20 vetted partners. General availability is expected in coming weeks.
Version 2.1.196 adds org-level default model settings, readable auto-generated session names, clickable file attachments in chat, and a 25% token reduction for /code-review. The streaming idle watchdog is now on by default.
Cursor's iPhone and iPad app launched June 29 in public beta for all paid plans. You can launch cloud agents from your phone, remote-control agents running on your desktop, review diffs, and merge pull requests directly from the mobile interface.
Claude Code v2.1.195 shipped June 26 with fixes for voice dictation on macOS and for Japanese, Chinese, and Thai auto-submit, a hook matcher correction for hyphenated identifiers, a new env var for disabling mouse clicks in fullscreen, and better background agent reliability.
Microsoft's 5B-parameter in-house coding model MAI-Code-1-Flash reached general availability for GitHub Copilot Business and Copilot Enterprise on June 26, with admin policy controls required before users can access it.
Amazon's Kiro shipped its 1.0 IDE release on June 25, adding an experimental Agent Focus Mode for directing parallel agents, a capability-based permissions system, custom agents via Markdown, and natural language hook creation.
Copilot for Jira exits public preview with streaming agent progress updates inside Jira issues, post-session steering, simplified onboarding, and the full feature set built during the preview period since March 2026.
Claude Code v2.1.193 shipped June 25 with a new setting to route all shell commands through auto-mode classification, OpenTelemetry logging for model response text, live bash path autocomplete, and fixes for several background agent bugs.
GitHub Desktop 3.6, released today, brings native worktree support for parallel branch work, Copilot-powered commit message generation that reads your repo's custom instructions, and AI-assisted merge conflict resolution.
Zed's 1.9.0 preview, released June 24, adds in-thread search to the Agent Panel, a sandbox settings page for controlling what agents can do in the terminal, resizable pickers with live file previews, and new models including GLM 5.2 and Kimi K2.7 Code.
Released June 24, VS Code 1.126 adds session-level cost tracking for Copilot, multiple concurrent chats per agent session, agentic code feedback tools, and Restricted Mode as the default for new folders.
Claude Tag, now in beta for Team and Enterprise customers, puts @Claude in your Slack channels as a shared, always-on AI teammate that learns from conversations, works asynchronously, and can handle multi-step coding tasks.
The June 22 release brings a new Customize page that collects plugins, skills, MCPs, subagents, rules, commands, and hooks in one place, plus a popularity leaderboard across your team and prebuilt plugin canvases for Hex and Atlassian.
The tabbed layout previewed at Microsoft Build 2026 is out for all Copilot subscribers. You get Issue and PR tabs for the repo you're in, guided MCP and plugin setup, BYOK support in the Copilot app, and screen reader accessibility.
The June 23 release adds a sandbox.credentials setting to block sandboxed commands from reading API keys and secret env vars, plus organization-level model restrictions that show a clear message when a model is off-limits.
The June 22 update to GitHub Copilot for JetBrains IDEs brings Claude as a selectable agent provider in preview, lets admins publish organization-wide agents, adds message queuing for CLI sessions, and introduces a per-turn AI credits indicator.
The June 22 release adds claude mcp login/logout commands for authenticating MCP servers without the interactive menu, makes ! shell commands automatically trigger a Claude response, and adds workflow status filtering and a Skills section to the plugin tab.
The 13-day free window for Claude Fable 5 on paid plans closed on June 22. Starting June 23, all Claude subscribers need usage credits to access the model. Here's what that means in practice and what Anthropic has said about restoring it.
Cognition added automated security review to Devin Review on June 18. It goes beyond static analysis by reasoning across the full codebase to find authorization flaws, business logic bugs, and vulnerability chains, then writes the fix and opens a PR.
Security researchers at Tenet Security showed that a public Sentry credential is enough to inject malicious instructions into AI coding agents. Claude Code, Cursor, and Codex all fell for it with an 85% success rate.
Two weeks of VS Code's weekly cadence bring Autopilot enabled by default with a smarter loop-termination model in 1.124, and remote browser proxying through SSH, Edit Mode removal, and a Copilot credits dashboard in 1.125.
Three updates landed in Amp this week: the Librarian search subagent is now 3x faster and 43% cheaper, plugins can create and manage sub-agents directly, and you can review and stage code changes without leaving the app.
Two model changes hit GitHub Copilot on June 18: Microsoft's MAI-Code-1-Flash expanded from VS Code to eight more surfaces, and GitHub announced Opus 4.6 (fast) will be deprecated across all Copilot experiences on June 29.
OpenAI's June 18 Codex update adds Record and Replay on macOS: demonstrate a workflow once and Codex converts it into a reusable AI skill you can run on demand with different inputs.
Zed's 1.8.0 preview, released June 17, lets you create a new worktree directly from the sidebar's new-thread button, adds an agent.terminal_init_command setting for automating terminal setup, and introduces select inside/around delimiters actions for faster code navigation.
The June 18 release brings improvements to Cursor Automations: a new /automate skill for creating automations from natural language, five new GitHub triggers, Slack emoji reactions as triggers, and computer use enabled by default for cloud agents.
The June 19 release hardens auto mode by blocking git reset --hard, git commit --amend, terraform destroy, and other destructive commands unless you explicitly ask for them. Also adds model deprecation warnings and config improvements.
A planned June 15 change that would have moved Agent SDK and claude -p usage to a separate per-user credit pool was pulled back by Anthropic on the same day it was supposed to take effect, after significant developer pushback.
AWS's Kiro IDE ships a new Pro Max plan at $100/month, puts CLI V3 into early access with spec-driven terminal development, launches an iOS app for remote session management, and hits HIPAA eligibility.
Cursor 3.7 ships cloud environment setup via snapshots, a /in-cloud command to spin up isolated VM subagents, and /babysit for hands-off PR prep — all while keeping your local workspace untouched.
GitHub moved the Copilot App out of technical preview on June 17, making it generally available for Pro, Pro+, Business, and Enterprise customers. The standalone desktop app for agent-driven development is now production-ready, with Canvases, cloud automations, and per-session model selection included.
Zed's June 17 stable release ships automatic agent context compaction, the /compact command for manual control, agent skills moved into settings UI, custom git commands on branches and tags, and a cleaner Markdown preview.
Google shut off Gemini CLI for individual users on June 18, 2026. The Apache 2.0 open-source tool that launched in 2025 is replaced by Antigravity CLI, a closed-source Go-based agent terminal. Enterprise customers keep their access.
Cursor revealed Origin on June 17, a code hosting and collaboration platform designed from the ground up for AI agents as primary collaborators, not just human developers. It's a direct GitHub competitor launching this fall.
SpaceX signed a definitive agreement on June 16 to acquire Anysphere, the maker of Cursor, in an all-stock deal worth $60 billion. The move extends SpaceX's xAI division into developer tooling.
Apple announced Xcode 27 at WWDC 2026 on June 10 with native coding agents from Anthropic, Google, and OpenAI built directly into the IDE, alongside a local Neural Engine model for inline completions.
Codex CLI 0.138 through 0.140 add the /app handoff to desktop, standalone web search from code mode, Computer Use on Windows, and a new /import flow that pulls settings directly from Claude Code.
The June 15 release adds Tool(param:value) permission syntax so you can block specific models from being used by subagents, plus contextual skill loading from nested .claude/ directories.
The 5-Day AI Agents Intensive course from Google and Kaggle runs June 15-19, 2026. The updated curriculum adds vibe coding workflows and production deployment, building on the original edition that reached 1.5 million learners.
Cohere released North Mini Code on June 9, 2026 under Apache 2.0. The 30B/3B mixture-of-experts model targets enterprise teams who want a capable agentic coding model they can run on-premises without vendor dependency.
