The Pulse
Autoheal Raises $7.9M to Govern Self-Improving AI Agents
Autoheal raised a $7.9 million seed round to develop a platform for deploying and improving enterprise software agents, including agents for incident response and vulnerability remediation. Its system evaluates agent runs and proposes chang
AI.info Team · Written with AI · fact-checked ·
AI coding agents can speed up software delivery, but Autoheal argues that the work after code is written—incident response, security fixes and cost control—still needs human oversight. The startup announced a $7.9 million seed round on September 28, 2026, to build a platform that evaluates operational agents and proposes changes to improve them. Its central constraint is that those changes do not go live without an engineer’s approval.
Autoheal puts an approval gate on agent self-improvement
Innovation Endeavors led the round, with participation from Emergent Ventures, U&I Ventures, Darkmode Ventures, Batch Ventures and Param Hansa Values, according to Autoheal’s announcement. The company also named angel investors Shawn Kung, Sumeet Arora, Anshu Sharma, Savin Goyal and Srikant Gokulnatha. Autoheal says its team previously built and scaled the DevOps platform at Harness.
The company describes its product as a shared system for building, running and governing agents across software engineering. It connects to customers’ repositories, CI/CD pipelines, observability tools, cloud environments, issue trackers and coding agents. Autoheal says its context graph links repositories, services, tickets and releases, while a registry stores the skills, agent instructions and memories used for tasks.
Autoheal distinguishes its platform from coding agents developers already use: it says it does not replace them. Instead, the platform aims to give agents shared engineering context and automate repetitive work after a commit, such as incident response and vulnerability remediation. The company says teams can deploy the system in their own cloud or in an air-gapped environment, with access policies and an audit trail for tool calls.
Evaluator scores runs; Healer proposes changes
The company’s proposed improvement loop has two parts. An Evaluator scores an agent’s work against downstream signals, including review comments, repeated CI runs, alerts and an incident’s eventual root cause. Autoheal says evaluations use customers’ own runs and outcomes inside their security boundaries.
A Healer then turns weak scores into proposed changes to skills, agent instructions, stored memories, tool permissions or model selection. Each change becomes a pull request and, according to the company, is tested against run history and promoted only if it clears a blast-radius threshold. Autoheal says changes are version-controlled in Git and require an engineer’s sign-off before going live.
Customer gains are presented as early evidence
Autoheal says Nomura Bank and AvidXchange use its incident-response agents. At Nomura, the company says, agents gather evidence from multiple systems while operating within the bank’s access controls. Sameer Jain, CIO, Wholesale at Nomura Bank, described the product’s effect:
“Our production operations teams spend valuable time triaging alerts and managing incidents, while also pulling engineers away from their software development activities. Autoheal gives us a platform that takes investigation timelines down from hours to minutes. The fact that it runs entirely within our own cloud, in compliance with our controls, made it a natural fit for how we operate.”
Autoheal also says AvidXchange reduced incident root-cause investigation to minutes and is expanding use of the platform across its software-development lifecycle. These are customer statements published by the startup, rather than results from a comparative study in the announcement.
Company plans customer-specific models
Autoheal says it is building toward enterprise-specific small language models trained with reinforcement learning on each customer’s private software-development data. Its longer-term vision is for engineering teams to set targets for quality, security, reliability and cost, while agents work toward those targets under company governance. For now, the approval rule is explicit: an agent can propose a change, but an engineer must approve it before it reaches production.