1. The work never ends, and somehow you don't want it to
Here's the usual pitch: if AI makes work faster, you get more free time.
But once you can hand off research, brainstorming, writing specs, building, fixing, and testing to AI one after another, the opposite can happen.
You don't spend the free time resting. You start inventing new work: "Oh, I could build this too." "I could fix that." "What if I hooked these two things together?"
And it doesn't feel like a grind. An idea hits on a walk. Another one hits while you're out with friends. You talk it through with an AI on your phone, turn it into a spec, and hand it to a coding agent. By the time you get home, something has moved forward.
It feels less like work eating your life and more like the real world turning into a map where you farm ideas.
The biggest difference from a regular game: a regular game has an endgame someone else built. Here, you can add the endgame content yourself.
It's gum that never loses its flavor.
2. Even the five-hour limit feels like a cooldown, not "game over"
On paid Claude plans, Anthropic says session-based usage limits reset every five hours, and there's a separate weekly usage limit that resets at a fixed time assigned to your account.[1]
The funny part: hitting a limit no longer has to mean "I'm done for today."
If you've broken the work into separate pieces, one session can stop while you move to another independent task. Or you pick it back up after the reset. The limit starts to feel like stamina or a cooldown timer in a game.
To be clear, this isn't about getting around anyone's terms or limits. It's about splitting allowed work into independent tasks and using the wait to sort out what depends on what, or to plan the next idea.
So your head goes from "the AI stopped" to "this worker's on break, so let's run another line."
3. Parkinson's Law, but for ideas instead of time
Parkinson's Law is usually summed up as "work expands to fill the time available." It started as a tongue-in-cheek essay in 1955, and later research tested versions of it at the individual level.[2]
With AI, the pattern bends a little.
Before, if you had eight hours, a task might take all eight. If AI cuts the actual work down to one hour, that doesn't mean you suddenly have seven free hours.
In those seven hours, you might:
- spot a new problem
- think up a new feature
- figure out how to connect another service
- raise your quality bar
- bring back ideas you'd dropped because they used to cost too much
So it's not just old tasks stretching to fill the time. Brand-new tasks sprout out of the time that got freed up.
The faster AI gets, the more your knack for coming up with ideas becomes the factory floor.
4. More productivity can mean more work: a Jevons-style rebound
This looks a lot like the Jevons paradox and the "rebound effect."
Those ideas come from energy debates: when better efficiency makes something cheaper to use, people use more of it, which can cancel out some or all of the savings you expected. How much that happens depends on the field and the situation, and a full "backfire" doesn't always happen.[3]
Calling this a scientific law of AI work would be a stretch. As a metaphor, though, it's spot on.
If building one thing takes three days, you throw out nine of every ten ideas.
If building one thing takes thirty minutes of instructions and review, those nine come back.
So instead of "productivity went up 10x, so we need a tenth of the work," you end up with "now we do all the stuff that never used to pay off."
The cheaper it is to build, the bigger the appetite for ideas.
AI can be a machine that removes work and, at the same time, a machine that creates work.
5. Why is this so fun? Freedom, getting better, and fast feedback
Self-Determination Theory says three things matter a lot for motivation and well-being: feeling in control of your choices (autonomy), feeling capable (competence), and feeling connected to others. Choosing what you do and feeling yourself get better are tied to the healthiest kind of motivation.[4][5]
That's a big part of why AI side projects turn into games so easily.
You're not filling in a spreadsheet someone dumped on you.
You decide what sounds fun.
The spec comes together in minutes.
Something working shows up within the hour.
If it breaks, you fix it.
And the fix keeps working from then on.
That loop keeps feeding you freedom, a sense of leveling up, and clear feedback, over and over.
In a game, your XP bar is a number.
In AI building, your XP bar is a real thing: something you couldn't do yesterday is running on its own today.
No wonder it's addictive.
6. Why it beats games: you get to build the endgame
A regular game, no matter how huge, is a world the developers built for you.
You beat the final boss. You finish your build. You grind the hardest content. Eventually you hit "there's nothing left to do."
With an AI-powered personal project, that's exactly when you add a new quest.
Built a site?
Now translate it.
Translated it?
Now get it out there.
Got it out there?
Now track how people react.
Tracked it?
Now have the next ideas generated automatically from those reactions.
Automated that too?
Now branch out into audio, video, new markets, new products.
Every time you hit the win condition, the player builds the next DLC.
It fits right in with the "build the system" fun of Minecraft or Factorio, except the systems you build affect real readers, customers, sales, what you learn, and hours of work saved.
The feedback isn't in-game gold. It comes from real life.
That's why the flavor doesn't fade.
7. You keep getting promoted: from project manager to product owner to boss
At first you act like a project manager:
"How do I get this task done?" "Where's it stuck?" "Who should do what?"
Then you turn into more of a product manager or product owner:
"What should we build?" "What do users actually get?" "Which improvement goes first?"
Once the AI also handles breaking down tasks, handing them out, building, and the first round of testing, your questions climb another level:
"Which product gets the resources?" "Which market do we grow?" "Which project do we shut down?" "Which systems should talk to each other?" "What part of the company do we build next?"
At that point you're not polishing one feature anymore. You're designing a small company.
| Stage | Main question for the human |
|---|---|
| Builder | How do we make it? |
| PM | How do we finish it? |
| PdM / PO | What should we make? |
| Executive | Where do the resources go? |
| Portfolio owner | What do we grow, and what do we kill? |
AI didn't make you important.
