DigitalOcean Launches Agent Learning and Data Services: What Is in Public Preview, What Is Private, and What Is Missing
DigitalOcean wants agents that improve with every run: record each step, flag failed sessions, test fixes on replayed conversations, then route each task to the best-scoring model. None of it is production-ready yet, and there is no pricing.
TL;DR
DigitalOcean announced Agent Learning Services and Agent Data Services on 12 October 2026. Learning Services has three parts in public preview: Insights records every step an agent takes, Signals flags sessions where an agent stalled, looped, or failed, and Evaluations & Simulations tests changes on realistic conversations built from those sessions. Agent Data Services, a managed, searchable data layer for agents, and an Inference Router optimiser that picks the best-scoring model per task are in private preview. DigitalOcean says preview services are not supported for production and has given no pricing or general availability date.
DigitalOcean's announcement adds a loop around the agents developers run on its cloud: watch what agents do, find where they fail, test a fix, and only then ship it. The company frames the goal as agents that get better with every run.
What launched, and its status
| Product | What it does | Status |
|---|---|---|
| Insights | Records every step an agent takes | Public preview |
| Signals | Flags sessions where an agent stalled, looped, lost the user, or failed | Public preview |
| Evaluations & Simulations | Tests proposed changes on realistic conversations built from flagged sessions before real users see them | Public preview |
| Agent Data Services | Managed data layer: transcribes and describes video, reads and indexes documents, assembles logs and agent sessions with configurable redaction, all searchable by meaning in one call | Private preview |
| Inference Router optimiser | Picks the best-scoring model per task for cost, quality, or your own evaluations, and reruns evaluations when new models ship | Private preview |
How the pieces connect
- Signals writes flagged sessions into Agent Data Services, which keeps data in a per-customer catalogue with access governance.
- An optimiser uses that catalogue for training and test data only if the customer opts in.
- Evaluations & Simulations replays flagged sessions as test cases, so a fix is checked against the conversations that actually failed.
- DigitalOcean says it plans to extend the automated loop to agent workflows, alongside existing human approvals. That is a plan, not a current feature.
What's missing
There is no pricing for any of the five services and no general availability date. The only named customer is Amplitude, whose CEO Spenser Skates said it chooses "the model for each task based on the results it delivers, not just its cost". DigitalOcean says it has more than 680,000 customers, but not how many are using Managed Agents.
What this means if you're choosing tools
If you already run agents on DigitalOcean, the public-preview trio addresses a real gap: most teams can't see why an agent failed or test a fix on the failing conversation. Don't build production on it yet. If you're comparing platforms, the opt-in rule for training on your sessions is worth checking against any rival's defaults.
AI Industry Reporter
Priya covers model releases, industry announcements, and the gap between what labs claim and what independent evaluators actually find. She reads the primary source - the paper, the system card, the benchmark org's own statement - before writing a word.
More on AI Automation & Agents
Ai2's StateBench: Frontier Models Fall to 13% or Lower When Long Documents Keep Changing
Priya Nair · 4 min
Anthropic's Mythos 5 Report: The Model Kept Attacking a Real GitHub Project 98% of the Time Once It Started, Newer Claude Models Far Less
Priya Nair · 5 min
Anthropic Opens a Waitlist for Claude Duet, an Always-On Agent for US Pro and Max Subscribers
Priya Nair · 5 min