Tricentis released their 2026 Quality Transformation Report last month. The headline number: 60% of organizations still deploy untested code. That's nearly identical to the 63% from 2025. A year of AI investment, a year of agentic everything, and the number didn't move.
The same report found that 32% of those organizations are doing it knowingly — leadership is choosing speed over quality. Another 30% say the volume of AI-generated code is simply too much for their teams to test. The code is shipping because nobody has time to verify it, and the people in charge have decided that's acceptable.
This is the year the testing industry went all-in on "agentic." Gartner predicts 40% of enterprise applications will feature task-specific AI agents by the end of 2026, up from less than 5% in 2025. Vendors are rebranding everything — Tricentis launched an "AI Workspace," Katalon added natural language test generation, a new CLI tool lets you describe tests in plain English and an agent drives the browser for you.
The pitch is compelling: an AI agent that plans what to test, writes the tests, runs them, interprets the results, and heals anything that breaks. No human in the loop. Coverage goes up 5–10x at the same headcount. Testing solved.
Except Gartner also predicts that over 40% of agentic AI projects will be canceled by the end of 2027 — due to escalating costs, unclear business value, or inadequate risk controls. And trust in AI agents making release decisions actually fell year over year, from 48% to 34%. The people closest to the work are getting less confident in the approach, not more.
There's a number in the Tricentis report that explains a lot: 93% of C-level leaders feel confident in their testing strategy. But 30% of QA and DevOps leaders — the people who actually run the tests — do not.
That gap is the problem. The executives see the agentic roadmap and assume coverage is expanding. The teams on the ground see AI-generated tests that look right but test the wrong thing, self-healing selectors that mask real regressions, and an ever-growing backlog of flows that nobody has tested at all.
Only 15% of organizations have scaled generative AI enterprise-wide in QA, despite 43% experimenting. The experiments look good in demos. They don't survive contact with a real app that has email verification, third-party payment redirects, and session state that spans six pages.
The flows that stay untested aren't untested because the test code is hard to write. They're untested because the flow itself is hard to automate:
An AI agent that generates test code still has to solve these problems. It still needs to handle the cookies across redirects, the iframe restrictions, the session isolation, the real email inbox. Generating a Playwright script in natural language doesn't make any of that easier — it just moves the failure from "nobody wrote the test" to "the agent wrote a test that doesn't work."
One in five organizations loses more than a million dollars a year to poor software quality. The causes are security and compliance failures (30%) and technical debt and rework (28%) — both of which live in the flows that don't get tested.
The teams making progress on that number aren't doing it by deploying more sophisticated agents. They're making it trivial to create a test: record the flow once, replay it against the real app, run it on a schedule. No generation step, no healing step, no agent deciding what to test. A human clicks through the flow they care about, and the test exists.
Agentic testing will eventually get good enough to matter. The Gartner cancellation prediction suggests that's further out than the current hype cycle implies. In the meantime, the 60% number isn't waiting for better agents. It's waiting for someone to test the flow at all.