NOIDA -- November 13, 2026 -- Only 41% of software teams run any automated check on AI-written code before merge, while 44% report more escaped defects or production incidents as AI-written code has grown, according to the State of AI in Quality Engineering 2026 report from TestMu AI (formerly LambdaTest).
By our calculation, 59% of the teams surveyed merge AI-written code with no automated pre-merge check. At 40% of teams, AI now writes more than half of production code, the report says. Overall, 97% of teams use AI in testing or plan to, which leaves about 3% outside the trend.
Teams without separate CI tests report rising escaped defects 11 points more often
Manual review is the most common safeguard, used by 55% of teams. A QA hand-off after the code is written follows at 51%. Both checks happen after the code exists, and the shares overlap, so some teams use both.
Escaped defects rose at 49% of teams without separate CI tests, against 38% of teams that run them. That is a gap of 11 percentage points, or roughly 29% more often by our calculation.
Understanding lags behind merging. The report finds that 81% of teams have merged AI-generated code or tests they did not fully understand. Among developers the share is 93%.
TestMu AI sells an agentic quality engineering platform. The release gives no sample size, survey dates or respondent mix.
Teams that use one AI model to grade another merge code they do not understand nearly twice as often
Teams that use LLM-as-a-judge, where one model scores another, regularly merge AI output they do not understand at a rate of 29%. Other teams do so at 15%. The ratio works out to about 1.9 times.
The report also finds that AI-generated tests are "almost right, but not quite" a median 65% of the time.
Half of teams put most of their AI testing budget into generation, against 17% into verification
The two budget shares count teams, not dollars. By our calculation, about 2.9 times as many teams put most of their AI testing budget into producing AI output (50%) as into verifying it (17%).
AI agents add a second exposure. Some 77% of teams build agents they need to test. Of those, 37% have no systematic way to evaluate them, and only 28% test agents against users who try to break them.
CEO Asad Khan, Co-founder of TestMu AI, says output from AI should be validated "by a system independent of the model that produced it". He says surveyed teams that follow this practice see fewer defects reach production. The company expects the approach to become standard.
Only 13% of teams have an incident-response plan for AI errors, which leaves about 87% without one by our calculation.