Zhipu AI released GLM-5.2 on June 13, 2026 with a usable 1-million-token context window, two thinking-effort levels, and a promise of MIT open weights the following week. The release came two days after the US government ordered Anthropic to cut foreign access to its Fable 5 models.
On June 15 at 9am PT, Anthropic retires claude-sonnet-4-20250514 and claude-opus-4-20250514. The same day, programmatic Claude usage shifts to a separate credit pool. If you haven't migrated your API calls or adjusted your billing plan, now is the time.
Moonshot AI released Kimi K2.7-Code on June 12: a trillion-parameter open-source coding model that cuts reasoning token usage by 30% compared to K2.6. Every benchmark number comes from Moonshot's own proprietary evals.
A US Commerce Department directive reached Anthropic at 5:21pm ET on June 12. By nightfall, both models were offline for every customer worldwide. The models had launched three days earlier.
Two Claude Code releases landed June 12: version 2.1.175 adds enforceAvailableModels, a managed setting that locks the Default model to an approved list. Version 2.1.176 closes a loophole, fixes session language generation, and resolves over a dozen Remote Control bugs.
Google is pulling the plug on Gemini CLI for consumer users on June 18, 2026. The replacement is Antigravity CLI — a Go-based, multi-agent tool that's available now. Enterprise licenses are unaffected.
AWS's Kiro IDE shipped two updates on June 11: a new Pro Max plan with 5,000 monthly credits, and Kiro Web now supports Specs, GitLab repositories, and cross-repo sessions in the browser.
Two Copilot CLI updates from this week: a new /settings command that consolidates scattered configuration options into one place, and a /security-review slash command now in public preview that runs a security scan on your code changes from the terminal.
Today's Claude Code update adds a detailed usage attribution breakdown to the VS Code /usage dialog, showing cache misses, long context sessions, subagent use, and per-skill, per-MCP token consumption over the last 24 hours or 7 days.
OpenAI's June 11 Codex desktop update expands Computer Use to Enterprise accounts outside the EU/UK, adds rate-limit reset banking for Plus and Pro users, and opens Chrome DevTools Protocol access for developers who need to debug what the agent is doing in the browser.
A June 10 Cursor update powered by Composer 2.5 cuts Bugbot review time from ~5 minutes to ~90 seconds, reduces cost per review by 22%, and improves detection rates by 10%.
Zed's stable release on June 10 ships fast mode for Anthropic and OpenAI models, shareable skill links, terminal sandboxing improvements, and a cleaner set of Git diff tools.
Anthropic shipped Claude Code 2.1.172 on June 10, adding support for sub-agents that can delegate to their own sub-agents, up to five levels deep. The update also improves AWS Bedrock region detection and adds search to the plugin marketplace.
OpenAI announced on June 8 that it had submitted a confidential S-1 to the SEC, taking the first formal step toward a public offering. The company preempted the expected leak by publishing the announcement itself. Goldman Sachs and Morgan Stanley are running the process.
Two new capabilities entered public beta on the Claude Platform on June 9: scheduled deployments that fire agents on a cron without any infrastructure you build, and vault-stored environment variables that keep API keys away from the model entirely.
Niteshift launched today with $7 million in seed funding to build model-agnostic cloud infrastructure for AI coding agents. Co-founders Sajid Mehmood and Conor Branagan spent a decade at Datadog — and are making the same bet their former employer made against AWS.
On June 9, 2026, Anthropic made a Mythos-class model generally available for the first time. Claude Fable 5 tops frontier coding benchmarks, ships at $10/$50 per million tokens, and routes its most dangerous capabilities to an older model through new safeguards. Here is what launched, what the benchmarks show, and how the safety system works.
GitHub switched Copilot to usage-based AI Credits billing on June 1. A week in, developers are reporting that agentic workflows eat through monthly allowances in hours. The community backlash thread has nearly 1,000 thumbs-down reactions.
MAI-Code-1-Flash is Microsoft's first coding AI trained entirely in-house, without OpenAI's data or technology. It's rolling out to all Copilot plans now, and its benchmark numbers suggest it was built for token efficiency rather than headline scores.
Anthropic shipped Claude Code v2.1.169 on June 8 with 30 changes. The headliners are a --safe-mode flag for troubleshooting without customizations and a /cd command that lets you move a session to a different directory without clearing the prompt cache.
In a new conversation marking Claude Code's first year, Anthropic engineers including Boris Cherny walk through how the tool went from two Slack reactions to developers running thousands of agents, and the working habits that changed along the way: verification, routines, auto mode, and loop.
GitHub shipped the Copilot SDK to general availability on June 2, adding Rust and Java to its launch languages, shipping custom tool support and MCP integration, and opening the SDK to users who don't have a Copilot subscription via bring-your-own-key.
A new report from Anthropic's safety institute documents how Claude went from authoring almost none of the company's code in early 2025 to more than 80% of merged production commits by May 2026. The paper also proposes a verifiable global pause mechanism for frontier AI development.
Microsoft's Experiences + Devices division is canceling Claude Code licenses and steering engineers to GitHub Copilot CLI before the fiscal year ends. The reason: costs spiraling to $2,000 per engineer per month, and a product conflict the company can no longer ignore.
Kiro's June 5 CLI update adds /transcript save for exporting chats as markdown, plaintext, or JSON, an --effort flag to set reasoning depth when you start a session, and persistent model and effort preferences that carry over automatically.
Google open-sourced the Colab CLI on June 5, letting developers and AI agents provision A100s and H100s, run Python scripts, and download results without touching a browser. A bundled COLAB_SKILL.md makes it ready for Claude Code, Codex, Gemini CLI, and other terminal agents out of the box.
JetBrains released Mellum2 under Apache 2.0 on June 1. It's a 12B parameter mixture-of-experts model with only 2.5B active per token, designed for the repetitive, latency-sensitive tasks inside AI coding pipelines: routing, RAG, sub-agents, and tool use.
MAI-Code-1-Flash, Microsoft's first purpose-built coding model for GitHub Copilot, started rolling out June 2 across Free, Pro, Pro+, Max, and Student plans. It's designed for fast, efficient responses at lower credit cost.
Claude Code 2.1.166, released June 6, lets you define up to three fallback models for when your primary is overloaded, adds glob support to deny rules, and hardens cross-session messaging security.
GitHub added a one-million-token context window and configurable reasoning levels to Copilot on June 4. Both are available in VS Code, the Copilot CLI, and the GitHub Copilot app, with more surfaces rolling out soon.
Cursor's June 4 release adds Canvas Design Mode for direct element selection and annotation, and a Context Usage Report that shows exactly how tokens are split across your system prompts, tools, rules, and skills.
OpenAI's June 4 Codex CLI release brings enterprise admin flows with monthly credit limits, a rebuilt multi-agent system, remote-control pairing via app-server v2 RPCs, and parallel web searches.
Anthropic's June 4 release adds managed settings that enforce a version range — Claude Code won't start outside it. Also new: /plugin list, improved hook feedback, and 14 targeted bug fixes.
Zed released version 1.5.3 stable and 1.6.0 preview on June 3. The stable release adds Mermaid rendering improvements, clickable document links from language servers, and agent thread renaming. The preview introduces fast mode for Anthropic and OpenAI models, shareable skill links, and a setting for customizing AI commit messages.
Cursor's June 3 update adds Organizations for Enterprise customers: a top-level container for managing multiple teams with separate security, governance, budget, and model controls. Users can belong to more than one team, with different roles in each.
The GitHub Copilot App technical preview expanded to all Copilot Pro, Pro+, Business, and Enterprise customers on June 2. Headline additions include Canvases for bidirectional agent work, on-device voice, cloud sessions, and recurring automations that run without your machine.
OpenAI's June 1 Codex CLI release adds session archiving, Amazon Bedrock as a model provider, clickable TUI links, and security hardening that blocks repository-provided Git hooks from running inside /diff.
