AI Writes Half the Code for 42% of Developers [2026]

A new workforce survey out of nearshore software giant BairesDev has put a hard number on something engineering managers have sensed all year: generative AI has stopped being an autocomplete novelty and started writing a meaningful share of production code. In its Q3 2026 Dev Barometer, published in mid September and still generating fresh coverage this month, BairesDev found that 42 percent of surveyed developers now say AI writes at least half of their code, up from just 12 percent one year earlier in the Q3 2025 edition of the same study.

The survey, fielded in August 2026 among 705 developers in more than 60 countries alongside a companion poll of 41 enterprise chief technology officers, is one of the more granular looks yet at how fast AI coding tools have moved from the margins of the software development lifecycle to its center. It lands alongside a wave of recent product news, including GitHub’s October 19 deprecation of four Copilot models, Cursor’s shift to a new subscription structure, and continued jockeying among GitHub Copilot and Cursor for enterprise seats. Taken together, the numbers suggest the AI coding story in late 2026 is no longer about whether developers will adopt these tools. It is about what happens to engineering organizations once adoption crosses the halfway mark.

·

Google · Preferred Sources

Don't miss new tech stories on Google

Add TrendinTech once in the Google app and our stories appear in your news suggestions.

Add Now

What the Q3 2026 Dev Barometer actually measured

BairesDev describes the Dev Barometer as a recurring quarterly research series tracking how artificial intelligence is reshaping software engineering work. The Q3 2026 edition surveyed 705 software developers across more than 60 countries in August 2026, then layered on a companion survey of 41 enterprise CTOs drawn from Fortune 500 and mid-market organizations. That combination, developers reporting from inside the work and executives reporting from the budget and governance side, is what gives the report its texture. It is not simply a tool-adoption tracker. It is an attempt to measure where the actual hours in a developer’s week are going now that AI assistance is routine rather than exceptional.

The headline statistic, 42 percent of developers saying AI now writes at least half their code compared with 12 percent in the Q3 2025 survey, represents a more than threefold increase in a single year. BairesDev’s Q3 2026 report framed it directly: developers are getting back nearly two working days a week, and the share who say AI writes at least half their code jumped from 12 percent to 42 percent in a year. A second figure fills out the picture: 72 percent of developers now say AI assists with at least a quarter of their code, more than double the roughly 33 percent who said the same a year earlier. Read together, those two numbers describe a workforce where partial AI assistance has become close to universal and majority AI authorship has gone from rare to common.

It is worth being precise about what the survey is and is not. BairesDev itself has noted that the developer sample drew heavily on applicants moving through its own engineering screening process rather than a random global census of working developers, and the CTO panel, while useful for an executive perspective, numbered just 41 people. That does not undercut the direction of the finding, which is corroborated by independent outlets that picked up the report within days of release, but it does mean the precise percentages should be read as directional signals from a reasonably large, internationally distributed sample rather than a scientifically weighted census of the global developer population.

The 42 percent number, broken down

The topline adoption figure is only one piece of a broader shift the Barometer documents. BairesDev also reported that nearly 80 percent of developers now spend less than half their week coding, which is the mirror image of a related finding that only 21 percent of developers spend more than half their week writing new code from scratch. Developers reported saving roughly 13 hours per week through AI-assisted coding, up from approximately 7 hours a year earlier, nearly doubling the time returned to each engineer.

But that time has not simply evaporated into free capacity. The same survey found that 67 percent of developers reported spending more time reviewing AI-generated code than they did 12 months earlier, and 52 percent reported spending more time fixing problems that AI-generated code introduced. On the executive side, 78 percent of the 41 surveyed CTOs said they had increased spending on code review, quality assurance, and validation over the past year specifically to keep pace with AI-generated output. The clearest way to read the data: the hours developers are no longer spending typing code are being redirected into verifying it, not eliminated.

Metric	Q3 2025	Q3 2026	Change
Developers saying AI writes at least half their code	12%	42%	+30 points
Developers saying AI assists with at least a quarter of their code	~33%	72%	More than double
Average coding hours saved per week	~7 hours	13 hours	+6 hours
Developers spending more time reviewing AI-generated code	n/a	67%	New this edition
Developers spending more time fixing AI-introduced problems	n/a	52%	New this edition
Surveyed CTOs who increased review/QA/validation spending	n/a	78% (of 41)	New this edition
Developers spending more than half their week writing new code from scratch	n/a	21%	New this edition

BairesDev’s research also touched on career and compensation effects tied to AI fluency. Twenty percent of developers said AI fluency helped them earn more over the past year, while 12 percent said AI was cited as a reason to limit a raise, reduce headcount, or restructure a role, and 61 percent reported no direct compensation impact either way. That split suggests AI skill is becoming a differentiator for some developers while simultaneously creating downside risk for others, rather than uniformly lifting or depressing pay across the board.

