Summary
Signals, which identifies failures and performance gaps;
News
DigitalOcean Launches Learning and Data Services to Enable Self-Improving AI Agents
Oct 26, 2026
2
min read
Follow us on
DigitalOcean has announced the public preview of Learning Services and Agent Data Services, enabling developers to build, evaluate, and continuously improve AI agents within a unified, AI-native, cloud platform.
The new services build on DigitalOcean's Inference Engine and Managed Agents to help agents learn from real-world interactions, identify weaknesses, test improvements, and leverage company data, the company says:
We built Inference Engine so agents could think, Managed Agents so they could act, and now Agent Learning Services and Agent Data Services so they can improve, on one open cloud and on the customer’s own data.
Learning Services combines three capabilities: Insights, which records agent activity; Signals, which identifies failures and performance gaps; and Simulations, which tests potential improvements through realistic conversations simulated before deployment.
Although separate from Learning Services, the enhanced Inference Router plays a key role in DigitalOcean's approach to continuously improving agents. It optimizes model selection based on task requirements, balances cost and quality, and reruns the same evaluations on new model versions.
Agent Data Services provides a secure, managed data foundation for AI agents. It processes and indexes documents, video, audio, logs, and agent sessions, making this information searchable and accessible through a unified catalog with built-in governance and access controls. It also integrates with Learning Services to support continuous agent improvement:
Agent Learning Services are built on [Data Services]: Signals writes sessions into it, and Optimizer draws its training and test data from it only where the customer expressly opts in.
Commenting on the announcement on LinkedId, LakeOps co-founder Jonathan Saring observed:
The write-back loop is the part I find most interesting. Once traces, tool calls and outcomes feed evaluation or learning, they stop being “just logs” and become production data.
On a similar note, senior engineer Deepanshu Udhwani argued:
The write back what happened part is underrated compared to the read side everyone talks about with agent data layers. Most teams build great retrieval and skip logging outcomes properly, so six months later nobody can tell which tool calls actually worked versus which ones just looked fine in the moment.
Summarizing DigitalOcean's launch clip, TinyFish co-founder Keith Zhai, emphasized how Signals and Simulations can turn a failed user session into a fix before hundreds of other users encounter the same issue:
Signals and Simulations pull that session into a fix before the four-hundredth person hits the same wall. [...] Green tool calls next to abandoned users is still where most agent stacks break. The bottleneck is whether anyone can read the session the user actually lived.
DigitalOcean Insights, Signals, and Evaluations, including Simulations, are available in public preview. Inference Router optimization and Agent Data Services are available in private preview.
About the Author
Sergio De Simone
The InfoQ Newsletter
A round-up of last week’s content on InfoQ sent out every Tuesday. Join a community of over 250,000 senior developers.
View an example
Source: InfoQ
