Example · QA Engineer
Writing the tests is cheap now. Knowing what they prove still takes you.
24 of your 40 hours still need you
- 13 stay yours
- 11 faster with AI
- 16 tools could take
A typical week at the middle pace. Your plan uses your own tasks.
Is QA testing a dying career?
By our task-by-task count, 24 of a typical 40-hour week as a QA engineer still need a person. Tools could already take about 16 hours. 41% of your week is the part you were hired for, and it is the part a model does well: tests from a ticket, cases from acceptance criteria, the regression script, the bug write-up. If you have watched a test-generation pilot, you have seen it write dozens of tests in minutes.
Read more
The same pilots show what holds. A good share of the green tests prove nothing, and it takes a person who knows the product's domain to tell which ones. Your job now is deciding what is worth testing, telling a real defect from a business rule working as designed, and being the name on "go" the night before a release reaches customers. Many QA engineers already do all three, informally, and their title does not say so. Two of the three are usually the hardest parts of the week: the red pipeline that is not a real bug, and release night. This report helps you hold the pen in those rooms.
The hard truth
Testers earn less than the developers they work with.
QA pays well, but BLS data show it pays less than development and is growing more slowly. The median QA analyst earns about $30,000 less than the median developer.
Sources (2)
- Median pay in May 2025 was $104,300 for QA analysts and testers and $135,980 for software developers. BLS Occupational Outlook Handbook, Software developers, QA analysts and testers, 2026
- BLS projects 6% job growth for QA analysts and testers versus 10% for developers, 2025 to 2035. BLS Occupational Outlook Handbook, 2026
What stays yours
"Is this a bug, or is that how the domain works?"
The answer lives in domain rules (tax rules, a customer's configuration) and years of watching what broke. A model can quote the rule, but it cannot know which edge case took down a release last year.
Knowing which edge cases break your product's core engine
A wrong calculation can look exactly like a right one until it reaches a customer. Your test data encodes where the bodies are buried, and usually nobody wrote that down.
Saying "go" on release night
Where your company is audited, the sign-off is evidence in a change-management control that customers' auditors rely on. Someone named has to be answerable for it. Often, informally, that is QA.
Finding the generated tests that tested the wrong thing
A green test that asserts nothing is the most expensive kind of mistake, because it hides the problem. The person who can read a test and say "this proves nothing" is the constraint on every test-generation tool.
Working a customer escalation from support
The customer's configuration, their calendar and the support rep's relationship are all undocumented. A model can form a guess, but it cannot call the customer's admin.
What's next for a QA engineer?
Plan A is your best move inside the work you do now. Most sessions in the 13-week plan build toward it.
- A
Become the named release-risk owner
Why it fits
Why you. If you already say "go" and already know which edge cases break, a record of catching domain bugs before customers do is standing nobody else at your level has. The gap between what you do and what your title says is the smallest of the three moves.
The gap. Usually no written risk assessment exists. The sign-off is a checkbox, and nobody has named a QA-level owner, so it can be anybody.
The trade-off. You become the person blamed, in writing, when a release goes wrong. The stress stays the same, but you have more authority on paper. You will also ship fewer tests for a quarter, and someone will count.
Other moves that fit
These are also good bets in this field. Read them if Plan A doesn't suit you.
Own the quality of machine-written tests
Why it fits
Why you. Finding the generated tests that tested the wrong thing is a skill with no name yet, and test-generation pilots are going team-wide.
The gap. You have no mutation testing, no written standard for what a test must assert and no agent guidelines yet. The flaky rate is a guess rather than a measured number.
The trade-off. You will read more than you write, and some weeks it feels like proofreading for a machine. It sits close to a senior SDET's territory; if they want it, you need the release-risk move more than this. An executive can still buy a "test quality" dashboard; the plan makes that deal worse than keeping you, though it stays possible.
The portable hedge: domain-correctness specialist
Why it fits
Why you. Knowing which edge cases break a domain engine (payroll, tax, benefits, fintech) is worth money at every company in that domain. Domain QA is the corner of this occupation where the hiring bar is domain context.
The gap. You have no public artifact yet and no idea what the market pays.
