DigitalOcean Opens Agent Learning Preview, With Self-Improving Workflows Still Ahead
Teams can preview tools to trace agent failures and test changes; the managed data layer is private, and automated workflow optimization remains planned.
A feedback loop for deployed agents
DigitalOcean announced Agent Learning Services on October 12, pitching a way for teams to improve AI agents using records of what happens after deployment. According to its English press release, Insights captures agent steps, Signals identifies sessions where an agent stalled or failed, and Evaluations and Simulations test proposed changes before they reach customers. These components are available in public preview. The company also says an enhanced Inference Router can choose models against cost, quality or a team’s own evaluations, but that optimizer is in private preview.
Data access and preview limits
The companion Agent Data Services product is in private preview. DigitalOcean’s English product blog describes a managed catalog that ingests documents, media, structured records and agent traces, then exposes them through search, SQL or an API. The company says access controls govern which people and agents can retrieve each piece of data. Its press release describes automated optimization of agent workflows as a next step that would build on current human approvals; it is not part of the preview already available. DigitalOcean says private-preview services have no service-level agreement or standard compliance certification and may change or be discontinued. The launch establishes the preview offerings, not independent evidence that agents improve themselves in production.