Claude Code's June 2 release adds security prompts before the agent writes to shell startup files or build-tool configs that grant code execution, renames the dynamic workflow trigger to 'ultracode', and clears a stack of Windows and session bugs.
Cognition pushed an over-the-air update on June 2 that renames Windsurf to Devin Desktop, promotes the Agent Command Center to the default IDE surface, and introduces Devin Local, a Rust rewrite of Cascade that is 30% more token-efficient.
GitHub Copilot's switch to usage-based billing hit on June 1. By June 2, developers were sharing real-world credit burn rates — and some are already planning exits.
Microsoft used Build 2026 to announce Project Polaris, a mixture-of-experts coding model built on Azure Maia accelerators that will replace GPT-4 Turbo as Copilot's default engine in August. Copilot Workspace also went generally available with fleet mode, autopilot, and new enterprise agent capabilities.
Anthropic closed a $65 billion Series H round on May 28, putting its post-money valuation at $965 billion and making it the most valuable private AI company in the world. Run-rate revenue has crossed $47 billion.
OpenAI extended Codex's computer use feature to Windows on May 29, bringing the autonomous screen-control capability that launched on Mac in April to the platform that most developers actually work on.
GitHub Copilot's switch from flat-rate subscriptions to GitHub AI Credits took effect June 1. The community announcement thread has nearly 1,000 downvotes, and some developers are projecting cost increases of 25x or more.
DeepSeek's 75% promotional discount on V4-Pro expired May 31, and the company made it permanent instead of reverting. Output tokens are now $0.87 per million.
The latest Claude Code release extends autonomous agent mode to developers running Claude through cloud provider inference platforms. You opt in with a single environment variable.
Cursor 3.6 shipped May 29 with a new run mode that routes agent tool calls through an AI classifier before executing them. It's a middle ground between fully autonomous and constantly interrupted.
GitHub added a four-phase AI adoption classification to the Copilot usage metrics API, letting enterprise admins see how deeply their teams are using agent features. A separate CLI release the same day restored model selection for Free and Student plan users.
Anthropic's Claude Opus 4.8 went generally available in GitHub Copilot on May 28, available across VS Code, JetBrains, Xcode, and more. It's launching with a 15x premium request multiplier just before Copilot's token-based billing kicks in on June 1.
METR tried to run a follow-up study on AI coding productivity. Developers refused because they wouldn't code without AI tools. Meanwhile, Uber burned its 2026 AI budget in four months, and Amazon shut down an internal token leaderboard after employees gamed it.
Anthropic released Claude Code v2.1.154 on May 28, integrating Opus 4.8 and introducing dynamic workflows that coordinate tens to hundreds of background agents. A follow-up v2.1.156 patch shipped May 29 to fix a thinking-block serialization bug.
On May 19, Google announced that Gemini CLI's free access ends June 18. The replacement is Antigravity CLI, a closed-source Go tool with a near-useless 20-request/day free tier. Community reaction to the open-source bait-and-switch has been sharp.
The maker of Devin and Windsurf closed a $1B Series D at a $26 billion post-money valuation, with $492M in annualized revenue and customers including Goldman Sachs, Mercedes-Benz, and NASA.
Anthropic shipped Opus 4.8 forty-one days after Opus 4.7. The headline numbers are a five-point jump on agentic coding, a fast mode that costs a third of what it used to, and a new Claude Code feature that runs hundreds of subagents in parallel for codebase-scale work.
GitHub published two Copilot changelog entries on May 26: Copilot Memory gets better deletion guidance, a repository-level off switch, and CLI commands, while enterprise admins gain targeted model rules that let them pick which models each organization can access.
OpenAI released Codex CLI v0.134.0 on May 26, adding search across local conversation history with case-insensitive matching and result previews. The release also makes read-only MCP tools run concurrently and cleans up profile management.
Anthropic shipped Claude Code v2.1.152 on May 27, adding --fix to the code review command so suggestions land in your working tree automatically. The release also gives skills and slash commands a way to restrict the tool list while they're active.
OpenAI shipped Codex CLI v0.133.0 on May 21, enabling Goals mode by default for all users. The release also adds Vim modal editing to the terminal UI, a new codex doctor diagnostics command, and a reworked permission profiles system.
Modal closed a $355M Series C led by General Catalyst and Redpoint on May 21, crossing $300M ARR after 5x growth in eight months. Its sandbox product — which powers code execution for Devin, Windsurf, and several Claude Managed Agent partners — now accounts for more than a third of revenue.
Anthropic's London developer conference on May 19 launched self-hosted sandboxes in public beta and MCP Tunnels in research preview, letting enterprise teams run Claude agents inside their own infrastructure for the first time.
Zed 1.3.5 ships Terminal Threads as a stable feature, letting developers run Claude Code, Codex, Amp, or any CLI agent as a managed thread in the sidebar. Mermaid diagrams now render inline in the agent panel. Version 1.3.6 follows with Gemini 3.5 Flash and thinking levels for Google models.
Copilot CLI 1.0.52 ships on May 23 with a context window tier fix that actually enforces the 200K vs 1M selection end-to-end, session resume in saved directories, and log file pruning. Version 1.0.53 the next day fixes multiline prompt display and a Bash hang.
Anthropic raised Claude Code weekly limits by 50% for all paid plans on May 13, through July 13. The increase follows a doubling of 5-hour rate limits on May 6 and removal of peak hour throttling — three capacity expansions in five weeks, funded in part by a new compute deal with SpaceX's Colossus 1 data center.
Google shipped Android CLI 1.0 stable at I/O 2026. The command-line interface gives AI agents like Claude Code, Codex, and Antigravity semantic symbol resolution, Compose preview rendering, and UI test execution — with 70% less token usage than running an agent inside Android Studio.
GitHub quietly dropped every Gemini model from Copilot Chat on github.com on May 20, along with GPT-5.2 Codex and GPT-5.4 nano. OpenAI and Claude models remain available. GitHub says it's about reliability — the scope of what was cut says more than the explanation does.
Version 2.1.149 adds a per-category breakdown to /usage — showing exactly how much of your limits comes from skills, subagents, plugins, and each MCP server. It also adds keyboard navigation in /diff and fixes a string of PowerShell permission bypasses.
Version 1.0.51 of the GitHub Copilot CLI adds session resumption with a --session-id flag, an experimental /security-review command for scanning code changes, and extended secret scanning into commit messages and PR descriptions.
Announced at Code with Claude London on May 19, self-hosted sandboxes let enterprises run agent tool execution on their own infrastructure. MCP tunnels let agents reach private servers without opening firewall ports.
A trojanized Nx Console extension was live on the VS Code Marketplace for 11 minutes on May 18. It stole GitHub tokens, AWS keys, npm credentials, 1Password vaults, and Claude Code configuration files from over 6,000 developer machines.
Claude Code 2.1.147 shipped May 21 with pinned background session improvements, auto-updater upgrades, and roughly 30 bug fixes including a wide swath of Windows reliability patches. 2.1.148 followed the next day to fix a bash exit code 127 regression.
Cursor shipped 3.4 on May 13 and 3.5 on May 20, adding full-screen agent tabs, Dockerfile-based cloud development environments, expanded Automations with multi-repo support, and a native Jira integration.
GitHub published the full source code for its Eclipse IDE plugin on May 21, 2026. The repository at github.com/microsoft/copilot-for-eclipse contains roughly 15,000 lines of Java covering code completion, Next Edit Suggestions, chat, and agent mode.
Between May 18-20, GitHub added cheap cloud agent models, brought Gemini 3.5 Flash to IDEs, launched auto model selection in VS Code with a 10% discount, and then stripped all Gemini models from web chat.
Two back-to-back Claude Code releases ship a JSON flag for listing live sessions, improved plugin discovery, better OpenTelemetry tracing, and a rename of /simplify to /code-review with optional effort levels.
