Autoheal just closed a $7.9 million seed round led by Innovation Endeavors. The goal is to scale its self-improving software factory for enterprise platform teams.
The round also featured Emergent Ventures, U&I Ventures, Darkmode Ventures, Batch Ventures and Param Hansa Values. Harpinder Singh joined the board as part of the deal.
Its platform tackles the post-coding work that still drains more than a third of engineering time. That includes incident response, vulnerability remediation and LLM cost control.
Before this funding, the founders built enterprise engineering and AI platforms at Harness, Microsoft Azure, ThoughtSpot and AppDynamics. Their work helped Harness surpass $200 million in ARR.
That experience made the team realize the real challenge isn’t building a single agent, but running dozens across the SDLC without fragmented tools or lost context.
Off-the-shelf point agents stumbled with early design partners Nomura Bank and AvidXchange. Strict security and production access limits prevented them from scaling.
Autoheal ties together existing coding agents, repositories, CI/CD pipelines, observability tools, cloud runtimes and issue trackers. The result is a shared engineering context graph.
Two meta-agents work nonstop. An Evaluator scores each worker run against downstream signals-review comments, CI failures and incidents-while a Healer opens pull requests to tweak prompts, skills, tools or model choices.
Every change made by a Healer is version-controlled in git and must be approved by an engineer. Autonomy grows only as reliability proves itself.
Customers are already seeing faster triage. Nomura Bank reports investigation times dropping from hours to minutes while staying entirely within its own cloud.
AvidXchange says root-cause analysis now takes minutes thanks to trusted evidence. Developers are freed up for feature work, and usage is expanding into other SDLC stages.
The $7.9 million round looks modest next to bigger bets on the same idea. Rival Factory raised $200 million at a $5 billion valuation to scale self-improving software development, using the same “software factory” framing as Autoheal.
Autoheal says the next frontier is sovereign intelligence. It wants to use reinforcement learning on private engineering data to train enterprise-specific small language models inside the customer’s security boundary.
That step, plus an extension beyond software into data and security engineering, will test whether a healing loop can cut cost per successful task without adding governance overhead. Yet it remains unclear how easily the loop can be adopted at scale.
