AI coding succeeds creating new CIO challenges

The numbers should trouble any CIO who approved an AI coding rollout expecting a productivity windfall. Some 84% of developers now use or plan to use AI tools, according to Stack Overflow’s most recent published survey of more than 49,000 developers. Yet productivity gains have plateaued at about 10%, according to a 2026 study from developer intelligence platform DX Research, even as 93% of the 121,000 developers surveyed reach for AI.

The productivity numbers tell only part of the story. They are a symptom of a deeper shift: AI isn’t just changing how fast software gets built. It’s changing what developers do, how teams are structured, and — most consequentially — how the next generation of engineers learns the craft.

Developers become designers and reviewers

Kai Chuang, CIO of Circles, said he has seen this firsthand. His developers’ work has shifted from hands-on coding toward design and systems architecture. Developers at the workplace hospitality services provider spend less time “on literal programming,” he said, and more time specifying what to build and testing whether it works. The pace of change caught him off guard. Once developers began trusting the output, “the changeover to nearly full AI code generation happened rapidly on its own,” with no top-down mandate, he said.

The scarce skill is no longer writing code. Erik Brown, a senior partner at management and technology consulting firm West Monroe, explained, “It’s knowing what should be built, how it should be architected, whether it’s secure, and whether it actually advances the business outcome.” Rather than simply writing more code, “the companies that get this right will redesign the software development lifecycle around AI. The ones that simply hand developers tools will get more activity, not necessarily better results,” Brown said.

At UiPath, an enterprise automation software company, “well over the majority of production-deployed code is authored by coding agents already,” said chief technology and product officer Raghu Malpani. “Developers are transforming from code writers to reviewers and system designers. They’re defining intent, validating outputs, and shipping more code, faster,” instead of writing every line of code, Malpani said. He calls it “a shift in one of the core parts of the developer identity.”

When coding is no longer the slow step, the bottleneck moves upstream to design, which AI is reshaping as well. That puts new demands on business analysts and product managers to have concepts “shovel-ready,” Chuang said. Using AI to explore use cases and mock up interfaces before involving developers lets them “deliver a much better, more refined design,” he said.

For CIOs, the practical consequence is that their teams need a different mix of skills — less raw coding ability, more architectural thinking and business judgment. The developers who thrive will be the ones who can ask the right questions, not just type the right syntax. That shift is already reshaping hiring and training, and it’s happening faster than many organizations expected.

Beyond productivity metrics

If productivity looks flat, CIOs should first assess whether they’re measuring the wrong things. Consider what Cornerstone Research, an economic and financial consulting firm that supports high-stakes litigation, found in its own data. Across more than a million billable time records, “the answer so far is essentially no change,” said chief technology and innovation officer Phil Leslie. But that conclusion, while accurate, is also misleading.

“AI use has not measurably reduced the analyst share of hours,” Leslie said. “But what it has done is shift the mix: analysts report less time on coding and debugging, and more on interpretation, methodology and thinking. The job feels different, even though the hours have not moved.”

Some organizations are reporting substantially larger productivity gains. At Bank of America, which invests nearly $14 billion annually in technology, the AI-powered coding assistance used by more than 18,000 developers is generating efficiency gains of more than 20%, according to the company. But raw speed isn’t the point, said Hari Gopalkrishnan, the bank’s chief technology and information officer. “The need for talented people who can solve complex problems, exercise judgment and build relationships will remain critical,” Gopalkrishnan said.

Most standard AI coding metrics still count effort rather than results: seats deployed, tokens consumed, lines of code generated, self-reported hours saved. “Those are activity metrics,” Brown said. “The better question is whether the business and engineering outcomes moved.” Brown recommended a dashboard that tracks cycle time from idea to production, deployment frequency, change failure rate, escaped defects, security vulnerabilities and the share of AI-generated code requiring material human correction. “The goal isn’t more code,” he said. “It’s faster, safer, higher-quality delivery tied to business outcomes.”

The data shows why quality matters. Developer tools maker GitClear’s analysis of 211 million lines of code found code churn nearly doubled between 2020 and 2024, while refactoring dropped from 25% to less than 10%. A 2026 benchmark from software delivery firm Opsera found AI-generated pull requests take 4.6 times longer to review and contain 15% to 18% more security vulnerabilities than human-written code. The time saved writing code often reappears later in review queues and security fixes.

The junior developer time bomb

The most serious risk won’t appear in this year’s metrics, however. It’s lurking two or three years out. The routine work AI now absorbs — bug fixes, documentation and test coverage — was exactly how junior developers sharpened their skills. Strip that away “without a new apprenticeship model to replace it, and companies will create a talent gap two or three years out,” Brown warned.

Cutting entry-level roles on the theory that AI will replace juniors is “close to a one-way door,” cautioned Leslie. “The apprenticeship model is how almost every profession grows the judgment its senior people eventually rely on.”

The fix isn’t to stop hiring junior developers but to redefine the role. The best early-career developers “won’t just know how to write code — they’ll know how to ask the right questions, understand the business intent behind the software, and evaluate whether AI-generated output actually solves the problem,” Brown said.

The hiring calculus changes, too. Chuang said he now favors developers who are “more interdisciplinary and interested in solving underlying business problems.” Malpani added that as coding gets cheaper, “the judgment of what and how to code becomes a valuable asset,” with the premium going to developers who understand system design and can keep automations “secure, compliant and maintainable over time.”

Governance moves to the center

As AI-generated code proliferates, oversight shifts from the margins to the core of the job. “The emphasis is moving from reviewing every line of handwritten code to governing the entire software lifecycle: testing, deployment, permissions, auditability and runtime behavior,” Malpani said.

“Enterprises will need platforms that provide consistent oversight, control and traceability, regardless of which coding agent produced the code,” he said, stressing that agents “need guardrails and experienced reviewers.” It is a lesson that runs against intuition. Coding agents “haven’t eliminated the need for low-code or enterprise development platforms,” Malpani said. “They’ve made [those platforms] more valuable. Faster code generation increases the need for reviews, judgment, governance and collaboration.”

Speed is part of the payoff. But the full value of AI coding tools doesn’t accrue until the work around software development changes — how teams are built, how work is measured and how the next generation learns to judge what the machines produce.