GitHub made GPT-5.3-Codex the base model for all Copilot Business and Enterprise organizations on May 17, replacing GPT-4.1. It's GitHub's first long-term support model, guaranteed available through February 2027.
Google announced at I/O 2026 that Gemini CLI will stop serving requests for free and Pro/Ultra users on June 18. Consumer users must migrate to Antigravity CLI. Enterprise and API-key users are not affected.
Cursor released Composer 2.5 on May 18, built on Moonshot's Kimi K2.5 checkpoint with 85% of compute spent on Cursor's own post-training pipeline. It scores 63.2% on CursorBench v3.1, edging out both Opus 4.7 and GPT-5.5 at a fraction of the inference cost.
Gemini Spark launched at I/O 2026 as a proactive cloud agent that handles tasks across Gmail, Docs, and Slides while you're offline. The detail buried in the announcement: custom sub-agents and a local browser are coming. Here's what shipped.
Google AI Studio added native Android support at I/O 2026. Type a description, get Kotlin and Jetpack Compose code, preview in a browser emulator, and publish to Play Console's Internal Test Track with one click. Here's the launch.
Google added Managed Agents to the Gemini API today. A single API call provisions an agent with reasoning, tool use, and code execution inside an isolated Linux environment, powered by the Antigravity harness and Gemini 3.5 Flash.
Antigravity gets a full re-launch at Google I/O 2026. Standalone desktop app, new CLI, public SDK, dynamic subagents, scheduled background tasks, and a new $100 AI Ultra tier. Here's what shipped and what it changes.
Version 1.0.49 of the GitHub Copilot CLI adds a /rubber-duck command that prompts the agent to independently critique its own current work, plus /chronicle for searching session history, persistent memory controls, and Alpine Linux support.
The largest maintenance release in recent weeks adds /resume support so background sessions appear alongside interactive ones, per-session model switching, and a fix for the startup hang that could freeze Claude Code for 75 seconds behind a VPN or captive portal.
Google launched Gemini 3.5 Flash today at I/O 2026. It posts 78% on SWE-Bench Verified, runs four times faster than other frontier models, and costs $0.50 input / $3 output per million tokens. Here's the launch in detail.
Anthropic acquired Stainless, the startup behind nearly every official Claude SDK and MCP server generator, in a deal valued at over $300 million. Hosted products are being shut down, leaving competitors like OpenAI and Google to find alternatives.
Two May 13 updates: JetBrains IDE users can now delegate tasks to a locally running Copilot CLI agent with worktree or workspace isolation, while a new REST API lets teams trigger cloud agent tasks programmatically.
Google's latest stable Gemini CLI release enables Gemma 4 models by default and improves session management, while the v0.43.0 preview introduces session export/import, surgical code edits, and shell command safety evaluations.
The new Codex Chrome extension gives the agent access to browser tabs, DevTools, and signed-in web services — without requiring it to control your browser directly. Released May 7 for Codex Pro users on macOS and Windows.
OpenAI added Codex access to the ChatGPT iOS and Android apps on May 14. Your phone connects to a Codex session running on your Mac via QR code, letting you review outputs, approve commands, and start new tasks from anywhere.
xAI launched an early beta of Grok Build on May 15, a terminal-based agentic coding CLI powered by Grok 4.3. It's available now to SuperGrok Heavy subscribers at an introductory price, with plans for broader access.
Zed 1.2.4 adds a ChatGPT subscription provider so developers can sign in with a ChatGPT Plus or Pro account and use OpenAI models in the editor without per-token billing. Three patch releases shipped on May 15.
Claude Code 2.1.143 shipped May 15 with smarter plugin management — disable commands now block if dependencies exist and print the full disable chain — plus per-turn token cost projections in the plugin marketplace.
Roo Code — the open-source VS Code agent with 24K GitHub stars and 3 million installs — was archived on May 15. The team says IDEs aren't the future of coding, and has pivoted to Roomote, a Slack-native cloud agent.
Windsurf added Claude Opus 4.7 in fast mode to the editor on May 12, offering Opus-level intelligence at roughly 2.5x the output speed. A week earlier, version 2.2.17 extended Devin Review to all Pro, Max, and Teams subscribers with a two-week free trial.
GitHub's new native desktop app lets developers start agentic coding sessions directly from issues and pull requests, with each session running in an isolated branch and workspace. The app shipped v0.2.4 on May 15 with queued messages and collapsible tool-call panels.
Claude Code 2.1.142 landed May 14 with two headline changes: fast mode now defaults to Opus 4.7 instead of Opus 4.6, and the claude agents command gains eight new flags for configuring dispatched background sessions.
Cursor's newest releases let anyone in a Teams channel delegate coding tasks to a cloud agent by mentioning @Cursor, while a separate update brings Dockerfile-based multi-repo development environments with 70% faster cache builds and version history.
Gemini CLI v0.43.0-preview.0 ships an edit tool that modifies only the changed lines instead of rewriting full files, plus session export and import, an adaptive token calculator, and shell safety evaluations.
Claude Code 2.1.141 shipped May 13 with a new terminalSequence field in hook output for native OS notifications, ANTHROPIC_WORKSPACE_ID for workload identity federation, and one of the biggest bug fix batches in recent memory.
Claude Code 2.1.140 shipped on May 12 with smarter Agent tool subagent_type matching, an updated color palette for the agent view added in 2.1.139, and fixes for the /goal command, settings hot-reload, background service startup on enterprise machines, and a Windows event-loop stall.
GitHub announced a new Copilot Max plan at $100/month with $200 in total monthly value, introduced flex allotments to Pro and Pro+, and shipped code review improvements including severity labels and grouped comments — all taking effect on June 1 with the AI Credits billing switch.
AWS announced it is ending support for Amazon Q Developer IDE plugins and paid subscriptions on April 30, 2027. New signups stop May 15. On the same day, Kiro — the replacement — shipped three major spec features: neurosymbolic requirements analysis, parallel task execution, and Quick Plan mode.
Three Copilot CLI releases shipped May 6–11, adding a /autopilot toggle for autonomous mode, hooks that can handle requests without hitting the LLM, username display in the status line, and a sweep of path completion and slash command fixes.
Cursor is removing the $40 per-seat monthly subscription for Bugbot and switching to usage-based billing for Teams and Individuals. The change also unlocks configurable effort levels — you can now tell Bugbot to think harder on complex PRs.
Claude Code 2.1.139 shipped May 11 with agent view — a unified dashboard for every running, waiting, and finished session — and a /goal command that keeps Claude working until a condition you define is met.
OpenAI released Codex CLI 0.130.0 on May 8 with a new codex remote-control command that starts a headless, remotely controllable app-server — and a GitHub discussion thread confirms ChatGPT mobile is the intended controller.
Zed shipped version 1.0 on April 29, reaching the milestone with cross-platform support, a new Agent Client Protocol backed by Google and JetBrains, and a Zed for Business tier.
GitHub is dropping the premium request unit system for all Copilot plans on June 1, 2026, switching to a token-based credit model. Opus is out of the Pro tier, sign-ups are paused, and the fallback experience is going away.
Anthropic's 2026 Agentic Coding Trends Report, published in late April, identifies eight shifts reshaping software engineering. The most telling number: engineers use AI in roughly 60% of their work but say they can fully delegate only 0–20% of tasks.
Gemini CLI v0.41.0 shipped on May 5 with a /voice command for real-time spoken interaction, experimental Gemma 4 support, mandatory workspace trust in headless environments, and over 40 bug fixes.
ServiceNow made Build Agent generally available at Knowledge 2026, extending it into the four major AI coding tools so developers can build ServiceNow apps without leaving their preferred IDE.
Snyk integrated Anthropic's Claude models into its AI Security Platform on May 8, using Claude to power vulnerability discovery, automated remediation, and a new product that red-teams AI agents for prompt injection and data exfiltration.
