How to Cross the AI Code Productivity Divide

When AI coding tools debuted, the productivity promise dazzled. However, in the four years they’ve been publicly available, both researchers and developers find the actual productivity benefits are mixed, at best.

While AI coding tools accelerate tasks such as prototyping and the development of boilerplate and repetitive code, the same tools actually slow down developers when they are debugging complex or runtime issues, writing code that is security-sensitive, and reviewing and debugging AI-generated code—and more.

Even when developers report productivity gains from using AI tools, those gains aren’t necessarily improving organizational bottom lines. While many startups find AI coding tools so beneficial that they base their business models on them, other organizations have yet to benefit from integrating AI coding tools into developer workflows.

“AI helps developers do more,” said Anastasiia Orovetska, head of TM Metric at Devart, a data management company. “More code. More tickets. Faster drafts. More iterations. But more output doesn’t always mean better delivery, higher quality, or more business value.”

The question of how, why, and when AI enhances productivity—and when it doesn’t—is complex, existing within an environment that is always changing as upgraded large language models (LLMs) and coding tools are released. Other factors in the productivity equation include the challenges of measuring developer productivity in general, how, why, and when developers and organizations use AI coding tools, and organizational and development context in which AI coding tool use takes place.

Whether or not AI delivers a productivity boost depends on the organizational development ecosystem within which a developer works, said Ankit Jain, co-founder/CEO at Aviator, a developer productivity startup, and host of The Hanger DX podcast. “AI is really amplifying your software development process,” he added, “which means, if you have a good software development process, you can do a lot more with AI, but if you don’t, many factors can bring your productivity down.”

Early reports on the productivity boost delivered by AI tools were positive. A landmark 2023 controlled experiment published by Microsoft revealed that developers using GitHub Copilot took 55% less time to complete an HTTP server implementation task than a control group that did not use AI.

In contrast, a randomized controlled trial conducted by METR (Model Evaluation and Threat Research) which evaluated the capabilities and risks of AI models found that when experienced open-source developers working on their own repositories used AI, tasks took 19% longer than when they did not use AI. The 2025 Stack Overflow Developer Survey found that 16.3% of developers currently using AI tools reported increased productivity, while 41.4% reported little impact on productivity.

On an organization-wide basis, the Faros AI/DORA AI Productivity Paradox Report revealed that teams that heavily employed AI coding tools completed 21% more tasks and merged 98% pull requests, but pull request review times leapt by 91%, creating bottlenecks further down in the development life cycle. The study also reported slower review times and no increase in delivery time compared with development without AI tools.

Sunita Verma, chief technology officer at Ironclad, said the verdict on AI coding tools and productivity is nuanced. “AI coding increases developer productivity as long as AI is integrated into the entire software development lifecycle including code generation, review, testing, building, deployment, documentation and debugging,” she said. “AI will not work if it is only integrated as a point solution for generating code alone.”

To determine whether AI coding tools increase productivity, developers, teams, and organizations need to assess variables beyond speed, said Pavel Tsarikov, CEO of Expert Software Development. “The better indicators are rework rate, review burden, code quality, and confidence in changes after release,” he said. “Without these checks, productivity gain can be very real at the start and very expensive later on.”

Measuring Developer Productivity

Measuring developer productivity—with or without the use of AI coding tools—presents challenges. While frameworks exist, the factors involved in measuring developer productivity are complex. Frameworks include DORA (DevOps Research and Assessment), DevEx (Developer Experience), and SPACE (Satisfaction, Performance, Activity, Communication/Collaboration, and Efficiency and flow). The development community hasn’t bought into one specific framework or tool.

The absence of an agreed-upon measure of developer productivity without AI coding tools makes it difficult to assess how AI coding tools impact productivity.

“A lot of the executives I speak with would have a really hard time going to their boardroom and showing that they were a good steward of their engineering budget,” said Laura Tacho, senior principal technologist, developer experience at Amazon Web Services. “That’s because they can’t look into their code editor and say how much a line of code is going to bring back in revenue.”