Why review time is becoming the real bottleneck

The Dev Barometer’s most consequential finding may not be the 42 percent headline but the review-burden data sitting underneath it. When two in three developers say they are spending more time checking AI output than they were a year ago, and more than three in four surveyed CTOs say they have had to pour additional budget into review and validation, the practical bottleneck in software delivery is shifting. It is moving away from how fast code can be produced and toward how fast it can be verified, tested, and made accountable to a human reviewer.

That finding tracks closely with independent research published earlier this year. A widely cited study from METR, the nonprofit AI evaluation group, found that experienced open source developers using AI coding agents on tasks within their own codebases were measurably slower than developers working without AI assistance, even though the developers themselves believed the tools were speeding them up. TrendInTech covered that METR study on AI coding agents in detail, and the gap between perceived and measured productivity it documented lines up with what the BairesDev CTO data implies: the time saved writing code is being partially offset, and in some cases more than offset, by the time required to review, correct, and take responsibility for what the AI produced.

There is also a trust dimension layered on top of the time dimension. Engineering leaders who have spoken publicly about agentic coding workflows have repeatedly emphasized that the decision to ship code, and the liability that comes with it, still rests with a single named engineer, regardless of how much of that code an AI model generated. As the Dev Barometer’s own executive summary put it: “In Q3 2025, 12% of developers said AI wrote half their code. Today it’s 42%, and the decision to ship it still sits with one engineer, for far more code than it used to.” That framing captures why review time is rising even as raw coding time falls: accountability has not been automated, only authorship has.

How this compares with other 2025 and 2026 developer surveys

The BairesDev findings do not exist in isolation. Stack Overflow’s developer survey work has tracked a similar adoption curve from a different angle, reporting that AI tool usage among developers rose from 44 percent in 2023 to 62 percent in 2024 and 79 percent in 2025, according to a retrospective published by Stack Overflow’s own blog in October 2026. A separate 2025 Stack Overflow survey found that 84 percent of respondents said they use or plan to use AI tools in their development process, and that 51 percent of professional developers said they use AI tools daily.

GitHub’s own research has reported even higher saturation at the enterprise level. A GitHub survey conducted by Wakefield Research found that more than 97 percent of enterprise software development respondents said they had used AI coding tools at work at some point. Those figures measure a broader question, whether a developer has touched an AI tool at all, rather than BairesDev’s narrower and arguably more revealing question of how much of a developer’s actual code output originates from AI. That distinction matters. Tool usage saturated years ago. What changed in the past twelve months, according to the Barometer, is the proportion of code those tools are now responsible for generating rather than merely suggesting.

Survey	Publisher	Period	Key finding
Q3 2026 Dev Barometer	BairesDev	Fielded August 2026	42% of developers say AI writes at least half their code, up from 12%
Developer Survey retrospective	Stack Overflow	2023 to 2025	AI tool usage rose from 44% (2023) to 62% (2024) to 79% (2025)
2025 Developer Survey	Stack Overflow	2025	84% use or plan to use AI tools; 51% use AI tools daily
AI wave survey	GitHub (Wakefield Research)	2024	97%+ of enterprise respondents had used AI coding tools at work

The tools behind the shift: Copilot, Cursor, and Claude Code jockey on price

The timing of the Barometer’s release lines up with a period of unusually fast repricing and repositioning among the three tools most commonly cited in enterprise coding-agent deployments. GitHub Copilot, as of October 19, 2026, listed a Free tier at $0, a Pro tier at $10 per month, a Pro+ tier at $39 per month, and a Max tier at $100 per month, alongside a Business tier priced at $19 per user per month. That same week, GitHub deprecated four older models, Gemini 3.7 Flash, GPT-5.5, GPT-5.4, and GPT-5 mini, across most Copilot surfaces including chat, inline edits, and agent mode, continuing a pattern of rapid model rotation inside the product.

Cursor, built by Anysphere, made a more structural change. On October 12, 2026, it began rolling out what it calls the X Subscription, replacing its legacy Pro plan at $20 per month with a new tier structure: a Free tier, Plus at $8 per month, Premium at $30 per month, Super at $100 per month, and Ultra at $200 per month. The older $20 Pro plan and the former Pro+ and Ultra tiers were closed to new signups as part of the transition, though existing subscribers were reported to retain their prior terms until they switch over. Anthropic’s Claude Code, meanwhile, has been reported at a Pro tier priced at $20 per month as of mid October pricing trackers, positioning it between Cursor’s Plus and Premium tiers on a pure subscription basis.