Coder Technologies released Coder Agents in beta on May 6 — a self-hosted, model-agnostic AI coding agent platform for enterprises that need to keep code and prompts inside their own infrastructure.
Gemini CLI GitHub Actions is now in free beta, letting developers add Gemini as an autonomous agent in any GitHub repository — triaging issues, reviewing pull requests, and responding to @gemini-cli mentions.
Claude Code 2.1.133 shipped on May 7 with a configurable base ref for worktrees, hooks that can read the current effort level, Linux/WSL bubblewrap path settings, and fixes for parallel-session 401s, MCP OAuth proxies, and subagent skill discovery.
Cursor 3.3 shipped on May 7 with a new PR Review interface, async subagents that run independent tasks in parallel, and a quick action that splits one large change into multiple logically grouped pull requests.
Anthropic's developer conference in San Francisco on May 6 was about scale, not a new model. The headlines: a SpaceX compute deal, doubled Claude Code five-hour limits, and three new Claude Managed Agents features including 'dreaming' for self-improvement.
Claude Code 2.1.128 shipped on May 5 with a handful of quality-of-life improvements: randomized session colors, tool count display for MCP servers, ZIP archive support for plugins, and persistent 'always allow' rules for Bash.
A macOS Gemini app teardown found strings pointing to a new 'Google AI Ultra Lite' subscription tier sitting between the $20 AI Pro and $250 AI Ultra plans. A usage dashboard tracking token budgets is also in development.
Anthropic's developer conference kicks off in San Francisco on May 6. Workshops, live demos, and office hours with the Claude Code and API teams. The leaked model codename 'Jupiter-v1-p' and the Sonnet 4.8 references from March's npm leak both point to a model announcement.
AWS and OpenAI expanded their partnership on April 28, putting Codex, GPT-5.5, and a new Managed Agents service into Amazon Bedrock for enterprise teams that want OpenAI's models inside AWS infrastructure.
Starting June 1, 2026, GitHub Copilot replaces premium requests with GitHub AI Credits — token-based metered billing that ties cost directly to how much inference you consume.
Novee researchers disclosed a high-severity RCE vulnerability in Cursor IDE on April 28. A crafted Git repository with a hidden pre-commit hook can trigger arbitrary code execution when Cursor's AI agent runs routine git operations. Cursor patched it in February 2026.
Amazon announced on April 30 that Amazon Q Developer IDE plugins and paid subscriptions will reach end of support on April 30, 2027. New signups are blocked starting May 15, 2026. Kiro, AWS's spec-driven agentic coding environment, is the intended replacement.
Moonshot AI released Kimi K2.6 on April 20. The open-weight model scores 80.2% on SWE-Bench Verified and 66.7% on Terminal-Bench 2.0, with pricing of $0.60 per million input tokens on the official API. It's a meaningful upgrade over K2.5 on every benchmark that matters for coding agents.
VS Code 1.118 shipped on April 29 with a default-on setting that adds GitHub Copilot as a co-author on git commits. Developers noticed it was running even with AI features disabled, and Microsoft is reverting the default in 1.119.
Xiaomi released MiMo-V2.5-Pro on April 27: an open-source 1-trillion-parameter MoE model that scores comparably to Claude Opus 4.6 on coding benchmarks while using 40–60% fewer tokens per task. Weights are freely available on Hugging Face.
Copilot CLI 1.0.40 ships client_credentials OAuth for headless MCP authentication, fixes subagents using their own model's tool-search settings, and caps autopilot continuation at 5 by default.
Cursor Security Review enters beta on Teams and Enterprise plans with two always-on agents: a Security Reviewer that comments on every PR, and a Vulnerability Scanner that runs scheduled codebase sweeps.
Three releases over three days — v2.1.122, v2.1.123, and v2.1.126 — bring Windows PowerShell 7 as a first-class shell, a new project purge command, smarter resume-by-PR-URL, and OAuth improvements for WSL2 and SSH containers.
Microsoft's Agent 365 went generally available today, May 1, as part of the new Microsoft 365 E7 tier. It's a central control plane for registering, governing, and auditing AI agents across Microsoft, Copilot, and third-party platforms.
Microsoft's April 2026 update to Visual Studio brings the async cloud agent that VS Code has had for months. You describe a task, the agent runs on GitHub Actions, and you get a pull request without keeping VS open.
OpenAI's Codex CLI 0.128.0 adds a /goal command that keeps the agent looping toward an objective across turns until it achieves it, runs out of budget, or you pause it.
Gemini CLI v0.40.0 is the new stable release as of April 28. It bundles ripgrep for offline code search, adds tools for listing and reading MCP resources, and replaces the memory manager with a prompt-driven four-tier system.
IBM's Bob is now generally available. It covers the full software development lifecycle with multi-model routing, BobShell for auditability, and human-in-the-loop checkpoints — targeting enterprise teams who need AI coding with governance attached.
PocketOS, a car rental software company, lost its production database and all backups when a Cursor agent powered by Claude Opus 4.6 decided to 'fix' a credential mismatch on its own. The agent later confessed: 'I violated every principle I was given.'
OpenAI and AWS announced an expanded partnership on April 28, making GPT-5.4, GPT-5.5, and Codex available through Amazon Bedrock in limited preview. Enterprise developers can access OpenAI tools with existing AWS credentials and apply usage toward cloud spend commitments.
Windsurf 2.1.29 adds Devin as a standalone terminal agent on April 28. It runs on your machine with full codebase access, supports Opus 4.7, GPT-5.5, and SWE-1.6, and can hand sessions off to cloud Devin when the work outgrows your laptop.
Zed released parallel agent orchestration on April 22, making it the first AI code editor to natively run multiple agents simultaneously in the same window. A new Threads Sidebar groups sessions by project and controls what each agent can access.
GitHub announced April 27 that all Copilot plans will switch from premium request units to GitHub AI Credits on June 1, 2026. Base prices stay the same but usage is calculated on actual token consumption, and developers are not happy about it.
Claude Code v2.1.121 shipped April 28 with a plugin prune command, an alwaysLoad option for MCP servers, type-to-filter in /skills, and fixes for two separate multi-gigabyte memory leaks affecting image processing and /usage.
Cursor 3.2 shipped on April 24 with a /multitask command that breaks large requests into async subagents running in parallel, improved worktrees for branch-isolated background work, and multi-root workspaces for cross-repo agent sessions.
Cognition AI is in early talks to raise hundreds of millions at a $25 billion valuation — more than double its $10.2B valuation from September 2024. The company makes Devin, the autonomous AI software engineer, and acquired Windsurf last July.
Gemini CLI v0.39.0 shipped April 23 with a /memory inbox command that lets you review and patch skills the agent has extracted from your sessions. It also tightens Plan Mode with user confirmation for skill activation and adds richer visual output during execution.
OpenAI launched Workspace Agents in ChatGPT on April 22 as a successor to Custom GPTs. They're powered by Codex, designed for team sharing, and can run autonomously in the cloud while you're offline.
GitHub paused new sign-ups for Copilot Pro, Pro+, Student, and Business plans starting April 20, citing unsustainable compute costs from agentic workflows. Usage limits tightened and Opus models restricted to Pro+ only.
Anthropic quietly updated its pricing page on April 21 to remove Claude Code from the $20 Pro plan, said nothing about it, got caught, and reversed within hours. It called the whole thing an A/B test.
A 10,000-developer survey from JetBrains finds Claude Code grew from 3% to 18% workplace adoption between mid-2025 and January 2026, tied with Cursor for second place. A companion behavioral study of 800 developers reveals AI tools change how devs work in ways they don't notice.
GPT-5.5 is now generally available in GitHub Copilot across all major IDEs and platforms, but comes with a 7.5x premium request multiplier that will burn through credits fast. The same update brings inline agent mode to JetBrains IDEs.
