What's in It for Developers? Sharing AI Productivity Gains
The question developers don't ask out loud
"I use AI more and more, and I ship more than ever. But the work isn't lighter – it's heavier. I hold more context in my head, review more code I didn't write, and make more decisions every hour. I end the day more tired, just in a different way. And next sprint, I get more tickets. So let me ask plainly: what's in it for me? Why should I work more and harder with AI, delivering more to the company, if I get nothing out of it?"
Few developers voice it, but it explains much of their resistance to AI. It is usually not about disliking new tools. It is a rational judgment about who harvests the gains.
Why developers keep their AI gains quiet
Wharton professor Ethan Mollick calls people who use AI quietly, without telling their employer, "secret cyborgs" (Detecting the Secret Cyborgs, 2023; Making AI Work: Leadership, Lab, and Crowd, 2025). They fear the time they save will be filled with new work, that higher productivity will justify job cuts, or that they will look lazy or like cheaters.
His conclusion: companies must change incentives openly. That means rewarding people who share what works, guaranteeing no one loses their job because of productivity gains, and letting employees keep part of the gain, for example as time for more interesting projects. Otherwise the gains stay hidden in individuals' hands instead of spreading across the organization.
Joe O'Connor, who researches the four-day week, calls this "performance punishment": the most efficient people simply get more work (Allwork.Space, 2026). He argues that giving people back part of the time AI saves reverses the mechanism. He also cites LinkedIn data: 51% of workers say keeping up with AI feels like a second job.
Economists have long had a name for this: the ratchet effect. Show that you can do more, and your quota goes up—so the rational strategy is not to show it.
What good companies do, seen from the developer's side
The practices below are grouped by what a developer actually feels, not by what the company wants to achieve.
"I earn more"
This is where examples are scarcest, which is a finding in itself: companies share time and tools with employees far more readily than money. The clearest case is Klarna. Its CEO says part of the savings from AI goes back into salaries, and average pay has risen 60% since 2022. The catch is that the company shrank at the same time, through natural attrition and a hiring freeze.
"I have more time"
Several companies protect time for learning instead of hoping people will find it. Duolingo sets aside two hours every Friday for AI experiments, with an "AI captain" in each function who prepares material for the session. Canva gave everyone a full week, because people said they simply had no time to tinker.
Zapier paused regular work for a week-long, company-wide AI hackathon, after which AI usage jumped from about 10% to over 50%. The boldest move comes from the Dutch software company AFAS, which since 2025 has run a four-day week at full pay, justified by productivity gains from technology and AI, and made it permanent after a pilot (MT/Sprout, Accountancy Vanmorgen).
"I do less tedious work"
The most convincing gains remove work nobody wants to do. At Stripe, agents handle dependency bumps, configuration changes and small refactors (InfoQ). Datadog's CEO describes AI as a way out of firefighting and false alarms, so that engineers have time to build (Pigment).
Intercom runs automated jobs that remove dead code and stale feature flags (Jam). It also lets AI approve the simplest pull requests, cutting review and waiting, and it put the reclaimed time into its defect backlog, which shrank by 54% (Fin). Carta took a similar route with a hackathon dedicated to clearing tech debt with AI (LeadDev).
"I do more interesting work, on my own terms"
AI doesn't just take away tedious tasks; it opens up projects there was never time for. When Anthropic studied its own engineers, 27% of AI-assisted work turned out to be things that would not have been done otherwise, such as revived ideas, internal tools and dashboards (Anthropic). Stripe built Harbor, an internal tool for rapid prototyping with an AI agent, and Datadog engineers used AI to build a platform that tests changes against production traffic (Datadog Blog).
But "interesting work" means different things to different people: prototypes and architecture for some, writing code by hand for others. Good companies leave room for that choice. At Duolingo, each team decides what to work on during its Friday experiments, without pressure (heise). IBM deliberately did not mandate its internal coding agent, because forcing usage risks shallow adoption; the stated goal was for engineers to enjoy the tools and feel more capable, not just more productive (LeadDev).
Craft matters too. Anthropic's study found that some engineers deliberately practice without AI to keep their skills sharp, and the company published those concerns openly. Shopify keeps a simple rule: to submit a pull request, you have to understand the code, even if AI wrote it (The Pragmatic Engineer).
