Notes

Focus on becoming an engineer who judges quality. Don’t be a person who executes test cases. AI already drafts cases, writes scripts, and heals locators. That work is being commoditized. What still pays is owning risk, systems, and verification when the output is cheap and often wrong. Here is what I would do, in this order: Code: master one language for real. TypeScript or Python. Write helpers, fixtures, reviews of AI-generated tests. Automation shows up in most serious QA ads. The language is just a filter. API and contracts first, then UI. Most defects are not in buttons. They are in schema, auth, data, integrations. Learn Playwright/request or equivalent plus SQL. UI automation without API thinking comes late and costs more. CI as a product: GitHub Actions (or similar): the system that runs tests on every push and pull request. You design how tests run there. Parallelism: split the suite across many machines so a 40-minute run becomes 8 minutes. Slow feedback kills quality. Quarantine: isolate known-flaky tests so they don’t block the whole team, while you still track them. A flake that fails the build randomly trains people to ignore the pipeline. Cost: every extra minute and every extra browser or container costs money. Cut waste: cache deps, fail fast, and don’t rerun the entire UI suite for a readme file change. Signal-to-noise: the pipeline must tell the truth. High noise = flakes, timeouts, env issues, tests that fail for the wrong reason. When people stop trusting CI, they ship anyway. Own the pipeline. You decide what runs, when it runs, what is allowed to fail, and whether a red build actually means “do not ship.” Use AI as a force multiplier, then reject most of its output. Cursor/Claude or others to generate and debug. You stay accountable. Those days, prompting without review is how teams ship garbage tests. Learn to test AI products. Non-deterministic output, evals, hallucinations, bias, agent behavior, prompt regression. Explicit “AI skills” in QA ads are still a minority, but those roles pay a large premium. Observability. I mean: traces, logs, request IDs, metrics. Screenshots are not evidence. ...

Buying a services firm and running it with AI agents is just about pricing the model. You’ll still need the process, and the story changes there. You can buy the clients. But you can’t treat the firm as a stack of tasks waiting for AI agents. Walk the work for a week, and you see something else: the same job taking three different paths depending on who opened the email and whether the file arrived complete. Nobody wrote the exceptions down because the exceptions are the job. ...

Your AI adoption numbers come from the vendor who sold you the tool. Seats activated, weekly active users, prompts per head. Those numbers exist because a software company needed to know whether it had captured a customer. They arrive free, already formatted for your steering committee. The advice going around right now goes one level deeper: ignore the seat count, watch an ordinary Tuesday: does the AI assistant come up on its own in people’s work, without anyone nudging them to use it? ...