v2026.10.08
TL;DR
You can now agree on a plan with Thunders AI before it writes a test, so the test matches what you meant. Thunders MCP also gets new tools that help AI agents write, find, and update your tests.
FEATURE
- Thunders AI plan: You now check the plan before Thunders AI writes a test. Thunders AI shares the steps and the defaults it will use, asks only what you must decide, and offers to browse the site first.
- Thunders AI questions: Thunders AI now shows the questions that it asked next to the answers that you selected.
- Thunders MCP: new
get_authoring_guidetool teaches AI agents using Thunders MCP about the step types, the variable syntax, to author reliable tests. - Thunders MCP: AI agents now find the right test case quickly, also in a large project. The new
search_test_casestool searches by text, label, and status. - Playwright export: An admin can now export the web test cases of a project as a Playwright project. Use the
/api/admin/script-exportendpoint with a machine-to-machine API key of your organization.
IMPROVEMENT
- Thunders MCP: You now control the cost of step generation. Agents tell you the cost with and without live browsing, and let you choose.
- Shared runs: A link to a shared run now opens the run in the app when you have access to it.
- Guided recovery in Thunders AI: When a browse or agent step fails, Thunders AI now tells you why and gives you the next action: try again, browse again, or set up the app.
- Test assets: You can now filter the Uploaded on column of test assets by uploader and by date.
- Stuck runs: Thunders now ends a test run that stays in the Running, Waiting, or Queued state for too long. The run shows as Interrupted, and its schedule can run again.
FIX
- Element detection: Thunders now finds a file input that sits next to a drop zone. Thunders now refuses a click on a disabled element. Thunders now reads the live value of a form control inside an iframe. Thunders now finds an element when its label has an ampersand or a quote mark.
- Reports: The used persona for execution now appears in the exported test runs report.
- Improved performance for long lists: Lists with more than 500 items now load in full.
Migration steps
- Pull the new images and redeploy, starting with the API Service (it applies any database updates automatically).
upgrade urgency
Recommended
Ready to ship faster with smarter testing?
