Agent Experience (AX)

Open Technical → Agent Experience (AX) to review a recorded agent journey alongside site-readiness checks. These screenshots use illustrative local data.

​

Access and starting a run

AX requires Team Max or a higher plan and access to Technical. Starting a run also requires write permission. A visible navigation item does not mean the current plan can run audits. From the history page, choose Run new audit, select Agent and model, and choose a Journey prompt task. Review the prompt for the project domain before choosing Run audit. The report opens while the run progresses. Start another run from the history page; individual reports do not have a run command.

​

Read the overview and history

The overview shows the latest completed journey. Trends compare runs with the same prompt and model; changing either changes the comparison. Use the history row’s status, prompt, model, steps, and duration to choose a comparable run.

• Answer from site: whether the answer used site evidence.
• Answer efficiency: how efficiently the agent gathered the evidence for its answer.
• Followed site links: whether navigation followed the site’s links.
• Steps and Duration: the amount of work and time recorded for the journey.

Inside a report, Site agent-readiness provides technical readiness evidence separate from the journey outcome. A successful run does not guarantee every task or agent will succeed. Check the recorded answer and trace before assigning fixes. Pending checks and incomplete provider analysis are not confirmed failures.

​

Agent answer tab

This is the default report tab. Compare the answer with the saved prompt and model. Inspect the recorded journey and discovered URLs to find where the agent succeeded, took a detour, or stopped. Use replay controls when a recorded trajectory is available. If no answer or trajectory was supplied, treat that as missing evidence. Do not infer a completed task from site-readiness scores alone.

​

All checks tab

Use All checks to investigate the technical conditions behind a journey. Expand a finding to read its observed evidence and recommended action. Work through Priority fixes, then rerun the same task and model after the site changes.

​

By layer tab

Read checks by technical area to find clusters of related problems. An optional signal and a required access check do not carry the same meaning.

​

By status tab

Separate failed and partial checks from pending, error, and not-applicable results. An error means the check lacks a reliable result; it is not proof that the site failed it.

​

Methodology tab

Review the provider rubrics and report provenance before comparing scores. Different rubrics use different checks and weights. Use the captured target and timestamps to understand the scope of the evidence.

​

Where to go next

• Technical overview
• LLM audit
• Crawler logs