You just handed off the lower layers, so now one person can touch the higher layers that used to need staff and capital.
That's what makes it fun.
8. One person, and still an org chart
Even better: you don't have to manage every AI worker yourself.
You can set up layers like this:
- Human: goals, priorities, what's fun, final calls
- Chat AI: brainstorming, research, rough specs
- Manager AI: breaking down tasks, tracking what depends on what, handing out work, planning how it all fits together
- Coding AI: code, tests, fixes
- Monitoring: checking results, comparing changes, catching failures
It's not human → AI worker.
It's human → AI manager → AI workers.
Once that works, your job moves even further from micromanaging and toward "what do I have the company do next?"
Zero employees. A very impressive org chart.
The one-person company of the future is a pretty weird creature.
9. Don't just add more agents; keep them from crashing into each other
Add more AI workers and a new problem shows up.
Ten super-fast workers all fixing the same wire isn't fast.
It's how things break.
A 2026 study of AI-agent pull requests on GitHub saw merge conflicts between agents working at the same time. In the sample it looked at, pairs made of different agents working in parallel had a higher rate of text conflicts than pairs from the same agent.[6]
Another 2026 study tested an approach called CAID for long, complex software work: one central place hands out tasks, work runs in parallel without waiting, each worker gets its own separate workspace, and results are merged in an organized way. It scored better than a single agent working alone.[7]
What matters isn't "how many AIs can I spin up," it's:
- not letting two workers touch the same file at once
- drawing clear lines between who owns what
- figuring out what depends on what first
- giving each worker its own space
- merging and testing at the end
In other words, the next bottleneck isn't computing power. It's how you organize the team.
And once a manager AI starts handling that too, you really do get pushed into the CEO's chair.
10. What's scarce shifts from coding time to attention, taste, and direction
When AI can crank out huge amounts of building, what's in short supply changes.
It's not lines of code.
AI can even come up with plenty of ideas.
What ends up scarce is attention and judgment:
- What's actually fun?
- What's lame?
- How far should this go?
- Is now the time?
- What do we cut?
- What shouldn't we take on faith, and should test instead?
- Which failure is a one-off accident, and which one shows the system itself is broken?
It's not "AI can do 100 things, so do 100 things."
In a world where 100 things are possible, running the business means deciding which 3 actually matter.
The bottleneck moved: from building to direction.
11. The one trap of endless endgame: you also set your own game-over rules
No ending is the best part. It can also be the dangerous part.
Just because AI can run 24 hours a day doesn't mean you should.
You skip sleep because it's fun. You keep growing the scope because you can always improve something. You push tons of untested changes because you can run things in parallel. You add so many workers you lose track of who touched what.
At that point, automation meant to give you freedom turns into a high-performance sweatshop with one employee: you.
The fix can be designed like a game, too:
- put a checkpoint in front of risky, destructive changes
- don't let two workers edit the same area at once
- read back the results to confirm they actually landed
- test the things that matter
- build a way to recover when something fails
- count your own sleep and life as a dependency
- keep a "fun, but not now" list
The goal isn't to do less work.
It's to keep the fun going without breaking it, so you can play for a long time.
12. Bottom line: it's not that work is fun, it's that work where you can grow your own world is fun
When you catch yourself thinking "maybe work is actually fun," what's fun probably isn't work in general.
It's not grinding through tasks somebody else picked.
It's finding a problem yourself. Picking your next goal yourself. Handing the building to AI. Seeing something real come out of it. Looking at the results and coming up with the next thing.
That loop is the fun part.
Someone who used to think as a PM about "how do we improve this" hands off the breakdown and building to AI, and starts thinking about "what kind of company do I want this to be."
And once a manager AI starts handing out work to AI workers, you climb even higher.
The question left at the end is really simple:
What would be fun to build next?
In a regular game, the studio answers that.
In this one, the player does.
That's why it never ends.
And that's the best part.
References (7)
- Anthropic Help Center, “What is the Max plan?” / “How do usage and length limits work?” – Zurücksetzen der Sitzungslimits alle fünf Stunden und separate Wochenlimits. und https://support.claude.com/en/articles/11647753-how-do-usage-and-length-limits-work support.claude.com
- Brannon, L. A., Hershberger, P. J., & Brock, T. C. (1999). “Timeless demonstrations of Parkinson's first law.” Psychonomic Bulletin & Review, 6(1), 148–156 pubmed.ncbi.nlm.nih.gov
- Sorrell, S. (2009). “Jevons’ Paradox revisited: The evidence for backfire from improved energy efficiency.” Energy Policy, 37(4), 1456–1469; siehe auch spätere Übersichtsarbeiten zu Rebound-Effekten sciencedirect.com
- American Psychological Association, “Self-determination theory: A quarter century of human motivation research. apa.org
- Gagné, M., & Deci, E. L. (2005). “Self-determination theory and work motivation.” Journal of Organizational Behavior, 26(4), 331–362 selfdeterminationtheory.org
- Xu, G., Subramanian, A., & Karthik, N. (2026). “AI Agent Pull Requests on GitHub: Frequency, Structure, and Merge Conflict Rates.” arXiv:2607.04697 arxiv.org
- Geng, J., & Neubig, G. (2026). “Effective Strategies for Asynchronous Software Engineering Agents.” arXiv:2603.21489 arxiv.org
