QA to AI: why testers have a head start

Testing is half the job in AI work now. What carries over from QA, the gaps worth closing, and a 30-day plan.

By MT, AI Solutions Engineer. Updated 2026-10-03.

Key takeaways

  • AI writes more of the code, so proving it works has become a bigger share of the job.
  • Test cases, edge cases and regression thinking transfer almost directly to evaluating AI.
  • The gaps are scoping the problem, knowing the tools, and building something yourself.

Why testing just became the job

In April 2026, Google's CEO said 75% of all new code at Google is generated by AI and approved by engineers. That changes where the work is. Writing code got faster; knowing whether the result is right did not.

From my own projects, the effort is shifting toward roughly 40% requirements, 20% coding and 40% testing. That's my estimate, not a published statistic. But I see it every week: the hardest part of an AI project is proving it works on the messy cases nobody picked to make the demo look good. That is a tester's instinct.

What transfers directly

  • Test cases with known right answers. That's exactly what an AI evaluation set is.
  • Edge cases. AI tools fail on the odd inputs: the handwritten form, the date in a different format, the question whose answer isn't in the documents.
  • Regression thinking. Every time you change an AI instruction, something that worked can break. Testers already retest after changes.
  • Bug reports that people can act on. "Fails on 3 of 10 invoices when the total is on page 2" is the kind of finding that gets AI tools fixed.
  • Healthy skepticism. A demo that works on three examples doesn't impress you. That's the right attitude for AI. (How to explain evals simply.)

The gaps worth closing

  1. Scoping. Testers usually receive requirements. In AI work you often write them: who it's for, what goes in, what comes out, what done means. Here's how to scope a project.
  2. The tools. One chat model, one automation tool like n8n or Make, and a way to build a quick app. The toolkit I actually use.
  3. Building the first version yourself. Instead of only testing someone else's build, you build a small one with AI, test it, and show the before-and-after failure rate. That combination is rare and valuable.

A 30-day plan for testers

  • Week 1: Pick a repetitive task with a clear right answer, like sorting tickets or pulling fields from forms. Collect ten real examples and write the expected result for each.
  • Week 2: Build an AI step with six of them, test on the four you held back, and log every failure with its cause.
  • Week 3: Fix the worst failure, add a human-review step for anything the AI isn't sure about, and retest everything.
  • Week 4: Write a one-page case study with the failure rate before and after, and what the review step catches.

Use public or sample data for anything you share. Never use employer or client data in a public project.

How to describe the move

Lead with what you already do best, applied to AI: "Built an evaluation set and review process for an AI ticket triage tool; first run 6 of 10 correct, 10 of 10 handled correctly after the fix." The format is: built [what] that [did what], [result with a number]. Use your own real numbers.

Titles worth searching include AI Solutions Engineer, AI Quality Engineer, AI Evaluation Specialist and AI Automation Specialist. If a description mentions "evaluation," "quality," "reliability" or "guardrails," it's written for you.

Where to start this week

Not sure which skill to work on first? Take the 2-minute readiness quiz. Testers usually score highest on testing, and the plan shows what to build next.

For the guided version, the AI Portfolio Sprint walks you through three projects in 21 days, with the Break-It Rule built into every week.