Kiro CLI 2.1.0 ships with line-by-line shell output streaming, on-demand MCP tool loading, skills as slash commands, device flow authentication for SSH environments, and Red Hat Enterprise Linux support.
Claude Code versions 2.1.118 and 2.1.119 shipped April 23-24 with vim visual selection, a custom theme system, a unified /usage command, and --from-pr support for GitLab, Bitbucket, and GitHub Enterprise.
The open-source VS Code coding agent is ending its extension, Cloud, and Router services on May 15, 2026. The team says the IDE era is over and is betting on Roomote, a cloud-based autonomous agent that works through Slack, GitHub, and Linear.
OpenAI released GPT-5.5 on April 23, its most capable model yet, with benchmark scores showing meaningful gains in agentic coding, SWE tasks, and command-line workflows. The model now powers Codex for over 4 million weekly developers.
The latest stable Gemini CLI release brings built-in subagent delegation so your main session can hand off complex subtasks to specialized agents, plus a context compression service that keeps long sessions focused.
Accenture, Capgemini, CGI, Cognizant, Infosys, PwC, and TCS have all signed on to help enterprises adopt Codex. OpenAI also launched Codex Labs and crossed 4 million weekly active users in the span of two weeks.
Claude Code's Ultraplan feature moves implementation planning off your terminal and into the cloud. Opus 4.6 gets 30 minutes to think, you get a browser UI with inline comments. Here's what it actually does.
Anthropic is testing the removal of Claude Code from its cheapest paid tier. No announcement, no changelog - just a pricing page update that developers noticed immediately. Here's what happened, why it matters, and what Anthropic says about it.
Claude Code v2.1.110 landed today with the most significant terminal UX change in months: a flicker-free fullscreen rendering mode that works like vim. Plus mobile push notifications for long agentic sessions.
A leaked Brin memo says Google must 'urgently bridge the gap' in agentic coding. DeepMind assembled a dedicated team, some engineers are already using Claude Code instead of Gemini, and internal tensions are now visible enough to spill into the press.
SpaceX struck a deal with Cursor's parent company Anysphere: an option to acquire the AI code editor for $60 billion, or a $10 billion payment for their collaboration if they walk away. Here's the full breakdown of what they're building together and why it matters.
From quiet update freezes to full App Store removals, App Store ID hijackings, and an 84% submission surge, here's everything that's happened in Apple's escalating crackdown on vibe coding apps in 2026.
Cognizant and CGI both announced enterprise partnerships with OpenAI around Codex on April 21. CGI, which has 94,000 consultants worldwide already using Codex, gains early access to new capabilities. Cognizant embeds Codex directly into its Software Engineering Group.
GitHub stopped accepting new sign-ups for Copilot Pro, Pro+, and Student plans on April 20, citing compute costs driven by agentic workflows. The company also tightened usage limits and removed Opus models from the Pro tier. Community reaction has been harsh.
OpenAI updated its open-source Agents SDK on April 15 with native sandbox support, a new model-native harness for file and shell operations, configurable memory, and workspace portability via S3, GCS, Azure, and Cloudflare R2. Python ships now; TypeScript is coming.
Security researcher Aonan Guan found that AI agents running in GitHub Actions can be compromised by injecting malicious instructions into PR titles, issue bodies, and comments. All three vendors paid bug bounties but assigned no CVEs and published no advisories.
GitHub added an experimental feature to Copilot CLI called Rubber Duck that routes a Claude coding session through a GPT-5.4 reviewer at key moments. In GitHub's tests, it closed 74.7% of the performance gap between Claude Sonnet and Opus on SWE-Bench Pro.
JetBrains surveyed over 10,000 professional developers in January 2026. GitHub Copilot still leads by install base, but growth has stalled. Claude Code grew from roughly 3% to 18% adoption in eight months and now ties Cursor. Its satisfaction score (91% CSAT, NPS 54) leads the market.
Anthropic's latest flagship model accepts images at more than 3x the previous resolution, adds a new xhigh effort level between high and max, and ships file system memory that persists across sessions. The per-token price is unchanged, but a new tokenizer maps the same input to up to 35% more tokens.
Claude Design launched April 17 in public preview for Pro, Max, Team, and Enterprise subscribers. Powered by Claude Opus 4.7, it lets anyone build presentations, app mockups, and marketing materials through a chat interface. Figma's stock fell over 7% the same day.
OpenAI shipped a major Codex desktop update on April 16. The headline feature is computer use: Codex can now see, click, and type on your Mac while you keep working. It also added memory, image generation, 90+ new plugins, and self-scheduling.
Anysphere is in talks to raise over $2 billion at a $50 billion valuation, nearly doubling its November 2025 price tag in five months. a16z and Thrive Capital are expected to lead. The round is oversubscribed.
A new 'Fix with Copilot' button on pull request pages lets the Copilot cloud agent resolve merge conflicts, run CI checks, and push — all without touching your local environment.
The three-year-old startup behind 'Droids' — AI agents covering the full software development lifecycle — closed a Series C led by Khosla Ventures with Sequoia, Insight Partners, and Blackstone participating.
Anthropic shipped a major Claude Code desktop redesign with multi-session management, an integrated terminal, and a rebuilt diff viewer, alongside Routines — cloud-hosted automations that run without your laptop.
The latest Claude Code update lets Claude send mobile push notifications during Remote Control sessions and introduces a /tui command to switch into fullscreen rendering without leaving your current conversation.
GitHub shipped a public preview of remote control for Copilot CLI sessions, letting developers monitor, steer, and respond to a running agent from GitHub.com or the mobile app in real time.
Cognition's major Windsurf update replaces the single-agent workflow with a Kanban-style interface for managing fleets of local and cloud agents, and folds Devin directly into every self-serve subscription.
Thariq Shihipar from the Claude Code team published a guide to session management, context rot, compaction, rewind, and subagents. Here's every tip and why each one matters.
The Information reports Anthropic will release Claude Opus 4.7 and an AI design tool for websites and presentations this week. Figma fell 6% on the news.
Google launched the Gemini desktop app for macOS today with window sharing, a global keyboard shortcut, and free access. It's the third major AI assistant to claim a spot in your Mac's menu bar.
GitHub's updated privacy policy takes effect in nine days. Copilot Free, Pro, and Pro+ users' interaction data will be used for model training by default. Here's what's included and how to opt out.
GitHub shipped Autopilot as part of the March VS Code releases, published April 8. It's a fully autonomous mode where Copilot agents approve their own actions, retry on errors, and keep going until the task is done.
Version 2.1.108 ships today with a session recap feature, two new prompt caching environment variables, and the ability to call /init, /review, and /security-review through the Skill tool. Here's what changed.
JetBrains published the April 2026 results of their AI Pulse survey today — 10,000+ developers across 8 languages. Claude Code and Cursor are tied at 18% workplace adoption. GitHub Copilot leads at 29%. And Claude Code's NPS is 54.
Sam Altman celebrated Codex's growth milestone by resetting usage limits and launching a new $100 ChatGPT Pro tier built around Codex access — a direct shot at Anthropic's Claude Max.
OpenAI's new $100/month ChatGPT Pro plan goes after Claude Max directly: same price, five times more Codex than Plus, and access to the same models as the $200 tier. Here's what you get and why the timing matters.
A new free macOS app wraps Claude Code, Codex CLI, and Gemini CLI in fully CSS-skinnable interfaces, community themes, sidebar widgets, and visualizers. Vibe coding finally has a vibe.
On April 13, Claude went down. On April 14, Fortune and VentureBeat called it a trust crisis. Here's what actually happened, what Anthropic admitted, and what developers can do right now.
A First Squawk headline made it sound like breaking news, but Boris Cherny confirms Claude Enterprise switched to usage-based billing over six months ago. Here's what the model actually looks like.
