Autoheal AI Inc. has raised $7.9 million in seed funding to develop a platform for evaluating, governing and improving AI agents used across enterprise software development. Innovation Endeavors led the round, with participation from Emergent Ventures, U&I Ventures, Darkmode Ventures, Batch Ventures and Param Hansa Values, according to SiliconANGLE.
A control layer for agent fleets
The startup is targeting companies that use multiple specialized agents in software factories, where AI-generated code can increase delivery speed while also contributing to production incidents, software vulnerabilities and rising token costs. Autoheal says its platform gives those agents shared engineering context, evaluation infrastructure, cost controls and security constraints.
Evaluation and automated repair
Autoheal’s system is designed to run inside a customer’s private cloud and connect with coding agents, code repositories, continuous integration and continuous delivery pipelines, and observability tools. Its shared engineering context graph is maintained by two specialized agents: an Evaluator agent rates other agents using signals such as CI failures and incident reports, while a Healer agent proposes changes to their models, prompts, tools and skills.
The proposed fixes are version controlled in Git, checked against historical benchmarks and subject to human approval, according to the company’s description. That workflow is intended to keep automated changes auditable rather than allowing agents to alter production-facing behavior without supervision.
Early customers and claims
Autoheal had operated in stealth mode before announcing the funding. SiliconANGLE reported enterprise customers including Normura Holdings Inc., AvidXchange Inc. and Empiric Earth Inc.; those companies say the platform has reduced incident-resolution times and saved thousands of engineering hours. The source does not provide independently measured results or financial details about customer contracts.
Private data remains the next step
The company plans to develop reinforcement-learning techniques for training customer agents on private engineering data. Its longer-term proposal is to create enterprise-specific small language models that run within customers’ private cloud environments, potentially supporting software-agent fleets with more specialized knowledge and lower costs. Autoheal also says the architecture could eventually extend into data and security engineering, but those expansions are goals rather than reported product availability.
AI-generated editorial illustration; not a documentary photograph.
Source: Autoheal raises $7.9M to evaluate and fix AI agents with… AI agents
SIGNAL TITANS
Leave a Reply