DigitalOcean launches Signals and Simulations to turn failed agent sessions into tests

AIDigitalOcean launches Signals, which watches live agent sessions with rule-based checks for signs like repeated user messages, requests for a human, and looping tool retries. Flagged sessions go to Simulations, which replays them with a fake user against the fixed agent before release. The post says teams no longer need to pay an LLM to grade every conversation or write test cases by hand.

AIDigitalOcean launches Signals, Simulations, and Insights in public preview, and Agent Data Services in private preview. Signals reads live agent traffic and flags failed sessions, such as a successful tool call that queried the wrong region. Users can turn a flagged session into a test that Simulations replays with a proposed fix.

AIDigitalOcean's Signals flags production agent sessions where users got frustrated or tools failed and converts them into evals with one click. Simulations then replay those sessions against a fix on the agent's real harness before it ships.