Autoheal raises $7.9 million to evaluate and repair AI agents with…AI agents

Autoheal AI Inc., an artificial intelligence-based platform engineering startup seeking to advance the concept of “self-improving software factories,” today announced that it has raised $7.9 million in seed funding to make this possible.

Today’s round was led by Innovation Endeavors and included Emergent Ventures, U&I Ventures, Darkmode Ventures, Batch Ventures and Param Hansa Values.

Thanks to the widespread adoption of AI tools that can generate code faster than humans, companies are now shipping more new software than ever before. But this acceleration has caused as many problems as it has solved, as platform engineering teams contend with a dramatic increase in production incidents and vulnerabilities in their software, as well as rapidly rising token costs.

To address these challenges, many teams have adopted a “software factory” model supported by dozens of specialized AI agents to manage every step of development workflows. However, these agents often fail on their own because their rollouts are so large and lack common context and security constraints.

According to Autoheal, there is an urgent need for enterprises to adopt a unified platform for creating, managing and improving these software factory agents, including the all-important coding agents that require their existence. By managing every agent through the same platform, everyone has access to the same technical context, private evaluation infrastructure, and cost and security controls.

That’s exactly what Autoheal offers, explained co-founder and CEO Sid Choudhury. The startup has developed a unified operating model for building, governing and continuously improving AI agents that drive the entire software development lifecycle. Its platform can be hosted within a secure boundary in organizations’ private clouds and from there connect to any existing coding agents it uses, along with its code repositories, continuous integration/continuous development pipelines, and observability tools.

According to Choudhury, Autoheal creates a common engineering context graph that is managed by two specialized agents: an “evaluator agent” that evaluates employee downstream AI agents based on metrics such as CI errors and incident reports, and a “healer agent” that aims to fix low-scoring agents by opening pull requests that improve model selection, prompts, tools, and capabilities. Every change made by the healer agent is versioned in a Git, reviewed against historical benchmarks, and approved by a human supervisor.

Choudhury said he and his co-founders previously spent years building enterprise-grade AI and engineering infrastructure at companies such as Microsoft Corp., ThoughtSpot Inc. and Harness Inc. At Harness, they recognized the need to “manage agents as code, monitored by continuously learning meta-agents,” he said.

“Our experience has shown us that while building the first version of an AI agent is easy, scaling consistently across the enterprise SDLC is the real challenge,” Choudhury explained. “Platform engineers need a unified platform to deliver agents that not only perform tasks but continually improve alongside complex business operations.”

Although Autoheal has operated under the radar so far in “stealth” mode, it has gained popularity among enterprise customers such as Normura Holdings Inc., AvidXchange Inc. and Empiric Earth Inc. These companies all say they have used Autoheal’s platform to reduce incident resolution times and save thousands of hours of technical work.

Normura Bank chief information officer Sameer Jain said his production teams had previously been overwhelmed by alerts and had to spend hours endlessly testing them to deal with incidents. Their work often required them to pull engineers away from the things they were working on to help them solve problems. “Autoheal gives us a platform that shortens the investigation time from hours to minutes,” he said. “The fact that it runs entirely in our own cloud and conforms to our controls makes it a natural fit for the way we work.”

Autoheal’s goal now is to develop new reinforcement learning techniques that can train customers’ AI agents using their own, private technical data, Choudhury said. The idea is that customers can build enterprise-specific small language models that work entirely within their own secure private cloud environments.

Ultimately, these SLMs are intended to serve each customer’s fleet of software factory agents, reducing costs while increasing knowledge of the specific industries in which they operate. In the long term, Choudhury said, he is convinced that Autoheal’s architecture can be expanded beyond software development to data and security technology.

Harpinder Singh of Innovation Endeavor said he has come across many companies asking questions about how they can safely and efficiently operate AI agents at scale and trust them to run their software factories. “Autoheal is building the agent infrastructure layer that makes this possible,” he emphasized. “The opportunity is much larger than one agent or one workflow. It provides platform teams with a repeatable and scalable way to deliver specialized information across the technical organization.”

Image: Autoheal