Tool	Entry tier	Mid tier	Top individual tier	As of
GitHub Copilot	Free ($0)	Pro ($10/mo), Pro+ ($39/mo)	Max ($100/mo)	Oct. 19, 2026
Cursor	Free, Plus ($8/mo)	Premium ($30/mo)	Ultra ($200/mo)	Oct. 12, 2026
Claude Code	Pro ($20/mo)	n/a	n/a	Mid Oct. 2026

That pricing churn is itself a signal. When three of the dominant coding-agent products reshuffle their tiers within the same month that a major adoption survey shows AI authorship tripling year over year, it suggests vendors are racing to capture and monetize a user base whose habits are changing faster than their billing models can keep up. Readers who have already standardized on one tool and are weighing a switch can find a practical walkthrough in TrendInTech’s guide to migrating from Cursor to Claude Code, which lays out the tradeoffs in more procedural detail than a pricing table alone can capture.

Historical context: from autocomplete to agent in three years

It is easy to forget how recently AI coding assistance was limited to single-line autocomplete suggestions. GitHub Copilot’s earliest versions, which reached general availability in 2022, offered inline suggestions that developers accepted or rejected line by line. Stack Overflow’s own tracking shows the inflection point: 44 percent tool usage in 2023 climbing to 79 percent by 2025, a trajectory that roughly doubled in two years. What the BairesDev Barometer captures is the next stage of that curve, the point at which usage saturation gives way to authorship saturation. Developers are no longer just using AI tools. A plurality of them are now outsourcing the literal writing of more than half their code to those tools.

That progression mirrors the broader shift toward agentic tooling that TrendInTech has tracked across the second half of 2026, including OpenCode 2.0’s expansion into API access for more than 75 AI models and the rapid iteration on Claude Code’s own release cadence, which shipped dozens of incremental fixes in a single October update cycle. Each of these products has been converging on the same bet: that the next competitive battleground is not code generation itself, which has become commoditized across vendors, but the surrounding infrastructure of review, testing, and governance that the BairesDev data shows is now consuming the hours AI assistance freed up.

Enterprise response: governance spending rises alongside adoption

The CTO half of the Barometer survey is arguably the more actionable half for engineering leaders trying to plan budgets. Seventy-eight percent of the 41 surveyed CTOs said they had increased spending specifically on code review, quality assurance, and validation processes over the prior year, a direct response to the volume of AI-generated code now flowing through their pipelines. That is consistent with broader enterprise behavior documented elsewhere this year, including large organizations publicly committing to high AI-coding-agent adoption targets among their engineering staff over the next twelve to eighteen months.

The practical upshot is that the cost structure of software engineering is shifting rather than shrinking. A team that previously budgeted primarily for developer headcount now needs to budget separately for review tooling, static analysis, security scanning, and the human review capacity to act on what those tools surface. Static analysis and security-scanning vendors have been quick to position themselves around exactly this gap, and GitHub’s own CodeQL updates this year have explicitly framed expanded detection coverage as a response to higher volumes of machine-generated code moving through continuous integration pipelines.

The leadership backdrop at BairesDev

The Q3 2026 Dev Barometer landed just weeks after a leadership change at BairesDev itself. On August 19, 2026, the company announced that Darren Shimkus had become Chief Executive Officer, with co-founder Nacho De Marco moving into the role of Chairman of the Board. BairesDev, founded in 2009, has described itself in that announcement as having operated without outside capital and profitable every year since founding, framing the Barometer research as part of a broader push into AI-driven commercial offerings under its new executive team. That context matters for readers weighing how much institutional weight to put on the survey: it is published by a company with a direct commercial stake in the nearshore engineering-talent market the AI shift is reshaping, not an independent academic or standards body.

Still, the directional finding, that AI-generated code has moved from a minority to a near-plurality share of output in a single year, holds up against outside corroboration. Independent technology outlets picked up the 42 percent figure within days of release, and it lines up with the broader usage curves Stack Overflow and GitHub have documented over a longer period. The open question is not whether AI-generated code is rising sharply, which multiple sources now agree on, but how engineering organizations restructure review and accountability processes to keep pace with it.

Five predictions for the next twelve months

Based on the trajectory the Barometer documents and the surrounding product and pricing activity, several outcomes look likely over the coming year.

