Companies are not using artificial intelligence to shrink their engineering teams. They are using it to build more software, and they are paying a growing premium for the engineers who can manage it.
That is the central finding of Karat’s 2026-2027 AI Workforce Transformation Report, conducted by The Harris Poll. Ninety percent of U.S. technology leaders say they are using AI to increase output while maintaining or expanding engineering headcount. Only 10% say they are holding output steady with fewer engineers.
At the same time, 75% of global technology leaders now say a strong engineer is worth at least three times their compensation.
More software, but not faster software
The results fit the Jevons paradox, the economic idea that making a resource cheaper to use tends to increase total consumption of it. As code gets easier to produce, companies are producing more of it.
Among U.S. leaders, 82% report higher product and feature volume because of AI. But the speed gains are less clear:
67% say the time needed to bring new features to market has increased
44% report an increase in coding errors
The work itself is changing. Engineers spend more time reviewing AI-generated output, refining requirements, designing systems, testing, and managing the risk of ever-larger codebases.
AI is widening the talent gap, not closing it
Many expected AI to level the field between junior and senior developers. The survey points the other way. Organisations are putting more money into senior individual contributors and upskilling, while entry-level roles face more pressure.
“The most impactful engineers are increasingly applying skills we once associated primarily with engineering leaders,” said David Lau, Vice President of Workload Scaling at OpenAI. He said they are “constantly asking how best to combine the strengths of AI agents and human collaborators,” and they define success criteria that let agents show they’ve delivered the intended result.
Quality matters more as leverage grows. Fifty-seven percent of U.S. leaders are very satisfied with their full-time engineers, versus 38% for contractors. Still, 51% expect contractors to use AI to deliver more output at about the same cost.
A new line item: the cost of AI per engineer
Models, agents, tokens, coding tools and compute now sit beside salary and benefits in engineering budgets.
75% of U.S. leaders factor an engineer’s total cost, including AI usage, into workforce decisions
74% actively optimize AI spending at the individual engineer level
Leaders think they’re ahead, but few measure revenue
Nearly three-quarters of U.S. leaders believe their organizations are ahead of their industry at using AI, and 13% say they are significantly ahead. Just 5% say they are behind.
Yet what they report upward is mostly about inputs. Seventy-three percent report technology spend versus budget to CEOs and CFOs, but only 42% report product or feature revenue.
Preparedness appears to drive returns. Where employees are considered very prepared to use AI, 89% of leaders say at least half their engineers are seeing meaningful ROI. Where they are not, the figure falls to 65%, a 24-point gap.
What the wider data says
The Karat findings echo other recent research, though the picture is mixed. A study reported by PYMNTS found that firms with the highest AI spending per employee grew headcount 10.2% over the following two years, with entry-level headcount up 12%. The same report noted that almost all of those gains were in technology-sector firms and that nearly 102,000 job cuts have been attributed to AI this year, so short-term data remains mixed.
The new management challenge
For CIOs and CTOs, the question is no longer how much their engineers use AI. It is which combinations of talent and technology create real advantage, what that capability costs, and where it produces measurable business value.
Methodology: Karat’s 2026-2027 AI Workforce Transformation Survey, conducted by The Harris Poll, surveyed 157 technology leaders in the United States and China who are involved in AI and technology workforce planning. U.S.-specific figures reflect the U.S. portion of that sample.