The State of AI Coding Tool Use

More than half of developers surveyed in the Stack Overflow 2025 Developer Survey report using AI coding tools on a daily basis. Daily users most commonly employ AI coding to search for answers, generate content or synthetic data, learn new concepts or technology, and document code.

More than two-thirds of developers surveyed voiced frustration over “AI solutions that are almost right, but not quite,” and the time-consuming nature of debugging AI-generated code. Despite high adoption of AI coding tools, developer trust in those tools has declined because AI tool coding outcomes aren’t predictable—instead, AI coding tools produce variable outcomes based on probabilities, the Stack Overflow survey said.

As AI coding tools impact the software develop lifecycle, Gartner predicts that 90% of developers will use AI coding assistants by 2028, up from 14% in 2024.

Strategies for Improving AI Coding Tools’ Productivity

There’s no single approach for accelerating developer and organizational productivity. Instead, developers and organizations can experiment with a variety of scenario dependent strategies to juice the AI coding tool adoption and success curve.

Tacho views intentional and skillful change management strategies as the key to harnessing significant AI coding tool productivity gains. “The companies that succeed in turning individual productivity into organizational productivity are the companies with really strong change management skills,” she said. These are organizations that already have strong internal processes and understand their technologies so they can be regulated and compliant, versus the ones that use the “spray and pray” method by just turning the technology on, she added.

Along with change management, the practice of requirements engineering to develop clear specifications on how developers use AI coding tools is important, according to Marsha Chechik, professor of computer science and Bell University Chair in Software Engineering at the University of Toronto. She explained requirements engineering defines what specific problem needs to be solved, how it should be solved, and where responsibility lies for the solution.

“We as humans are not very good at understanding and expressing our intent well,” she said. “Vibe coding works quite well for a single developer building a standalone product, but coordination between multiple developers or building more complex systems or developing them with high quality require a more serious specification process.”

AI coding tools make requirements engineering even more necessary, which means developers need to study and practice requirements engineering without shortcuts to improve productivity, she added.

Praveen Kumar Cherukuri, head of professional services at Wavemaker, an agentic AI app generation platform, said that clear specifications are correlated with successful use of AI coding tools. “The winners in the AI era won’t be the companies that generate the most code,” he said. “They will be the companies that build the best guardrails—the specifications, the measurement frameworks, and the architectural literacy—that allow AI to amplify excellence rather than accelerate mediocrity.”

AI coding bottlenecks frequently occur around inappropriate or non-existent model use policies and uncontrolled token spending, said Jonathan Teplitsky, founder of Layer3Labs, a company that customizes AI systems for small businesses. “My view is that AI coding does increase productivity, but the organizations that benefit the most will be the ones that manage it like infrastructure: measure output, control costs, define review standard, and route the right task to the right model,” he said.

Determining how AI is used within an organization rather than promoting indiscriminate use improves productivity, said Gajendra Babu Thokala, an IEEE senior member and senior engineering leader at Apple. “Development teams are capturing legitimate productivity improvements by treating AI coding as a thinking partner and accelerator for well-understood problems, not as an independent coder for entirely new problems,” he said.

Jain recommended processes that emphasize making small changes within code bases. Google’s practices regarding small CLs are designed to make coding changes more seamless and easier to manage. “Every change you make with AI coding tools should be really small,” he said. “It makes it easier for reviewers to pay attention when they review. Now context switching is happening a lot more frequently than it used to, which means everyone is now spending a couple of minutes on their code review before they are pulled back into another agentic workflow, where they’re doing some coding.”

Continuous delivery is another practice that can accelerate productivity around AI coding tools, Jain said. “Smaller changes that are deployed more frequently tend to have more stability versus large chunks of changes,” he said. “That makes it easier to identify bugs and issues before they become really hot.”

Amy Buttell is a Silver Spring, MD-based technology, legal, and business journalist, content creator, writer, and ghostwriter.