Scripted testing is shrinking, but testing agents themselves is a growing need. Agents can write test cases, run regression suites and triage bug reports. Exploratory testing, judging whether a release is fit to ship, and testing how an AI agent behaves against real cases stay with a QA tester. Testers who move toward evaluating agents are well placed.
What can an agent take from QA work?
Much of a software tester's routine follows a pattern an agent can learn:
- Writing test cases from user stories and acceptance criteria, at depth 2, for a tester to check.
- Running regression suites and summarising failures, at depth 1 or 3.
- Triage of bug reports: removing duplicates, adding reproduction steps, assigning severity drafts.
- Updating test documentation after a release.
If your role is mainly executing scripted tests by hand, that part is going.
What stays with a tester?
- Exploratory testing. Finding the bug nobody wrote a test for needs curiosity and product sense.
- Release decisions. Whether a known defect blocks release is a trade-off with business risk.
- Severity judgement. An agent can propose severity; a person decides what it means for users.
- Testing the agents. Products increasingly contain AI agents, and someone must check that they behave correctly on real cases.
That last point is where the job is growing.
Why do agents need testers?
An AI agent is not tested like ordinary code. The same input can produce different outputs, and correctness depends on whether the agent used the right sources, stayed inside its scope and escalated when it should. In Pryme, Attest certifies an agent by replaying it against real cases from the customer's own operation and scoring every answer against a floor. There are seven gates, and each must clear on its own: grounded, refuses, depth, escalates, scope, adversarial and stable.
Each gate needs test cases. Does the agent refuse a request outside its role? Does it escalate when a customer threatens self-harm? Does it refuse to make a refund above the limit? Does performance drift when sources change? Designing those cases, judging the answers and reading why a run failed is QA work, done to a higher standard.
The OWASP Top 10 for LLM Applications lists risks such as prompt injection and excessive agency. Testers who can design cases that probe those risks are useful to any team shipping agents.
What does this look like for a tester?
Take a tester we will call Sam at a software company that sells a helpdesk product with a built-in agent. The agent writes Sam's regression tests now. Sam's week has shifted to building a set of difficult customer conversations: ambiguous refund requests, attempts to trick the agent into revealing another customer's data, messages with instructions hidden in pasted text.
When a run fails the scope gate, Sam reads the record, finds the agent followed an instruction embedded in a customer's attachment, and writes it up with a reproduction. That failure would have been invisible to a scripted test.
Which controls make testing agents meaningful?
| Control | What the tester checks |
|---|---|
| Grounding in sources | Every claim traces to a source the agent was given |
| Refusal and scope | Out-of-role requests are refused |
| Escalation | Risky cases go to a person |
| Drift | Results stay stable when sources change |
| Record | Every verdict is filed and can be reconstructed |
Which skills make a QA tester harder to replace?
Learn to design test cases for behaviour, not just function. Study how agents fail: hallucinated facts, ignored limits, instructions smuggled in through data. Practise writing clear failure reports with the case, the expected behaviour and the evidence.
How a QA tester can move into AI operations covers the wider move, and the jobs a QA tester can get with AI agent skills lists the titles. Reviewing agent decisions is close to testing; see moving from QA tester to AI agent reviewer. The guide to AI agent jobs and careers sets the context.
Where the Academy fits
For a QA tester, Pryme Intelligence Academy's Builder track is the closer fit: about 24 hours plus a capstone agent certified in Attest to depth 2 or higher, leading to Certified Pryme Builder. Start with the free Foundations track first, which covers depth and certification. Every lab runs in a free Build workspace, so you can practise without an employer.
Questions people ask
Is manual testing finished?
Scripted manual testing is shrinking. Exploratory testing and testing of agent behaviour need people and are growing in importance.
Do I need to learn machine learning?
Not deeply. You need to understand how agents use sources and tools, how they fail and how to write cases that expose those failures. That is testing skill applied to a new kind of system.
What should I show employers?
A set of test cases you designed for an agent, the failures they found and how you reported them. If you hold a certificate, employers can check it at source.