• The share of developers reporting majority AI authorship will likely cross 50 percent within the next one to two survey cycles if the current year-over-year growth rate holds, though the rate of increase may slow as the easiest-to-automate code categories saturate first.
• Code review tooling and AI-output verification products will see disproportionate investment relative to code-generation tools, since the CTO data already shows 78 percent of surveyed enterprises redirecting budget toward review and validation rather than generation.
• Pricing structures across GitHub Copilot, Cursor, and Claude Code will likely continue to fragment into usage-based and credit-based models rather than flat subscriptions, extending the shift already visible in Cursor’s October move to the X Subscription and GitHub’s shift toward per-seat upfront billing.
• Compensation effects tied to AI fluency will become more polarized rather than uniformly positive, echoing the Barometer’s finding that 20 percent of developers saw pay gains tied to AI skill while 12 percent saw AI cited in connection with reduced raises or restructured roles.
• Expect more enterprises to formalize AI-code review policies and staffing specifically around verification, mirroring the pattern already visible in large financial and technology firms publicly setting internal adoption and governance targets for coding agents.

What this means for individual developers and teams

For individual developers, the Barometer’s data suggests the most defensible skill investment right now is not prompt engineering or tool fluency alone, but the judgment required to review, correct, and take responsibility for AI-generated output quickly and reliably. The 67 percent of developers already spending more time on review than they were a year ago are effectively living that shift today. Teams evaluating which coding assistant to standardize on should weigh not just raw generation quality but how well each platform supports review workflows, since products across the market, including GitHub Copilot’s code review features and Cursor’s Security Review tooling, are increasingly differentiating on that axis rather than on generation speed alone.

Engineering leaders, meanwhile, face a budgeting problem that looks less like a simple headcount tradeoff and more like a reallocation exercise. The 13 hours per week developers report saving on coding is real, but so is the additional review burden absorbing a meaningful share of it. Organizations that treat AI coding adoption purely as a cost-reduction lever, without correspondingly investing in review infrastructure, risk recreating the exact gap the BairesDev CTO data describes: rising AI-generated code volume outpacing the organization’s capacity to verify it.

Frequently asked questions

What is the BairesDev Dev Barometer?
It is a quarterly research survey published by BairesDev, a nearshore software development company, tracking how AI tools are changing software engineering work. The Q3 2026 edition surveyed 705 developers across more than 60 countries plus 41 enterprise CTOs.

What percentage of code is AI now writing according to the survey?
42 percent of surveyed developers said AI writes at least half of their code, up from 12 percent in the Q3 2025 edition of the same survey, conducted a year earlier.

Are developers actually working less as a result?
Not necessarily. While nearly 80 percent of developers now spend less than half their week coding and report saving about 13 hours a week, 67 percent say they are spending more time reviewing AI-generated code and 52 percent say they are spending more time fixing AI-introduced problems.

How does this compare with other developer surveys?
Stack Overflow’s research found AI tool usage climbed from 44 percent in 2023 to 79 percent in 2025, and a separate 2025 Stack Overflow survey found 84 percent of developers use or plan to use AI tools, with 51 percent using them daily. GitHub’s Wakefield Research survey found more than 97 percent of enterprise respondents had used AI coding tools at work. Those figures measure broader usage; BairesDev’s narrower question is about the share of code AI actually generates.

Is the survey sample representative of all developers globally?
BairesDev has noted the developer sample drew significantly on candidates in its own engineering screening pipeline rather than a random global census, and the CTO panel included only 41 respondents. The findings are directional rather than a scientifically weighted global census.

What is happening with AI coding tool pricing right now?
GitHub Copilot listed Free, Pro at $10 a month, Pro+ at $39 a month, and Max at $100 a month as of October 19, 2026. Cursor began moving to a new X Subscription structure on October 12, 2026, with tiers from $8 to $200 a month, closing its former $20 Pro plan to new signups. Claude Code has been reported with a Pro tier near $20 a month.

Did BairesDev have a leadership change recently?
Yes. On August 19, 2026, BairesDev announced that Darren Shimkus became Chief Executive Officer while co-founder Nacho De Marco became Chairman of the Board, a transition that occurred shortly before the Q3 2026 Dev Barometer’s release.

What is the biggest risk the survey highlights for engineering teams?
The widening gap between AI-generated code volume and an organization’s capacity to review it. With 78 percent of surveyed CTOs already increasing review and validation spending, teams that scale AI code generation without a matching investment in review infrastructure risk accumulating unverified or poorly understood code in production.

Related Coverage

• Cloudflare Tops 10M Developers, Bets Big on Platform [2026]
• CMA Opens iPhone NFC API to Developers by Nov. 17 [2026]
• Sphinx Git Hook Quizzes AI Code, Hits 252 HN Points [2026]
• VS Code 1.142 Drops Sandbox, Adds Agent Rules [2026]
• PostgreSQL 19 Delay: AI Review Drops 8-10 Features [2026]