QA testers already think the way agent certification works: build cases from real use, find the edge cases, and refuse to sign off until every check passes. This CV section shows a software employer you can apply that to AI agents, testing what they may do and catching where they fail.
How does software testing map onto agent testing?
Testing an agent is closer to your work than most review roles. You still write cases, run them and report defects. The differences are that an agent's answers vary, so you test for behaviour rather than exact outputs, and that you test for things a normal application never does: whether it refuses what it should refuse, whether it escalates, whether it stays inside its scope, whether its answer is grounded in a source.
Pryme's Attest certifies agents by replaying them against real cases from the customer's own operation, scoring every answer against a floor across seven gates: grounded, refuses, depth, escalates, scope, adversarial and stable. A QA tester reading that list recognises a test plan. The guide on how a QA tester should show AI agent skills on a CV covers the positioning.
How should you fill it in?
Translate your experience into the agent equivalents, and claim only what you have done:
| Your QA work | Agent-era equivalent |
|---|---|
| Test cases from user stories | Cases from real past operations |
| Negative and boundary tests | Refusal, scope and escalation tests |
| Regression suites | Re-running cases after a brief or source change (the stable gate) |
| Defect reports with steps to reproduce | Failed decisions with the case, the answer and the expected verdict |
| Release sign-off criteria | A floor every gate must clear |
Pick three rows where you have strong evidence and write a bullet for each. Add one for any AI or agent testing you have done.
The CV section template
Agent testing and quality assurance
QA tester in [type of software], ready to test AI agents for [grounding / refusal / escalation / scope] and support their certification.
- Built [number] test cases from [real production data / user logs]; found [type of defect]
- Designed [negative / boundary] tests for [feature]; caught [type of failure]
- Maintained [regression suite]; [how often run / what it caught]
- [If applicable:] Tested [chatbot / agent / model feature] for [behaviour]; [result]
Methods: [test design], [exploratory testing], [automation], [defect triage]
Certificate: [name], issued [month, year], expires [month, year], verify: [link]
A filled sample
A sample for an invented tester, Kenji, at an invented software company making booking systems:
Agent testing and quality assurance
QA tester in booking and scheduling software, ready to test AI agents for grounding, refusal, escalation and scope, and support their certification.
- Built around 300 regression cases from anonymised real bookings rather than invented data; found time-zone defects that synthetic data had never triggered
- Designed boundary and negative tests for payments and cancellations; caught refunds issued twice when a cancellation was retried
- Ran the regression suite after every release and every configuration change; reported drift in results to the product owner
- Tested a support chatbot before launch for answers outside its scope; found it giving refund promises the policy did not allow and wrote cases to check every refund statement against the policy page
Methods: test design from real data; exploratory testing; test automation; defect triage
Certificate: Certified Pryme Operator, issued [month, year], expires [month, year], verify: [link]
The last bullet is the bridge. Kenji tested for scope and grounding without calling it that: the chatbot promised something no source supported. That is the same failure an agent certification gate is designed to catch.
What will you need to learn?
Two things. First, how agents are governed in operation, not just tested once: authority earned per agent, limits that expire, a record of every decision. Second, the reviewer's view: what a held decision looks like and how a person decides it. Testers who understand both can design tests that match how the agent will actually be used.
You do not need to learn model training. In governed systems, agents are improved by changing their brief, sources and limits, not their weights, which is good news for testers: those are things you can specify and test.
Where the Academy fits
For a QA tester, the Pryme Intelligence Academy path usually runs Operator then Builder. Operator teaches how a governed agent decides and is reviewed. Builder, about 24 hours plus a capstone agent certified in Attest to depth 2 or higher, is where your testing skills are used most directly.
Questions people ask
Should I go straight for building and certifying agents?
The tracks run in order, so start with Operator. The reviewer's perspective will make your test cases sharper.
How does this compare with a business analyst's or call centre agent's section?
A business analyst defines limits and processes; a call centre agent reviews customer outcomes. See the CV section for a business analyst and the CV section for a call centre agent.