AI in software testing in 2026: 5 changes that are actually shipping (not hype)

Every week brings a new “AI will replace testers” headline. We looked instead at what testers can actually use in 2026: features in Playwright’s release notes, ISTQB syllabi, job ads and EU law. Five changes stand out, and each one says something about what a tester’s job looks like now.

TL;DR

• Agents now ship inside the test framework: Playwright Test Agents (planner, generator, healer) since version 1.56.
• Playwright is built for coding agents: CLI debugging and trace analysis for agents (1.59), MCP server bundled (1.62).
• ISTQB split “testing AI” from “testing with AI”: CT-AI v2.0 (April 2026) and CT-GenAI v1.1.
• Employers ask for it: one QA job ad in five in France mentions AI.
• The AI Act high-risk deadline moved to 2 December 2027 (Annex III), after the AI Omnibus entered into force on 27 July 2026.

1. Are AI agents now part of the test framework itself?

Yes, in Playwright at least. Since version 1.56 (October 2025), Playwright ships Playwright Test Agents, three agent definitions that guide a language model through building a test:

• planner: explores the app and produces a Markdown test plan;
• generator: turns that plan into Playwright Test files;
• healer: runs the suite and automatically repairs failing tests.

One command, npx playwright init-agents, generates the agent definitions for your assistant (Claude Code, VS Code and others). The shift is important: AI assistance is no longer a third-party plugin bolted onto the framework, it is part of the framework’s own documentation and tooling.

2. What has Playwright added for coding agents in 2026?

Two releases made Playwright easier for an agent to drive and debug:

• 1.59 (April 2026): coding agents can run npx playwright test --debug=cli to attach and debug tests over playwright-cli, and npx playwright trace to explore a trace and understand failing or flaky tests from the command line.
• 1.62 (July 2026): Playwright now bundles the Playwright MCP server and playwright-cli, runnable with npx playwright mcp and npx playwright cli.

In practice, the loop “a test fails, the assistant reads the trace, proposes a fix” no longer needs custom glue code. Our Playwright MCP guide shows a first explore-then-generate scenario.

3. Why did ISTQB split its AI certifications?

Because they describe two different jobs. CT-AI v2.0, released on 17 April 2026, certifies the ability to test AI-based systems (data, models, generative AI products). CT-GenAI v1.1 certifies the ability to use generative AI to test (prompting, risks, LLM-powered tools). Since v2.0, “using AI for testing” belongs to CT-GenAI. Both require the CTFL. The English CT-AI v1.0 exam remains available until 21 April 2027. Details in our CT-AI vs CT-GenAI comparison.

4. Do employers really ask testers for AI skills?

Increasingly. In our count of 405 job ads with “QA” in the title published in France over 60 days (snapshot of 2 October 2026), 20.7% mention AI or generative AI. That word covers several realities: testing an AI-based product, using assistants to write tests, or simply the company’s context. For comparison, 24.2% mention Playwright and 40% mention Jira. The full figures are in our QA Skills Barometer.

5. What changed for the EU AI Act in 2026?

The calendar. The AI Omnibus, which amends the AI Act (Regulation 2024/1689), entered into force on 27 July 2026. According to the European Commission, rules for high-risk AI systems listed in Annex III now apply from 2 December 2027, and those for AI embedded in products covered by Annex I from 2 August 2028. For testers working on AI products, this is extra time to build risk management, data quality and robustness testing, not a reason to drop them.

What does it mean for a tester today?

The common thread: AI writes more tests, and humans review them. Before merging a test generated by an assistant, we check seven points:

1. a business assertion (not just “the page loaded”);
2. accessible locators (roles and names, not brittle CSS);
3. isolated test data;
4. no secrets in the prompt;
5. never run against production;
6. a limited exploration scope for the agent;
7. the test seen failing at least once, so you know it can catch a bug.

Test design (equivalence partitions, boundary values, risk-based prioritisation) matters more, not less: it is what you need to judge whether a generated test is worth keeping.

Over to you

Which of these have you actually used in a real project? Playwright's healer agent, MCP, an AI-generated test that made it to main? Tell me in the comments what worked, and what you had to throw away.

Sources: Playwright — Release notes (1.56, 1.59, 1.62) · Playwright — Test Agents · ISTQB — Certified Tester AI Testing (CT-AI) v2.0 · ISTQB — Testing with Generative AI (CT-GenAI) · European Commission — AI Omnibus enters into force · AutomationDataCamp — QA Skills Barometer, October 2026

Originally published on AutomationDataCamp. Written with AI assistance; every version number and date was checked against the sources listed above on 5 October 2026.

AutomationDataCamp is an online software testing academy (Playwright, API testing, CI, AI-assisted testing, ISTQB prep). More on our site.