AI code generation is racing ahead of the controls meant to govern it, according to a new GitLab study that surveyed more than 1,500 developers. While 91% of organizations now run two or more AI coding tools, nearly four-fifths (79%) say software delivery hasn’t sped up at the same rate as raw developer productivity — a gap GitLab calls the “AI paradox.”
The AI paradox: faster code, slower delivery
The headline numbers look encouraging at first. Around three in four developers believe they’re writing code faster (78%) and producing higher-quality code (73%). Yet the time AI supposedly saves is being consumed elsewhere — by reviewing, validating, and governing what the AI produces. In GitLab’s telling, the constraint hasn’t disappeared; it has simply moved downstream.
The data backs that up: 85% agree the biggest bottleneck is now code review and validation rather than code creation itself. For teams that follow enterprise AI adoption, that’s a crucial reframing. Generating code was never the hard part; trusting and maintaining it is.
Visibility is the real problem
As AI becomes embedded in daily workflows, developers are struggling to see what it’s doing. Two in five (43%) now find it hard to distinguish AI-generated code from human-written code, which complicates long-term security and quality. Three in four (73%) worry specifically about the maintainability of AI code over time.
That lack of visibility has real operational consequences. A third of developers (34%) say they can’t determine whether AI-generated code played a role in an incident — a serious gap when tracing the root cause of an outage or breach. Meanwhile, 92% report some form of governance challenge around “vibe coding,” and four in five admit they adopted AI tools faster than they put policies in place.
Trust becomes the differentiator
“The events of the past few months, including supply chain attacks, reliability issues, and regulators tightening expectations around AI traceability and provenance are making clear that speed without control is a liability, not an advantage,” said GitLab Chief Product and Marketing Officer Manav Khurana. His prescription: bake accountability into the platform rather than bolting it on afterward.
Developers appear to agree. Some 91% plan to invest in governance over the next year, and 98% say they’re allocating budget specifically for it. The message is that generating code is no longer the priority — governing it is. Teams building software delivery pipelines are increasingly treating traceability as a competitive feature, not overhead.
A pattern of caution
GitLab’s findings echo a broader unease. Stack Overflow’s 2025 developer survey found many developers turning away from AI over privacy, pricing, and quality concerns, with nearly 46% distrusting the technology to some degree versus just 33% who trust it. Coding has been a flagship use case for generative AI since the technology went mainstream in late 2022, yet the GitLab data shows policies and guardrails still lag well behind adoption. Together, the surveys suggest developers have accepted that AI has clear, valuable use cases — while insisting that traceability, accountability, and trust will decide who ships reliable software fastest.
The practical implication for engineering leaders is a shift in where to invest. If review and validation are now the true bottleneck, then tooling that improves visibility into AI-generated code — tracking its provenance, flagging it in reviews, and tying it to incidents — may deliver more value than simply generating more code faster. That reframing, from output speed to trustworthy delivery, is the core message of GitLab’s report.
Frequently asked questions
What is the 'AI paradox' in software development?
It’s GitLab’s term for the gap between faster individual coding and slower overall delivery, because time saved writing code is spent reviewing, validating, and governing AI output instead.
What is now the biggest bottleneck in AI-assisted development?
Code review and validation. In GitLab’s survey, 85% of developers said review — not code creation — is now the main constraint.
Are developers investing in AI governance?
Yes. GitLab found 91% plan to invest in governance over the next year, and 98% are allocating budget specifically for it.
