IBM Launches Bob, an Enterprise AI Coding Partner That Routes Tasks Across Claude, Mistral, and Granite

IBM shipped Bob to general availability on April 28. It is IBM’s answer to the question enterprise developers have been asking since Claude Code and Codex started showing up in procurement conversations: what does an AI coding tool look like when security, auditability, and compliance are not optional features?

Bob is not a code completion tool. It is closer in spirit to Devin or Codex than to Copilot. IBM positions it as an AI development partner that covers the full software development lifecycle, from planning and coding through testing, deployment, and legacy modernization.

Multi-Model Routing

The most technically interesting part of Bob is the model layer. Rather than committing to a single foundation model, Bob dynamically routes tasks to whichever model fits best, given accuracy, latency, and cost constraints at the time of the request.

The current mix includes:

• Anthropic Claude for complex reasoning and code generation
• Mistral open source models for tasks where open-weight models are sufficient
• IBM Granite for code reasoning, security analysis, and next-edit prediction

IBM also uses specialized fine-tuned models for specific subtasks. The routing is handled automatically, which means users do not pick a model per task. IBM’s system picks for them based on the task type.

This is a different bet from most enterprise coding tools, which typically offer a model picker. IBM’s argument is that task-level routing — not user-level selection — is what actually optimizes output quality.

BobShell and Auditability

For enterprise teams, the headline feature may be BobShell, Bob’s CLI layer. BobShell creates a self-documenting trace of every action the agent takes, in real time. Every file edit, every shell command, every API call is logged with full context.

The stated goal is compliance and traceability. If something breaks in production, you should be able to trace every step the agent took. IBM calls this “full auditability from start to finish.”

Bob also includes configurable human approval checkpoints. Teams can set these from fully manual (every action requires approval) to fully automatic (agent runs without interruption), with granular settings in between. The checkpoint configuration is intended to let teams match agent autonomy to their actual risk tolerance, rather than having a binary on/off.

Internal IBM Numbers

IBM says 80,000 employees are currently using Bob. Across surveyed users, IBM reports an average self-reported productivity gain of 45%.

Specific team-level results IBM has shared:

• IBM Instana team: 70% reduction in task time, roughly 10 hours saved weekly
• IBM Maximo team: 69% time savings on code updates and refactoring

Third-party results IBM cites include Blue Pearl, a cloud solutions company. Blue Pearl used Bob to complete a Java upgrade project that was scoped at 30 days in 3 days, saving over 160 engineering hours. Ernst & Young is also cited as a customer, using Bob for tax platform modernization.

These numbers come from IBM, so they carry the usual caveats that apply to vendor-reported productivity figures. That said, 80,000 internal users is a real deployment at scale.

What It Costs

Bob is available as SaaS with a 30-day complimentary trial. IBM offers both individual and enterprise plans. On-premises deployment is planned for a future release.

Pricing is not published on the product page. Enterprise pricing for tools in this category is typically negotiated, which is consistent with IBM’s positioning of Bob as a compliance-first enterprise product rather than a self-serve developer tool.

Context

Bob enters a market where the enterprise AI coding category is still taking shape. GitHub Copilot Business has the broadest enterprise adoption by user count, but it is primarily an inline completion tool. Codex and Claude Code are increasingly showing up in enterprise procurement conversations, but both require more configuration for compliance use cases.

IBM’s angle is that Bob ships with governance built in, rather than treating it as a configuration problem for customers to solve. The multi-model routing, BobShell auditability, and configurable checkpoints are all oriented toward the specific objections that enterprise security and compliance teams raise when AI coding tools come up for review.

The 30-day trial is live at bob.ibm.com.

---

Sources: IBM Newsroom, VentureBeat, The Register, DevClass