Developer tools · building
AgentTrace
OTEL-native harness so production agents keep traces, evals and runs tied together.
Challenge
Teams ship agents on a model alone, then lose the thread when a run fails: no shared harness, weak traces, and no durable link between what the agent did and why.
Stack
OpenTelemetry · Next.js · Agent harness
Result
Pre-launch waitlist for an OSS agent harness
Approach
- 01
Framed the product around the full harness (tools, memory, evals and telemetry) instead of model access alone.
- 02
Made OpenTelemetry the default so traces land in systems teams already run.
- 03
Tied every artifact back to the run that produced it, so debugging stays grounded in production behavior.
Outcomes
- 01
A clear wedge: Agent = Model + Harness, with harness as the product.
- 02
A waitlist-ready landing surface that explains the platform before the OSS cut ships.
- 03
A reusable pattern for AI infra products that sell clarity, not novelty.