DigitalOcean opened public previews of two services meant to close the gap between deploying an agent and improving it. Agent Learning Services and Agent Data Services sit on top of the company’s Inference Engine and Managed Agents.
Learning Services bundles three pieces. Insights records what an agent did. Signals finds the failures and performance gaps in those records. Simulations run realistic conversations to test a proposed fix before anyone ships it.
The company’s own framing is sequential: the inference layer let agents think, the managed runtime let them act, and this layer lets them improve on the customer’s data.
The write-back is the interesting part
Agent Data Services indexes documents, video, audio, logs and session transcripts into one searchable catalog with governance and access controls.
Signals writes sessions into it, and the optimizer draws training and test data from it only where a customer explicitly opts in. LakeOps co-founder Jonathan Saring noted that traces and tool calls stop being mere logs once evaluation reads from them.
An enhanced inference router rounds out the stack. It picks models by task requirement, weighs cost against quality, and reruns the same evaluations when a new model version lands.
A crowded shelf
Evaluation and observability vendors have made the same pitch for two years. DigitalOcean’s advantage is proximity, since its customers already rent the compute the agent runs on.
