The career ladder
Manual testing is shrinking as a standalone job, and teams are leaning on fewer people who can automate and who understand risk. The ladder below describes how responsibilities typically grow. Titles and pay differ widely between companies and regions, so use it to plan skills rather than salary.
| Stage | What you do | Skills to build |
|---|---|---|
| Manual and functional QA | Run regression passes, explore the product, file and triage bugs. | Test design, bug reports, exploratory testing, a test management tool. |
| Test automation engineer (SDET) | Write UI and API tests, remove flakiness, run tests in CI, build reusable components. | Playwright and TypeScript, locators, REST APIs, CI pipelines. |
| AI agent quality | Test agents that reason and call tools: guardrails, tool mocking, budgets and attack resistance. | Tool-call interception, LLM-as-judge rubrics, OWASP LLM Top 10, failure injection. |
A three-month roadmap
You do not need to become a full-stack developer. Focus on one theme a month. The pacing is a suggestion; go slower if you are working full time.
Month 1: the page, locators and recording
Goal: move from clicking through a product to understanding how it is built.
- Open your browser developer tools on your own application every day and inspect buttons, inputs and dialogs.
- Learn the difference between IDs, CSS classes, accessible roles and test attributes such as
data-testid. - Record a flow with a recorder and read the steps and assertions it produced.
- Rewrite five of your manual test cases as precise step lists with one expected result each.
Month 2: Playwright, TypeScript and reuse
Goal: stop copying steps between tests and run them automatically.
- Learn basic TypeScript: variables, functions,
asyncandawait, objects. - Turn repeated setup such as login and navigation into shared, reusable pieces.
- Export a suite to Playwright and run it from your terminal with
npx playwright test. - Add a CI workflow that runs the tests on every pull request.
Month 3: testing AI agents
Goal: learn to test software whose answers change from run to run.
- Understand how an agent loops between reasoning and tool calls, and what the Model Context Protocol (MCP) is.
- Write tests that intercept tool calls and return mocked responses, so nothing real is touched.
- Set limits on tool calls and run time so a runaway loop fails a test.
- Build a small prompt-injection suite based on the OWASP Top 10 for LLM applications.
Portfolio projects
A repository with clean tests and a clear README says more in an interview than a list of courses. Five projects that cover the ground:
- Multi-role checkout suite. Guest, signed-in and admin checkout flows using role-based locators and no fixed sleeps. Tools: Playwright, TypeScript.
- Reusable sign-in block. One module that logs in, caches the session and cleans up, reused across several suites.
- Pull request quality gate. A CI workflow that runs tests in parallel and posts a summary on the pull request.
- Agent tool-mocking suite. Tests for a support bot that intercept its refund tool, return mocked responses and check argument safety.
- Prompt-injection regression suite. A set of direct and indirect injection attempts against a conversational agent, scored by an LLM judge.
For every project, write down what risk each test protects against. That is the part hiring managers read.
Interview preparation
Two questions that come up often, with the shape of a strong answer. Put them in your own words and your own examples.
"A test fails intermittently in CI but passes locally. What do you do?"
Intermittent failures in CI are usually timing or shared state.
1. Open the trace or video and find the step that timed out.
2. Replace any fixed sleep with a web-first assertion:
await expect(locator).toBeVisible();
3. If tests share data, give each one its own unique records
(for example a new user per run) so order does not matter."How do you test an AI feature when the answer changes every run?"
When the answer changes on every run, assert on properties
rather than exact strings.
1. Structure: does the reply match the expected JSON schema?
2. Meaning: does an LLM judge, using a written rubric, find the
reply grounded in the source material?
3. Actions: if the agent calls tools, intercept the call and check
the arguments before returning a mocked response.More background on these ideas is in test automation foundations and agent testing 101.
Where to practice
You can build all of the projects above with open-source tools. If you would rather skip setup, CertiQA's 14-day trial gives you a place to record web flows, export them to Playwright, and build agent tests as plain steps with mocked tools. After the trial, access pauses until the account subscribes, so keep your exported code in your own repository.
FAQ
Can a manual tester become an SDET without a computer science degree?
Yes. Many engineering teams value product knowledge, test design and exploratory instincts as much as formal credentials. A practical path is to learn how the browser exposes page elements, pick up a modern tool such as Playwright, and build a few projects you can show. How long it takes depends on your starting point and how much time you can give it, so treat any fixed timeline as a guide only.
What portfolio projects should an aspiring automation engineer build?
Three good starting points are an end-to-end checkout suite with resilient locators, a CI workflow that runs the tests on every pull request, and a test suite for an AI agent that mocks its tool calls and checks how it handles prompt injection. Publish the code with a short README that explains what each test protects.
What is the difference between a QA engineer, an SDET and an AI agent quality role?
A QA engineer designs and runs tests, often manually. An SDET writes and maintains automated tests and the tooling around them. Teams that ship AI agents also need people who can test non-deterministic behavior: scoring answers against a rubric, checking tool calls, injecting failures and probing for prompt injection. These roles overlap, and titles vary between companies.
Do I need to know TypeScript to work in test automation?
It helps, since Playwright is most commonly written in TypeScript, and the basics are enough to start: variables, functions, async and await, and objects. Visual tools such as CertiQA let you build tests as plain steps first and export Playwright code to read alongside them.
CertiQA for QA engineersHow do I start testing AI agents?
Learn how an agent calls tools, then practice mocking those calls and checking that the right tool is called with the right arguments. Add an LLM-as-judge rubric for the wording of replies, and a small set of prompt-injection attacks drawn from the OWASP LLM Top 10.
AI agent testing in CertiQAIs this guide a certification?
No. It is a self-study guide. CertiQA does not issue professional certifications, and completing the roadmap does not confer a credential. A public portfolio and clear explanations of your work are what interviewers can actually evaluate.