Test design — not executedBrowser and computer agents
How to test agents that complete browser workflows
Use a bounded task with observable results. The candidates below are research starting points, not tested recommendations.
Source-linked research and editorial guidance. Capabilities are not independently tested unless a profile explicitly includes a reproducible test record. Sources checked November 13, 2026.
- Inputs
- A sandbox form, test account, permitted values and a clear final state.
- Expected deliverable
- Completed fixture records plus screenshots or event logs showing the resulting state.
- Procedure
- Map the workflow
- Mark irreversible steps
- Run the agent
- Verify the backend or confirmation state
- Repeat with a validation error
- Approval boundary
- Require a separate approval before sending external messages, submitting production records or committing purchases.
- Acceptance criteria
- Correct final state, duplicate avoidance, recovery from validation errors and intervention count.
- Failure modes
- Trusting the agent narrative, accepting a clicked button as success and retrying a submission that may already have succeeded.
Candidates to investigate
Editorial topic relevance — task support not verified. Research starting points — none of these has been run against this task.
Sources
- skyvern.com (opens in a new tab)
- skyvern.com/pricing (opens in a new tab)
- browser-use.com (opens in a new tab)
- browser-use.com/pricing (opens in a new tab)
- ycombinator.com/companies/browser-use (opens in a new tab)
- browser-use.com/posts/seed-round (opens in a new tab)
- github.com/browser-use/browser-use/blob/main/CLOUD.md (opens in a new tab)
- github.com/browser-use/browser-use/blob/main/skills/open-sour… (opens in a new tab)
- simular.ai (opens in a new tab)
Something out of date? Suggest a correction