title: "Cloudflare launches Agents with tracing and replay debugging" slug: "cloudflare-launches-agents-with-tracing-and-replay-debugging" published: "2026-08-16" beat: "Launches" tags: ["Launches", "Tools"] creator: "Agentry Newsroom" editor: "Susanne Sperling, Editor — Human in the Loop" tools: ["Claude (Anthropic)", "Perplexity Sonar"] creativeWorkStatus: "verified" dateReviewed: "2026-08-16" aiActArticle50: "compliant" humanView: "https://agentry.news/launches/cloudflare-launches-agents-with-tracing-and-replay-debugging" agentView: "https://agentry.news/agent/cloudflare-launches-agents-with-tracing-and-replay-debugging"
Cloudflare shipped Agents on August 4, 2026, a developer platform that includes OpenTelemetry-compatible tracing, session replay, and human-in-the-loop approval controls to help teams debug agent beha
Drafted by an AI agent. Verified by Susanne Sperling, Editor — Human in the Loop. AI policy.
Cloudflare launched Agents on August 4, 2026, introducing observability tooling designed to help developers debug and audit multi-turn agent behavior in production environments. The platform includes two core observability views—session replay and trace—alongside human-in-the-loop approval workflows Cloudflare Blog.
The session replay feature records full conversation context across agent turns without re-executing the agent. According to Cloudflare's official documentation, the Messages tab replays recorded data to show messages, reasoning, tool calls, and subagent activity Cloudflare Agents Docs. This eliminates the need to reproduce failures or rerun expensive model calls when investigating agent decisions after deployment.
Agent traces are available for the Agents SDK and capture agent turns, tool runs, and token usage alongside Workers runtime operations Cloudflare Changelog. The tracing system is built on OpenTelemetry, the open standard for observability instrumentation, and supports export to OTLP-compatible destinations for integration with third-party monitoring platforms. This approach enables teams to send agent telemetry to existing observability stacks without vendor lock-in.
Beyond observability, Cloudflare's Agents system includes human-in-the-loop approvals, allowing teams to require manual sign-off before agents take sensitive actions Cloudflare Blog. Combined with session replay and tracing, this creates a checkpoint for high-stakes agent operations where auditability and control are critical.
The launch addresses a core friction point in agent deployment: teams building autonomous systems need visibility into what agents decided, why they made those decisions, and what actions they took. Session replay eliminates costly re-execution, while OpenTelemetry export prevents teams from being locked into Cloudflare's proprietary monitoring. The human-approval layer reflects growing enterprise demands for oversight in agent deployments handling financial, legal, or operational decisions.
Cloudflare's positioning as an infrastructure provider—rather than an AI model or application layer—means these observability tools are meant to work with agents built on any foundation model or framework, making the August 4 launch a developer-infrastructure play in an increasingly crowded agent ecosystem.