IT Brief Asia - Technology news for CIOs & IT decision-makers
Asia
Most teams merge AI code they do not fully understand

Most teams merge AI code they do not fully understand

Fri, 20th Nov 2026 (Today)
Joseph Gabriel Lagonsin
JOSEPH GABRIEL LAGONSIN News Editor

TestMu AI has published its State of AI in Quality Engineering 2026 report. It found that 81% of software teams have merged AI-written code or tests they did not fully understand.

The report also found that, at 40% of teams, AI now writes more than half of production code, pointing to a broader shift in how software is built and tested.

Its findings suggest many organisations are expanding their use of AI in software development faster than they are putting checks around the output. Some 44% of teams said more bugs were reaching production as AI-written code became more common, while only 41% said they run any automated check on AI-written code before it is merged.

Manual oversight remains more common. The study found that 55% of teams rely on manual review and 51% depend on a QA hand-off after code is written, indicating that verification often still happens only after AI has generated software or tests.

Developers appeared to face the greatest exposure. Among those surveyed, 93% said they had merged AI output they did not fully understand, well above the overall figure across software teams.

Verification gap

The study argues that the biggest risk now lies in checking AI-generated software rather than producing it. Teams without separate continuous integration tests were more likely to report escaped defects: 49% said bugs reached production more often, compared with 38% of teams that run those tests.

Another area of concern was the use of one AI model to assess another. Teams using what the report described as LLM-as-a-judge methods regularly merge AI output they do not understand at nearly twice the rate of other teams, 29% versus 15%.

The report also found that AI-generated tests were "almost right, but not quite" a median 65% of the time. That points to a practical challenge for engineering teams, which may receive plausible-looking output from AI tools but still need to decide whether it is accurate enough for production use.

Budget priorities

Survey spending patterns showed more money going towards generating AI output than verifying it. Half of teams said they spend most of their AI testing budget on producing output, while only 17% said verification takes the largest share.

That split may help explain why many teams have yet to adopt stronger controls. Only 13% said they have an incident-response plan for AI errors, suggesting most organisations still lack formal procedures for dealing with mistakes introduced by AI-generated code or tests.

The report found that 97% of teams surveyed now use AI in testing or plan to do so. That near-universal level of expected adoption means the questions it raises extend beyond specialist AI companies to mainstream software teams across sectors.

Testing AI agents

The research also looked at AI agents, an area drawing growing attention from development teams. It found that 77% of teams build AI agents they need to test, but 37% have no systematic way to evaluate them.

Only 28% said they test agents against users who try to break them. The figures indicate that while AI agents are becoming part of software products and workflows, structured methods for stress-testing them remain limited.

The survey points to software teams leaning heavily on AI at several stages of development, from writing code to generating tests and deciding whether outputs are ready to ship. The findings suggest that dependence can create blind spots, with teams accepting output that appears useful without fully understanding how it works or whether it will hold up in production.

It also found that many teams still verify AI output either with a person after the fact or with another AI model. That raises questions about whether the checking process is sufficiently independent from the generation process.

"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 organisations shipping with AI," said Asad Khan, Co-founder and CEO, TestMu AI.