AI coding tools have made it faster than ever to build a new app feature or fix a bug on the spot. What they haven't done is make that feature easier to keep running once it's live โ€” and a new industry survey suggests the gap between building and maintaining is becoming the real bottleneck.

Google's 2025 DORA report, an annual survey of nearly 5,000 technology professionals that has tracked software development practices since 2014, found that teams using AI coding tools are shipping code faster than before. But the same teams report that once that code reaches production, the work of monitoring it, catching errors, and fixing what breaks hasn't sped up at a comparable rate. In some cases, a feature that took minutes to generate created problems that took engineers hours to untangle.

This isn't a story about AI tools failing to deliver. Code generation tools from companies like GitHub, Google, and a growing list of startups genuinely do cut the time it takes to write a working feature. The issue is what happens after the code ships: someone still has to understand how it behaves under real-world conditions, trace unexpected errors back to their source, and judge whether a quick patch introduces a new problem elsewhere.

That work โ€” commonly called operations or maintenance in software circles โ€” has traditionally scaled with the number of engineers a company employs and their familiarity with the codebase. AI-generated code doesn't automatically come with that familiarity. A human developer who writes a feature usually has a mental model of how it works. A developer reviewing AI-generated code often doesn't have that same depth of understanding, which can make troubleshooting slower, not faster.

This pattern echoes something the software industry has seen before with other productivity leaps. Low-code and no-code platforms in the 2010s made it easier for non-engineers to build simple apps, but companies quickly discovered that these apps still needed ongoing technical oversight once they were used by real customers. Cloud computing made it trivially easy to spin up new servers, which led directly to a wave of businesses struggling to manage costs and security across sprawling, hard-to-monitor infrastructure. The lesson in each case: tools that lower the barrier to creation don't lower the barrier to upkeep.

For small businesses, this matters most if you've leaned on AI tools to build internal software, a customer-facing app, or an automated workflow without a dedicated engineering team. The appeal of AI coding assistants is that they let a small staff โ€” or even a non-technical founder โ€” build something that would have once required hiring a developer. The DORA findings suggest that appeal comes with a hidden cost: someone still needs to watch that software after launch, and that someone needs enough technical skill to diagnose problems that AI didn't anticipate.

Practically, this means budgeting for monitoring and support, not just the initial build. If you used an AI tool to create a customer intake form, a booking system, or an internal dashboard, it's worth asking who checks on it weekly, what happens when it breaks at 2 a.m., and whether that person actually understands the code well enough to fix it quickly. A freelance developer on retainer, even for a few hours a month, is often cheaper than the cost of downtime during a busy sales period.

Watch for how AI coding tool providers respond to this gap. Expect more vendors to market "monitoring" or "observability" add-ons alongside their code-generation products, and watch whether enterprise software budgets start shifting line items from development toward maintenance and support over the next few quarters.

The bottom line: AI has made writing software faster, but it hasn't eliminated the need for ongoing technical attention once that software is in use. Businesses relying on AI-built tools should plan for the cost of keeping them running, not just the cost of building them.