TestMu AI (formerly LambdaTest), the Agentic AI Quality Engineering platform, today released its State of AI in Quality Engineering 2026 report. The central finding will unsettle anyone responsible for production stability: 81% of software teams have merged AI-generated code or tests they did not fully understand, and 44% report more escaped defects or production incidents as AI-written code has grown.
Adoption is close to universal. According to the report, 97% of surveyed teams use AI in testing or plan to. AI now writes much of the code and many of the tests, yet most teams still verify that work with a person after the fact, or with another AI model. For most organisations, the open question is no longer whether to use AI, but how its output gets verified before it ships.
The verification gap
The numbers show how thin the safety net is. Only 41% of teams run any automated check on AI-written code before merge. The majority lean on manual review (55%) or a QA hand-off after the code is written (51%). Independent testing appears to matter: teams without separate CI tests see escaped defects rise more often (49%) than teams that run them (38%).
Developers, the people closest to the code, are the most exposed. The report finds that 93% of developers have merged AI output they did not fully understand.
When AI grades AI
Many teams have tried to close the gap by using one model to score another, an approach known as LLM-as-a-judge. The data suggests this may widen the problem rather than narrow it. Teams that use the technique regularly merge AI output they do not understand twice as often as other teams, at 29% against 15%. Quality of generated tests is part of the issue: AI-generated tests are “almost right, but not quite” a median 65% of the time.
Agents ship on demos
The pressure is growing as teams build AI agents of their own. The report finds that 77% of teams build AI agents they need to test, yet 37% of them have no systematic way to evaluate those agents. Only 28% test agents against users who try to break them, which means many agents reach production on the strength of human review and demos.
Budgets and preparedness
Spending patterns reinforce the imbalance. Half of teams (50%) put most of their AI testing budget toward producing AI output, while just 17% prioritize verifying it. Preparedness for failure is similarly sparse: only 13% of teams have an incident-response plan for AI errors.
TestMu AI argues that the remedy is independence. “Software teams now rely on AI to write their code, generate their tests, and increasingly determine whether both are ready to ship. Verification, however, has not kept pace with that shift. We believe any output AI produces should be validated by a system independent of the model that produced it, before it merges and every time it changes. The teams in our survey that already follow this discipline see fewer defects reach production, and we expect it to become the standard for engineering organizations shipping with AI”, said Asad Khan, Co-founder and CEO of TestMu AI.
For CIOs, the report’s implications are practical. Teams that treat AI-generated output as trusted by default are accumulating risk they cannot see, while the organizations pulling ahead are those that fund verification as seriously as generation.