An AI agentAn AI system that doesn't just answer once but works toward a goal in a loop — taking actions, reading the results, and deciding what to do next.Full definition → can return HTTP 200 and still fail — wrong tool choice, stale multi-agentUsing several AI agents on one problem — splitting work in parallel, checking each other, or filling different roles like planner and reviewer.Full definition →context windowThe maximum amount of text a model can consider at once — its working memory for the current conversation or task.Full definition →, token-burning retry loops — failures that infrastructure telemetry never surfaces; agent-level traces capture model calls, tool executions, approvals, and per-turn token spend.
Subagent invocations nest inside the parent's trace waterfall next to Workers infra spans (D1, KV), so time and cost are attributable to model vs tool vs infrastructure in one view.
Session replay reassembles the whole turn from recorded data — system instructions, thinking, tool args and results — and payload capture is toggleable (storeMessages/storeTools) for sensitive data.
Launch support covers OpenTelemetry-compatible harnesses (Think, Flue, AI SDK) with export to any OTel destination, positioning traces as the feed for self-improving agent loops.
Terms in this piece · Glossary
sandbox — An isolated environment where AI-generated code or agent actions run without being able to touch anything real.
AI agent — An AI system that doesn't just answer once but works toward a goal in a loop — taking actions, reading the results, and deciding what to do next.
multi-agent — Using several AI agents on one problem — splitting work in parallel, checking each other, or filling different roles like planner and reviewer.
context window — The maximum amount of text a model can consider at once — its working memory for the current conversation or task.
Why it matters
Agent-level cost and tool-call tracing is the missing instrument for anyone operating agents in production rather than demoing them.
Key quotes
“An agent can return HTTP 200 and still fail. It may choose the wrong tool, pass stale context to a subagent, or spend tokens in a retry loop.”
“With agent-aware traces, you can now understand exactly what your agent is doing and what it costs: every model call, tool execution, and token is measured and presented here.”
“Plug this data into your agent development lifecycle , and you suddenly have autonomous, self-improving agents.”
“This is the vision for Cloudflare Agents: one place to deploy, observe, and continuously improve every agent you run.”
