Autoheal Raises $7.9M Seed to Build Self-Improving AI Software Factories

The company is targeting a problem created by the rapid adoption of AI coding tools. Enterprises are shipping more software than before, but platform engineering teams face more production incidents and vulnerabilities, along with rising token costs. Many teams have adopted software factory models powered by dozens of specialized AI agents, yet those agents often fail at scale because they lack shared context and security constraints, Autoheal said.

Autoheal provides a unified platform for creating, managing and improving software factory agents, including the coding agents that drive development workflows. By managing each agent through the same platform, the company says, teams can give agents access to the same engineering context, private evaluation infrastructure, and cost and security controls.

Co-founder and Chief Executive Sid Choudhury said Autoheal has built a unified operating model for building, governing and continuously improving AI agents across the software development lifecycle. The platform can be hosted in an organization’s private cloud within a secure boundary, connecting to existing coding agents, code repositories, continuous integration and continuous development pipelines, and observability tools.

At the center of the platform is a shared engineering context graph maintained by two specialized agents. An Evaluator agent rates downstream worker AI agents using metrics such as CI failures and incident reports. A Healer agent tries to fix low-scoring agents by opening pull requests that improve model selection, prompts, tools and skills. Each change is version controlled in Git, checked against historic benchmarks and approved by a human supervisor.

Choudhury and his co-founders previously built enterprise-grade AI and engineering infrastructure at Microsoft Corp., ThoughtSpot Inc. and Harness Inc. He said it was at Harness that they saw the need to manage agents as code, overseen by continuously learning meta-agents.

“Our experience taught us that while building the first version of an AI agent is easy, scaling it consistently across the enterprise SDLC is the real challenge,” Choudhury said. “Platform engineers need a unified platform to deploy agents that don’t just execute tasks, but continuously improve alongside complex enterprise workflows.”

Autoheal operated in stealth until now but has customers including Normura Holdings Inc., AvidXchange Inc. and Empiric Earth Inc. The company said these customers have used its platform to reduce incident resolution times and save thousands of hours of engineering work.

Normura Bank Chief Information Officer Sameer Jain said his production operations teams had been overwhelmed by alerts and spent hours triaging them, often pulling engineers away from other work. “Autoheal gives us a platform that takes investigation timelines down from hours to minutes,” Jain said. “The fact it runs entirely within our own cloud, in compliance with our controls, makes it a natural fit for how we operate.”

Autoheal now plans to develop reinforcement learning techniques that can train customers’ AI agents on their own private engineering data, Choudhury said. The goal is for customers to create enterprise-specific small language models that run entirely inside their secure private cloud environments. Those models could then power each customer’s fleet of software factory agents, reducing costs and improving knowledge of the industries they serve. Choudhury said he believes the architecture can eventually expand beyond software engineering into data and security engineering.

Editor's Summary

Autoheal AI Inc. raised $7.9 million in seed funding led by Innovation Endeavors for a platform that evaluates and fixes AI agents used in software development with other AI agents. The startup says its system gives enterprises shared context, evaluation and security controls, and it already counts Normura Holdings, AvidXchange and Empiric Earth as customers. Autoheal plans to add reinforcement learning and private small language models to train agents on customer-specific engineering data.