Make sure your AIdoes what it promises.
We review how your AI handles tasks from your brief or public materials. You get evidence of what needs fixing.
Example test: three simulated users ask an AI operations employee to check requests, update records, and notify the right team. Each takes a different route. Two finish the task. For the third, the AI marks the request complete, but no handoff is recorded in the receiving team’s queue. A cover lifts to reveal a break in the failed route. After your team fixes the issue, Unblo runs a new test and confirms the request reaches the right team.
Our founders worked on AI since 2018 at:
“We are already fixing some of the issues Unblo found. Our own tests didn’t catch them.”
Jean Roman
VP Operations, Stonly
Testing Stonly’s AI Knowledge Agent
Design partner since July 2026
Get the evidence to fix what your AI gets wrong.
Unblo tests an AI project coordinator in Northstar. The task: create a launch handover from Alex’s notes.
- Alex asks for two tasks, each assigned to its named owner.
- The AI says both tasks are saved and assigned correctly.
- Unblo checks the saved tasks against the notes.Verify backup is missing. Rehearse migration is assigned to Maya instead of Leo.
- Environment
- A test account in Northstar.
View full example
Know what actually got done.
Unblo checks the saved result against each requirement of the task.
Partly completed.1 of 3 requirements met.
What Unblo found
-
Met: Saved in Northstar.
Requirement: The handover is saved.
-
Not met: Verify backup is missing.
Requirement: Both launch tasks are created.
-
Not met: Maya is assigned instead of Leo.
Requirement: Each task has the correct owner.
Know what needs fixing.
Each issue names the gap between the request and the saved work.
Correct the saved handover.Two issues to address in this test.
-
Missing task
Create Verify backup.
Assign it to Maya.
-
Incorrect owner
Reassign Rehearse migration.
Change the owner from Maya to Leo.
Back every issue with evidence.
Review the source notes and saved result behind each finding.
Rehearse migrationThe saved owner does not match the notes.
- Source notes
- Leo
- Saved task
- Maya
See if your fixes worked.
Change a prompt, model, source document, or integration. Then test again to see what improved and what still fails.
Refine your product,
then run again.
- Build
Share your product context and documents. Unblo creates simulated Test Users and Tasks.
- Run
Test Users attempt Tasks in your product, each in a separate browser with a test account.
- Evaluate
See which requirements were met. Each Issue links to the steps and evidence behind it.
- Report
Review the evidence with your team and decide what to fix. Compare results from fresh attempts after changes.
Your first test is free. We’ll set it up with you.
Included in your first test
- One focused test
A task we choose together, tested in your product.
- An evidence-backed Report
What worked, what failed, and the evidence behind each finding.
- A founder walkthrough
Review the findings together and decide what to fix.
What we need from you
- Test access
A test account that lets us use your product.
- A task and success criteria
What you want tested and what a successful result looks like.
Help shape what we build next.
Work directly with the founders and share feedback each week to influence what we build next.
After your free test, decide together whether to continue as a design partner.
Start with a free test of your AI product.
Get a Report and a walkthrough with the founders. We’ll email you to agree what to test and arrange access.


