Autoheal raises $7.9m seed to build self-improving AI agent platform
Autoheal, a San Francisco-based startup, has closed a $7.9 million seed round to scale what it describes as a self-improving software factory for enterprise platform engineering teams. The round was led by Innovation Endeavors, with Harpinder Singh joining the board. Participating investors include Emergent Ventures, U&I Ventures, Darkmode Ventures, Batch Ventures, and Param Hansa Values. A cohort of angel investors from senior roles at Teradata, Skyflow, Outerbounds, and ThoughtSpot also joined the round.
The company was founded by Utkarsh Ohm, Sid Choudhury, and Puneet Saraswat, all of whom previously held engineering and product roles at Harness, Microsoft Azure, ThoughtSpot, and AppDynamics. Choudhury, who serves as chief executive, helped scale Harness to more than $200 million in ARR before co-founding Autoheal.
What the platform does
Autoheal connects existing coding agents, code repositories, CI/CD pipelines, observability tooling, cloud runtimes, and issue trackers into a shared engineering context graph. Two meta-agents operate continuously in the background: an Evaluator that scores each worker agent's output using downstream signals such as code review comments, CI failures, and incidents; and a Healer that raises pull requests to improve underperforming agents by updating their skills, prompts, tools, or model selections. All changes are version-controlled in git and require engineer sign-off before deployment, keeping human oversight in the loop.
The stated aim is to reduce the operational overhead that compounds as AI-assisted coding accelerates: more production incidents, more security vulnerabilities to remediate, and rising LLM token expenditure. Autoheal says repetitive software development lifecycle workflows already consume more than a third of a typical engineering team's capacity.
Named enterprise customers include Nomura Bank, AvidXchange, and Empiric Earth. Sameer Jain, CIO Wholesale at Nomura Bank, said the platform "takes investigation timelines down from hours to minutes" while running entirely within the bank's own cloud environment, satisfying its internal compliance controls. Krish Shetty, CTO and SVP at AvidXchange, said the tool reduced time to root cause for production incidents to minutes, with evidence engineers found trustworthy.
Market context
Autoheal enters a crowded but still forming market. Enterprise interest in agentic AI infrastructure has surged in 2026, with a growing number of vendors pitching orchestration, evaluation, and governance layers for AI agents across engineering workflows. Hyperscalers including Microsoft, Google, and Amazon have each embedded agent tooling into their developer platforms, while dedicated startups such as Langchain, Comet, and various observability vendors are competing on evaluation and observability for LLM-powered systems.
The distinguishing bet Autoheal is making is that the governing layer, rather than any individual agent, is the durable platform. Harpinder Singh of Innovation Endeavors framed the opportunity as "giving platform teams a repeatable, scalable way to deploy specialised intelligence across the engineering organisation," rather than backing a single-workflow agent.
The company also flags a longer-term product direction: training small, private models on each customer's proprietary engineering data to reduce dependence on frontier models for non-generative tasks. This positions Autoheal alongside a broader enterprise trend toward sovereign, cost-contained AI built on internal data rather than public-internet-trained general models. Whether a seed-stage company can execute model training at per-customer granularity alongside platform development will be a key question for investors as the next funding round approaches.
The round is undisclosed in terms of post-money valuation. Autoheal has not stated its current headcount or ARR, and no independent auditing of the customer claims above has been published.