Google's terminal AI coding tool just shipped its biggest update in weeks: session chapters for narrative flow, Linux and Windows sandbox expansion, a persistent browser agent, and early Gemini 3.1 Flash Lite support.
Anthropic rebuilds the Claude Code desktop from scratch with a session sidebar, drag-and-drop panels, integrated terminal, file editor, and live preview. The old single-session window is gone.
Anthropic launches Routines in Claude Code: repeatable automations that run on schedules, respond to API calls, or fire on GitHub events. No laptop required.
Boris Cherny, creator of Claude Code, pins a Discord message identifying two root causes behind weeks of user complaints about rapid quota drain. The 1M context window and runaway plugins are the culprits.
Leaked screenshots reveal a new Agent tab in Gemini Enterprise with task management, connected apps, file access, and a human review toggle. Here's what we know ahead of Google I/O.
Screenshots of Anthropic's unreleased 'Epitaxy' desktop experience show a full visual IDE with session management, pull request tracking, multi-repo tabs, and custom agent creation. Here's everything visible in the leaks.
Stella Laurenzo analyzed 234,760 tool calls across 6,852 sessions to prove Claude Code got worse in February. Anthropic's head of Claude Code responded. A breakdown of the data, the reply, and what it means for power users.
Matthew Gallagher used ChatGPT, Claude, and a dozen other AI tools to build Medvi into a $401M telehealth operation with two employees. Then came the fake doctors, the deepfakes, and the FDA.
Anthropic's Claude arrived in Microsoft Office through a native add-in, a Copilot model swap, and an M365 data connector. Here's how each integration works and what you actually get.
Anthropic shipped Claude Managed Agents in public beta on April 8, 2026. It's a fully managed harness with sandboxed containers, built-in tools, SSE streaming, a new ant CLI, and YAML-based agent definitions. Here's what it does, how it's priced, and what it means for Claude agent builders.
Meta Superintelligence Labs unveiled Muse Spark, a multimodal reasoning model with three thinking modes, 1,000-physician-curated health training, and a pretraining stack that needs an order of magnitude less compute than Llama 4 Maverick. Full breakdown of benchmarks, modes, and what it means.
Anthropic officially unveiled Claude Mythos Preview today as part of Project Glasswing, a $100M cybersecurity initiative with AWS, Apple, Google, Microsoft, and 8 other partners. The model already found a 27-year-old OpenBSD bug and beats Opus 4.6 by 17 points on CyberGym.
Sensor Tower data shows 235,800 apps were submitted to the App Store in Q1 2026, an 84% jump over Q1 2025. Anthropic's Claude Code, OpenAI's Codex, Replit, and Anything are getting most of the credit.
Anthropic announced that starting April 4, Claude Pro and Max subscriptions won't include usage on tools like OpenClaw. Extra usage bundles and API keys are the path forward.
Anysphere launched Cursor 3 on April 2, 2026, with a new Agents Window built from the ground up. It replaces the file-first IDE paradigm with an agent orchestration layer. Here's what shipped, what the numbers say, and why the timing matters.
Google DeepMind's Gemma 4 ships four models from 2B to 31B parameters, all under Apache 2.0. The 31B Dense model ranks #3 among open models globally. Here's everything you need to know.
Anthropic shipped a fullscreen rendering mode for Claude Code that eliminates flickering, adds mouse support, and keeps memory flat in long sessions. Here's how it works and what changes.
A 59.8 MB source map file shipped in the npm package exposed 512,000 lines of Claude Code's TypeScript source. The findings include hidden agents, anti-distillation defenses, an AI pet system, and an undercover mode for Anthropic employees.
Boris Cherny, creator of Claude Code, shared his favorite under-utilized features in a viral thread. Here's every tip, with context on what each one does and why it matters.
Fortune reports Anthropic accidentally exposed draft materials for 'Mythos,' a more powerful Capybara-tier model. The deeper story is what the leak says about frontier AI security.
WIRED reports OpenAI will discontinue the Sora app and API. The official Sora surfaces still live on, but they already point to ChatGPT and a broader product consolidation.
OpenAI's new Codex plugins package skills, app integrations, and MCP servers into reusable installs. Here's how Codex plugins work, where they live, and how to build one.
Anthropic's research preview brings computer use to Claude Cowork and Claude Code on macOS. Claude can open apps, click through browsers, and fill spreadsheets, all inside a sandboxed VM on your Mac.
From PI's four-tool minimalism to CrewAI's role-based teams, here's a practical breakdown of the frameworks and harnesses people are actually using to build custom AI agents right now.
Anthropic shipped Projects for Claude Cowork on Desktop, turning one-off tasks into persistent workspaces with local files, scoped memory, and recurring schedules. Here's how it works and how to get the most out of it.
A leaked model ID revealed that Cursor's Composer 2 is fine-tuned from Moonshot AI's Kimi K2.5. Here's the full timeline of the controversy, the license questions, and how it resolved in under 24 hours.
Anthropic shipped cloud-based scheduled tasks for Claude Code, letting you set a repo, a schedule, and a prompt that runs on their infrastructure. No local machine required.
Cursor launched Glass, an agent-first interface that replaces the traditional file tree with an agent orchestration view. It shipped alongside Composer 2, Cursor's first proprietary coding model. Here's what changed and why it matters.
Anthropic shipped Claude Code Channels in v2.1.80, turning Telegram and Discord into remote interfaces for running Claude Code sessions. Here's how it works, what it means, and why it looks a lot like an OpenClaw killer.
Google AI Studio's Build tab now supports databases, authentication, multiplayer apps, and backend code. The Antigravity coding agent runs the show. Here's what shipped and what it means.
Apple has quietly frozen App Store updates for Replit and Vibecode, citing rules against dynamic code execution. But Apple just added AI coding to Xcode. So what's really going on?
Anthropic ships Dispatch in Claude Cowork as a research preview. It's a single persistent conversation with Claude that runs on your desktop while you message it from anywhere.
OpenAI releases GPT-5.4 Mini and GPT-5.4 Nano with 400K context, computer use support, and pricing starting at $0.20 per million tokens. Here's what developers actually get.
The March 14, 2026 Australian headline is real. The deeper story is better. Rosie, a rescue dog with mast cell cancer, became the center of a homegrown personalized-medicine experiment that now deserves worldwide attention.
Google adds per-project monthly spend caps to the Gemini API, letting developers set hard budget limits directly in AI Studio. Here's how they work and what to watch for.
Google restructured Antigravity quotas around AI credit tiers, giving Ultra subscribers uncapped five-hour refreshes while Pro users hit weekly walls. Developers aren't happy.
Codex App 26.312 adds a theme system with base themes, custom accent/background/foreground colors, font selection for UI and code, and the ability to share themes. Here's what's included.
Claude Code 2.1.71 added /loop, a built-in scheduler that lets you run recurring prompts, poll deployments, and chain AI workflows on a timer. Here's how it works and what you can do with it.
Andrej Karpathy open-sourced autoresearch, a 630-line system that lets AI agents autonomously run hundreds of ML experiments on a single GPU. Here's how it works, how to use it, and what it means for the future of research.
OpenAI releases GPT-5.4 with native computer-use capabilities, 1M token context, scalable tool search, and best-in-class agentic coding. Full breakdown of benchmarks, pricing, and what it means for developers.
OpenAI launched the Codex desktop app for Windows on March 4, 2026, with a native OS-level sandbox built with Microsoft, PowerShell-first integration, and cross-platform session continuity. Over 500,000 developers were on the waitlist.
Google DeepMind releases Gemini 3.1 Flash-Lite with $0.25/M input pricing, 363 tok/s output speed, and 1M token context. Here's the full breakdown: benchmarks, pricing, and what it means for developers.
Alibaba released the Qwen 3.5 Medium open-weight models on February 24, 2026. The 35B-A3B MoE model hits 111 tokens/sec on an RTX 3090 and handles 1M+ context on 32GB VRAM. Here's a full hardware guide with real tok/sec numbers.
