Autoheal raises $7.9M to build a software factory where agents repair each other
Autoheal's $7.9M seed, led by Innovation Endeavors, funds a platform where an 'evaluator' agent scores worker agents and a 'healer' agent opens PRs to fix the underperformers.
The coding-agent gold rush is spawning a second layer: startups selling the infrastructure to run agents at enterprise scale. The latest is Autoheal, which announced a $7.9 million seed round Monday to build what it calls a self-improving software factory, the company said.
The round was led by Innovation Endeavors, with Harpinder Singh joining Autoheal's board, plus Emergent Ventures, U&I Ventures, Darkmode Ventures, Batch Ventures, and Param Hansa Values, and angels including Teradata CPO Sumeet Arora and Skyflow CEO Anshu Sharma. Autoheal's founders, Utkarsh Ohm, Sid Choudhury, and Puneet Saraswat, came from Harness, Microsoft Azure, ThoughtSpot, and AppDynamics; Choudhury is CEO.
The pitch addresses a real pain point. Coding agents have made the first draft of software cheap, but incident response, vulnerability remediation, and token-cost management are eating the savings. Autoheal says repetitive SDLC workflows consume more than a third of engineering capacity, and agent rollouts fail on fragmented tools, missing shared context, and security constraints.
Its answer is a platform that connects a company's existing coding agents, repos, CI/CD, observability, cloud runtimes, and issue trackers into one shared engineering context graph, with two meta-agents running in the background. An evaluator agent scores every worker agent's run, using downstream signals like review comments, CI failures, and caused incidents. A healer agent then fixes low-scoring agents by opening pull requests that improve skills, prompts, tools, or model selection, verified against historical benchmarks for regressions before an engineer reviews them. Every behavior change is version-controlled in git and requires engineer approval.
Traction is early but concrete: the company names Nomura Bank and AvidXchange as design partners. Nomura's wholesale CIO Sameer Jain said Autoheal took investigation timelines from hours to minutes and runs entirely within the bank's own cloud; AvidXchange CTO Krish Shetty said it cut root-cause time to minutes with evidence engineers trust.
Autoheal's longer-term bet is that the data captured inside the factory becomes the raw material for small private models per customer, trained on the enterprise's own implicit architecture decisions that frontier models never saw. The thesis is that every large enterprise will run its own population of specialized agents and needs one governed platform to build, deploy, and continuously improve them.
At $7.9 million, this is a small bet on a big thesis, and the "self-improving" loop is the part to watch: agent infrastructure that gets cheaper and more reliable the longer it runs is exactly what enterprise buyers say they need before trusting agents with the SDLC end to end.