"I'm not afraid for my job"
Security is the precondition for everything else, and it has to be said out loud. After its "AI-first" memo caused a backlash, Duolingo's CEO publicly stated that the company has never laid off full-time employees and doesn't plan to (Fortune). Shopify sent a different kind of signal by growing its internship program to over 1,000 people instead of cutting junior roles (CoderPad).
"I have a say in how I work"
Developers adopt the tools they help shape. Shopify removed spending limits on AI tools (First Round). At Intercom, developers co-create a shared library of skills and plugins, and 31% of R&D contributes to it (Fin). Datadog's CEO adds a cultural condition: admitting that something doesn't work should never hurt your career (Pigment).
"My market value grows"
This is the one benefit a company can't take back: your skills and track record go with you to your next job. The market prices them, too. Job postings that require AI skills pay an average 62% premium, up from 57% a year earlier (PwC Global AI Jobs Barometer 2026), and recruiters weigh a portfolio of real projects more heavily than certificates (Rework).
The strongest examples follow the same path, from an internal problem to a portfolio with the engineer's name on it. The path usually starts with a real, tedious problem. At Airbnb, it was migrating about 3,500 test files, a job estimated at 1.5 years of manual work. A hackathon team showed in 2023 that an LLM could convert hundreds of files in a few days, and the full migration was finished in six weeks (Airbnb Tech).
A measurable result turns a clever experiment into a story worth telling. At Datadog, engineers used AI to build a new parser that raised throughput by 10% and is projected to save about $2 million a year (Datadog Blog).
Then comes the step many companies skip: letting engineers tell that story under their own names. Charles Covey-Brandt wrote up the Airbnb migration on the company blog, and InfoQ later quoted him by name (InfoQ). Alistair Gray did the same at Stripe with the Minions coding agents, and Obie Fernandez at Shopify with the Roast framework. In Poland, Allegro, the country's largest e-commerce platform, runs an engineering blog in the same spirit (blog.allegro.tech).
The last steps are talks and open source. Block's Goose agent began as the project of one ML engineer, Bradley Axen, who was automating data migrations and repetitive tasks during a reorganization (Heroku podcast). Block open-sourced it, Axen presented it at the Data + AI Summit, and in 2025 it was donated to the Agentic AI Foundation at the Linux Foundation (Arcade). Shopify's Roast followed a similar arc: extracted from internal tools, open-sourced, and discussed by its author on The Ruby AI Podcast. At Intercom, engineer Brian Scanlan took the company's AI journey to the stage at DX Annual.
Each step leaves something with the engineer's name on it that they carry to their next job. The company only has to provide three things: time, a simple approval process for publishing, and a policy for open-sourcing internal tools.
What to avoid
Skeptical developers know the counterexamples too, and they show how easily AI becomes a ratchet. At Meta, an internal leaderboard ranked employees by token consumption. It rewarded burning tokens rather than results and was taken down once it became public (The Pragmatic Engineer).
Broken promises are just as corrosive. Atlassian's CEO said the company would have more engineers in five years; in March 2026 it cut 1,600 jobs, over 900 of them in R&D (The Next Web). Coinbase chose the stick and fired engineers who hadn't started using AI tools by a set deadline (TechCrunch).
Even strong adopters send mixed signals. Intercom has many of the good practices described above, but it measures merged pull requests per head and tiers engineers partly by how intensively they use AI (Fin). That is the kind of metric developers tend to read as a ratchet.
Where to start
Developers get out of AI exactly what their company chooses to give back openly: in money, time, security and growth. The first steps are cheap.
Start with protected time, such as two hours a week for AI learning and experiments. Agree openly on what happens to the time AI saves, so that part of it goes to tech debt, defects and people's own projects rather than straight back into the backlog. And measure outcomes, not usage: keep AI usage data out of individual reviews and skip the token leaderboards.
Then address the fears directly. Say clearly what higher productivity means for headcount, and keep your word. Give people permission and time to publish articles, give talks and open-source their tools under their own names. And if AI really does lower costs, let part of the savings show up in pay.
