Every Gantt chart has a bar for "Write code". For years it was the fat one. Now an agent can fill it in before your coffee cools, and the rest of the chart is still there, unchanged and looking at you.
That is the mood behind a post that hit Hacker News this week: AI promised less work. I've never worked more. It struck a nerve because plenty of people recognise the shape of it.
What happened
The author, engineering lead Orestis Ioannou, published the post using his own GitHub history as evidence. By Oct 27 he had logged 2,102 contributions, more than in any full year from 2021 to 2025. August 2026 alone came to 427, against a previous monthly high of 319. He describes bursts of output followed by burnout, a return to full-time-coder pace.
His diagnosis is simple. When coding is cheap, he takes on far more work in parallel. Each agent comes back with a question, a plan to approve or a diff to review, and the context-switching is what wears him out. There's a short summary on daily.dev if you want the quick version.
It lands alongside recent research. A Berkeley field study, published in HBR, found that people filled the freed-up capacity themselves. They widened their scope, ran more threads at once, and slipped "quick prompts" into lunch breaks and evenings. A Medium write-up covers the same study.
The delivery-side numbers point the same way:
- Faros AI telemetry reported 98% more merged PRs with AI, but review time rose 91%.
- A separate study of 100,000+ GitHub developers found that human bottlenecks after the code is written (review, merge, release) cancel out much of the gain from AI-written code.
LeadDev's take is that developers aren't more productive so much as busier.
Many commenters in the HN thread said it was always naive to expect efficiency gains to become free time. In their experience, the gain turns into more output demanded for the same pay.
Why engineers get it
The typing was never the whole job. It was just the part that fit neatly on a timeline.
Now the timeline has a new shape. Starting something costs almost nothing, so you start five things. Each one needs a decision, a review, a merge and a release, and those still go through a human who has only so much attention. The queue doesn't shrink. It moves downstream and gets wider.
The context-switching is the quiet cost. One agent wants a clarification. Another wants a plan approved. A third has a 600-line diff that looks plausible, which is the most expensive kind to review. None of those interruptions is large, but together they leave you tired without a single deep-work block to show for the day.
Then there's the expectations curve. If a feature takes an afternoon instead of a week, nobody says "great, take Friday off". They say "great, what's next?" The capacity gets spent before you've had a chance to notice it.
None of this makes the tools bad. It means the scarce resource was always judgment and attention, and we've just made the cheap part cheaper.
The design
The design is a Gantt chart in which "Write code" has shrunk to a sliver, while "Review", "Merge" and "Release" stretch off the edge of the page. An agent waves cheerfully from the left margin, delighted with its work. The printed line reads "Cheaper to start. Costlier to finish." It's for anyone whose planning doc has a very short first bar and a very long everything else.
Git Push and Pray