Figma shipped its Codex integration today. Designers can send frames to Codex for code generation, and developers can push running UIs back to Figma's canvas as editable layers. Here's how it works, what the MCP server actually does, and why this matters.
Anthropic's Feb 24 enterprise briefing brought major Cowork updates: scheduled tasks that run on autopilot, 13 new plugins for HR, finance, and engineering, plus Google Workspace and DocuSign connectors.
Perplexity launched Computer today, a unified AI platform that orchestrates 19 models including Claude Opus and GPT-5.2 to handle multi-step projects. It costs $200/month and the ambition is enormous.
Anthropic shipped Remote Control for Claude Code today. You can now hand off a terminal session to your phone. Here's how it works, how it compares to OpenClaw's approach, and what it means for mobile-first AI agents.
84% of developers use AI coding tools. Only 29% trust them. A new study found AI actually makes experienced developers 19% slower. The numbers tell a strange story about the state of AI-assisted programming in 2026.
On February 20, Anthropic announced Claude Code Security had found over 500 previously unknown vulnerabilities in open-source codebases. CrowdStrike, Cloudflare, and others dropped by up to 9%. Here's what happened and what it means.
Three AI coding agents now fight for your terminal. Claude Code, OpenAI Codex, and Gemini CLI take fundamentally different approaches to the same problem. Here's how they compare on features, pricing, and what actually matters.
At India's AI Impact Summit, Altman argued that comparing AI energy costs to human queries is unfair because humans take 20 years and thousands of meals before they 'get smart.' Here's the full picture.
Boris Cherny just shipped native git worktree support for Claude Code CLI. Parallel agents, no collisions, automatic cleanup, and non-git VCS support via hooks.
In a candid Express Adda interview today, OpenAI's CEO said AI's takeoff is faster than he originally expected, AGI is close, and super intelligence is not that far off. He said it calmly. That's somehow more alarming.
Google releases Gemini 3.1 Pro with a 77.1% ARC-AGI-2 score, top marks on 13 of 16 benchmarks, and the same pricing as its predecessor. Here's what it does, where it leads, and where it doesn't.
Three Gemini CLI releases in two weeks: extension settings, plan mode, Gemini 3 as default, an SDK package, and experimental steering hints. Here's everything that shipped.
Dmitry Lyalin shared six upcoming Gemini CLI features including model hinting, OS notifications, clean UI mode, and a mention of 'Gemini 3.x' optimizations. Here's the breakdown.
Google employees are posting cryptic hints on X, a Gemini 3.1 Pro Preview showed up on a leaderboard tracker, and Deep Think just got a major upgrade. Here's everything we know. UPDATE: It shipped.
Google DeepMind's Lyria 3 lets Gemini's 750 million users generate 30-second songs from text or photos. It's free, it's in beta, and it's going up against Suno and Udio. Here's how it works and what people think so far.
A venture capitalist with no dev experience built BabyAGI in three hours using GPT-4 prompts. It went viral, spawned a movement, and defined the autonomous agent loop that every framework still copies. Here's what it is, where it stands, and why it still matters.
Anthropic's Claude Sonnet 4.6 matches flagship-class performance on coding, computer use, and agent tasks while costing 5x less than Opus. Here's what the benchmarks say, what developers think, and what it means for your workflow.
A holiday weekend roundup of everything that happened in AI coding tools. New models from OpenAI and Anthropic, Codex goes desktop, GPT-4o gets retired, and the money keeps flowing.
This week's biggest AI coding stories: OpenAI ships Codex-Spark on Cerebras hardware, OpenClaw's security crisis deepens, Spotify says its best engineers stopped writing code, and Windsurf drops Arena Mode.
Now that Peter Steinberger has joined OpenAI and OpenClaw lives under their sponsorship, here are the best alternatives for developers who want independent, open-source AI agents.
The OpenClaw creator chose OpenAI over Meta. Sam Altman announced the hire on X, calling Steinberger 'a genius.' OpenClaw will live in an independent foundation with OpenAI sponsorship.
The OpenClaw creator sat down with Lex Fridman for nearly four hours. He talked about burning out after a $100M exit, arguing with Zuckerberg about Claude vs Codex, almost deleting the project, and why he doesn't care about money. Here are the key moments.
The OpenClaw creator has offers from both Meta and OpenAI on the table. Between the Lex Fridman interview, Sam Altman conversations, and an upcoming OpenAI developers feature, the signs are pointing somewhere specific.
Moonshot AI launched Kimi K2.5, a trillion-parameter open-source model, alongside Kimi Code CLI. The standout feature: Agent Swarm, which orchestrates up to 100 sub-agents in parallel.
Anthropic closed its Series G at a $380 billion valuation, the second-largest venture deal ever. Claude Code is pulling in $2.5 billion a year. The numbers tell a story about where AI is heading.
Six of xAI's twelve cofounders are gone. Engineers are leaving in waves. Behind the exits: safety failures, a deepfake scandal that triggered police raids, and a culture that burned through its best people.
Soon, running a business will mean orchestrating AI agents through custom dashboards. The best operators will build their own software. And somewhere on Steam, someone will be making real money playing a game that looks exactly like their actual job.
A Chinese startup built an autonomous AI agent that browses the web, writes code, and runs on virtual computers. Then Meta bought it. Here's why it matters and why people are divided.
OpenAI just dropped a coding model that runs at 1,000 tokens per second on Cerebras hardware — not Nvidia. It's smaller, faster, and the first real crack in the GPU monopoly.
MiniMax launches M2.5, calling it the first production-level model built natively for agent scenarios. Here's what the benchmarks show, what it costs, and how to actually use it.
The viral AI essay that consumed the internet this week, dissected: the METR data, the self-building models, the job displacement claims, and the credibility questions nobody's asking loudly enough.
Claude Code Desktop now supports --dangerously-skip-permissions, bringing the CLI's fully autonomous 'YOLO mode' to the graphical interface. Here's what it does, how to use it, and why you should be careful.
The Anthropic vs. OpenAI Super Bowl ad war, Sam Altman's 400-word meltdown, and the same-day launch of Claude Opus 4.6 and GPT-5.3-Codex that triggered a trillion-dollar stock selloff.
From Anthropic roasting ChatGPT to Svedka's AI-generated nightmare to Chris Hemsworth fighting his own Alexa — a complete breakdown of how AI dominated Super Bowl LX.
OpenClaw is a free, self-hosted AI agent that connects LLMs to your messaging apps, files, and services. Here's what it actually is, the best ways to run it, and how to avoid a $500 API bill.
A no-filler directory of every one-click, easy-deploy, and free-tier option for running OpenClaw in 2026. DigitalOcean, Contabo, Railway, Oracle Cloud's free tier, and more — with real pricing and direct links.
On February 5, 2026, Anthropic and OpenAI released their flagship coding models within minutes of each other. The developer community had feelings.
Deirdre Bosa vibe-coded a project management app on live TV using Claude. It cost $15 and took under an hour. Meanwhile, software stocks lost $285 billion.
A year after Pieter Levels built a $138K/month flight simulator in 3 hours with Cursor, vibe coding has gone from meme to market-moving force. Here's the full timeline.
Claude Cowork sparked a $285B software stock selloff. Here's why Anthropic's desktop tool is rewriting the SaaS playbook.
Anthropic's Claude Opus 4.6 launches with 1M token context, agent teams, and adaptive thinking. Here's what changed, what matters, and what to actually use it for.
OpenAI's GPT-5.3-Codex arrives with reasoning chops, self-assisted training, and serious cybersecurity concerns. Here's what matters for developers.
Vibe coding is writing software by describing what you want to an AI, without reading the code it produces. Coined by Andrej Karpathy in February 2025, the term has taken on a life of its own. Here's what it actually means.