Imagine a car that can drive ten times faster than any other vehicle. That’d be amazing, right? But there’s a catch: when stopped at an intersection, the car has to wait ten times longer for the light to change. Would you drive that car?
This is a similar predicament as the one software development and testing teams face when they lean heavily on AI to help write code. AI tools can churn out code with incredible speed. But often, that doesn’t translate to faster overall software release velocity; we examine why below.
The verification costs of AI-centric coding
The issue is not just that AI tools can create code faster than most teams can test it. It’s also that, in some cases, AI-generated code actually makes testing slower.
There are two basic reasons why.
The first is that AI-generated code, on average, contains something like 1.7 times as many errors as human-generated code. That translates to more bugs, which means developers and QA engineers have to run and analyze more tests to fix them all.
The second is that AI-generated codebases tend to be larger because the code that AI writes is less concise. This leads to more code to sort through when remediating a flaw discovered during testing.
Given these challenges, it’s no surprise that QA engineers are struggling to keep pace with AI-generated code, and that organizations that make extensive use of AI to write code report a median release velocity boost of merely 8%.
The financial fallout of using AI in software delivery
From a business perspective, these outcomes may not be a big deal if AI coding tools were free or low in cost.
But they are a financial problem, given that some companies report spending tens of thousands of dollars a year per developer in token costs alone. To generate bottom-line ROI, expensive AI tools need to result in faster, more efficient development pipelines. Too often, that’s not happening.
Better software testing means more cost-effective AI coding
It doesn’t have to be this way. When teams become as efficient at verifying AI-generated code as they are at generating that code, they begin seeing a positive ROI.
This is where platforms like Leapwork Play come in. By speeding the processes of building, executing, analyzing, and maintaining software tests, Play removes the pain points that have historically slowed down software testing and verification. In turn, it positions developers and QA engineers to be able to test as quickly and effectively as they can code.
Play also helps turn fast, one-off AI-generated Playwright output into governed, reusable, auditable test assets that the broader team can own and maintain. Reusable components and self-healing automation can reduce maintenance effort by 50–70% compared with raw AI-generated Playwright output, helping prevent the cost of testing from growing alongside the volume of code.
Businesses need to find ways to supercharge the speed and efficiency of software testing to turn AI-assisted development into a net benefit from a financial perspective. When that happens, testing becomes as fast and cost-effective as AI coding.
