AgentClash

Agent testing

AI agent testing that starts from real failures

AI agent testing should feel closer to software testing than prompt tweaking. AgentClash turns traces, golden datasets, and escaped failures into repeatable tests with baseline comparisons.

live eval
gate: pass

Candidate

92correct patch, low cost

Baseline

88stable reference run

Control

73missed edge case

replay timeline

1loaded task inputs and tool policy
2ran sandbox actions and captured artifacts
3scored trajectory and validator evidence
4attached scorecard and release verdict

ci verdict

Candidate clears release gate

Correctness improved, latency within budget, and required artifacts were preserved for review.

agentclash run create --follow

What production-grade agent testing covers

Built for reviewable agent decisions

Testing agents means checking behavior across the whole run, including tool calls, retrieval, cost, latency, and artifacts.

OpenTelemetry-compatible trace import

Pinned datasets and golden test cases

Baseline versus candidate regression checks

Replay trails for tool calls, outputs, and artifacts

Scorecards for correctness, cost, latency, and evidence

CI gates for prompt, model, RAG, and tool changes

Workflow

Test loop

Import the evidence

Start from OpenTelemetry traces, curated datasets, support transcripts, or a real failure your team already saw.

Pin the baseline

Record the current accepted behavior so every prompt, model, RAG, or tool change has a fair comparison point.

Replay the evidence

Inspect tool calls, outputs, artifacts, latency, cost, and judge evidence when a candidate gets worse.

Gate the release

Compare candidate and baseline runs, then fail CI before a regression reaches users.

FAQ

AI agent testing FAQ

Is AI agent testing different from LLM testing?

Yes. LLM testing often means scoring one response. AI agent testing evaluates plans, tool calls, artifacts, recovery behavior, and whether the task actually finished.

Can non-ML engineers review agent test failures?

Yes. Replay timelines, artifacts, and scorecards are designed for reviewers who need to understand what changed without reading raw model traces alone.

How do we keep tests from getting stale?

Promote escaped production failures into datasets, challenge packs, and regression suites so the same mistake stays covered after the next model swap.