Enterprise Agent platform CREAO has completed a new strategic funding round worth tens of millions of dollars, lifting its total financing to more than $30 million.
CREAO is currently concentrating on self-iterating harness systems, meaning it wants Agents to keep adjusting their execution systems based on errors and feedback from real tasks. In CREAO’s setup, the harness is the layer outside the model that controls how an Agent gets work done, including task planning, tool use, memory, model selection, and the way failures are handled.
Part of the automated correction loop is already running
CREAO said it has already put part of that automated correction process into operation. After an Agent finishes a task, the system automatically checks the result. If an issue is detected, it generates a ticket, and AI then investigates the cause, writes the fix, and validates the result.
New versions are first tested with a small amount of real traffic. If performance worsens, the system rolls back automatically. CREAO had previously disclosed this system publicly.
Next step is broader harness-level improvement
The company plans to expand this improvement process to more parts of the harness. If the same type of Agent repeatedly fails on a specific tool or task workflow, the platform can use data from a large number of real tasks to adjust the full execution method. Newly created Agents of the same type can then use those improvements from the start.
Some of that experience may later be moved into model training
CREAO also plans to train part of that accumulated experience into the model itself. It aims to carry out vertical post-training on open-source models so that common tasks can be handled more often by lower-cost specialized models, while more complex tasks are passed to frontier models.