The trade-off. Leaving resets the asset this report says protects you. Your first year anywhere new is your most exposed: context starts at zero and test-writing is all anyone can measure.
What's a plan B for QA testers?
Plan B is a job AI is creating that these skills already fit. One session a week in the plan trains you for it, so it is ready if your job shrinks or you want a change.
BPlan B, picked for this role
AI evaluation specialist
Deciding what a test actually proves is your daily work, and evaluating model outputs asks the same question of a new kind of system.
Other options
Agent simulation scenario designer
Writing edge-case scenarios from real defect reports carries straight into designing test conversations that push an agent until it breaks.
Domain AI red teamer
Hunting for the input that makes the release fail is close to red teaming, once you add a public write-up of your findings.
In your own plan you can switch to any of the 47 roles.
The first two weeks
The first 2 of 13 weeks. Most weeks have three sessions of 45, 60 and 90 minutes, plus one 40-minute Plan B session. In weeks 5 and 9 the third session is your Friday release, so move those weeks to match your release calendar. By week 13 you'll have two signed release risk assessments, a review checklist for AI-written tests, and a memo asking to be the named release-risk owner.
- Week 1: Measure your week
What you do
- ALog one week of your time. Sort each hour into the tasks listed in this report.
- APull numbers for your last three releases. Count tests run, flaky tests that fail at random, and bugs that reached support.
- AOpen a batch of AI-generated tests from a pilot, or generate some with an approved tool or on a public repo. Sort each test into proves something, always passes no matter what, or tests the wrong thing.
- BTake Anthropic's AI Capabilities and Limitations course. Pair each AI failure type with a bug type you already know.
- Week 2: Check the AI tool rules
What you do
- AAsk your manager in writing which environments AI agents may run in, and who decided.
- AFind who owns the change-management control, the audited rule for how releases ship. If you are audited under SOC 1 or SOC 2, two common audit standards, ask for last year's control description.
- ACompare the control text with what you actually do on release night. Write one page on the differences.
- BDo the first half of Claude Code in Action, a course on Anthropic's coding agent, using a public open-source repo. Have the agent write three tests.
Three courses from the plan
Checked 2026-09-22.
- ISTQB Certified Tester Foundation Level (CTFL) v4.0ISTQB via ASTQB / AT*SQA (US exam) · $229
Only if the postings you want ask for it. Some HR screens do. It teaches vocabulary more than judgment.
- ISTQB Certified Tester Advanced Level Test Automation Engineering (CTAL-TAE) v2.0ISTQB via AT*SQA · $249 when last seen; price not shown on the page today
No. It certifies the exposed half.
- ISTQB Certified Tester Test Automation Strategy (CT-TAS) v1.0ISTQB via AT*SQA · $199 when last seen; price not shown on the page today
Maybe, months 4–6, if “Own the quality of machine-written tests” lands and you want the strategy vocabulary.
Risks beyond AI: highCost-center risk
Forget AI for a moment. QA is an engineering cost center and has been the first team cut in every efficiency wave since continuous delivery, because its output is invisible when it works. Yahoo eliminated its dedicated QA teams in 2014–15, before any of today's tools existed, and the 2022–23 tech layoffs did it again. We score this separately because blaming AI for it would get the remedy wrong.
Where this could be wrong
- Exploratory testing may be more exposed than we scored. Our capability score assumes agents explore breadth more than risk. If an agent seeded with your escaped-defect list finds real bugs in Week 3, that task becomes more exposed and owning the quality of machine-written tests becomes urgent.
- We assumed the release sign-off is yours informally. If a release manager already signs and you merely report, the release-risk move means joining the sign-off rather than owning it, and that task gives you less protection.
- The BLS decline may be a revision artifact. The 201,700 and 187,600 figures come from different projection vintages; the direction matches job boards, but the size is not a clean measurement.
- A senior SDET, a developer who writes test code, could claim the machine-written tests move fast. More experience and the pilot's prompts in hand make a stronger claim than yours. If they want it, focus on release-risk ownership and the domain specialist hedge.
Also matched: qa engineer, test engineer, software tester. Research checked 2026-09-22. Page updated .