CREAO Raises Additional $10 Million, Enabling AI Agents to Self-Modify Their Harnesses After Each Failure.

Beating AI Express: Enterprise agent platform CREAO has secured a new strategic funding round of tens of millions of U.S. dollars, bringing its total raised capital to over $30 million. CREAO’s current focus is on iterative self-improvement of its "Harness" layer, which enables agents to continuously adjust their execution systems based on errors and feedback from real-world tasks. The Harness layer sits above the underlying model, governing how agents operate, covering functions such as task planning, tool invocation, memory management, model selection, and error handling. CREAO has already implemented partial automated error correction: after an agent completes a task, the system automatically reviews the output; if issues are detected, it generates a ticket, and an AI then investigates the root cause, drafts fixes, and validates the results. The new version is first tested with a small volume of real user traffic, with automatic rollback if performance declines. CREAO’s team has previously unveiled this system. Moving forward, CREAO plans to extend this improvement to more components of the Harness layer. For example, if a category of agents repeatedly makes errors with a specific tool or task workflow, the platform can adjust its entire execution approach based on data from numerous real-world tasks, and newly created agents of the same category can directly leverage these optimizations. Looking ahead, CREAO also plans to integrate some of these learnings into its models. The company plans to conduct post-training for open-source models tailored to vertical scenarios, so that common tasks can be handled more by lower-cost specialized models, while complex tasks will still call for state-of-the-art models.