AI Broke the Planning Fallacy
For decades, every project took longer than expected. AI just flipped the script. Now builders are overestimating timelines and underestimating what they can ship.
Key Points
- The planning fallacy said everything takes longer than you think. AI compressed execution so much that the opposite is now true.
- Builders are still scoping projects like it’s 2023. Two-week estimates for what takes an afternoon with the right tools.
- The new risk isn’t running out of time. It’s wasting time on work about work instead of shipping.
For decades, every builder I know operated under the same unspoken rule: whatever you think a project will take, add 40%. It was good advice. Psychologists called the underlying problem the planning fallacy. Humans are structurally incapable of estimating how long things take. We’re overconfident, we forget about coordination overhead, and we consistently underweight the likelihood of things going sideways.
I’ve shipped products across 23 businesses. I added the 40% religiously. And for most of my career, it was still barely enough.
Then something changed.
The Old Math Doesn’t Work Anymore
I rebuilt this entire website with Claude Code. Not a template. A custom Astro site with a bento grid layout, MDX blog system, dark mode, SEO, RSS, structured data. The kind of project I would have scoped at two to three weeks a couple of years ago.
It took a weekend.
That wasn’t a fluke. I’ve been building side projects, writing automation scripts, spinning up landing pages, and prototyping ideas at a pace that would have been physically impossible 18 months ago. The execution layer got compressed by something like 5 to 10x, and my planning instincts haven’t caught up.
The planning fallacy used to mean everything takes longer than you think. Now it means you’re still thinking in old timelines. You’re scoping two-week sprints for things that take an afternoon. You’re building project plans for work that doesn’t need a plan anymore, it needs a prompt and an hour of focus.
The Three Forces, Reversed
The classic planning fallacy has three components. Every one of them has flipped.
Overconfidence used to make you think you’d finish faster than reality allowed. Now the opposite is true. You assume a feature will take three days because that’s what your experience says. But your experience was built before AI could scaffold an entire component in seconds. Your reference class is outdated.
Coordination neglect used to be the hidden time sink. Syncing with designers, waiting on code review, back-and-forth on implementation details. AI collapses a lot of that. When I can go from idea to working prototype in a single sitting, the coordination layer shrinks dramatically. I’m not waiting on anyone. I’m iterating in real time.
Procrastination was the self-control failure that pushed hard tasks to tomorrow. But the hard parts got easier. The thing I used to procrastinate on, the tedious database migration, the CSS that never quite worked, the boilerplate nobody wants to write, is exactly the stuff AI handles best. The activation energy dropped. So I start things I would have put off.
The New Planning Fallacy
The old bias was optimism: thinking things would go faster than they actually would. The new bias is pessimism dressed up as experience. You’ve internalized years of “projects always take longer” and now you’re applying that heuristic to a world where execution speed fundamentally changed.
Here’s what I see this doing to builders around me. Founders who still plan quarterly roadmaps with the same velocity assumptions they used in 2023. Developers who estimate tickets the same way even though half the implementation can be generated. Side hustlers who haven’t started the project because they “don’t have time” for something that would take a weekend.
The constraint used to be execution. Now it’s decision-making. The bottleneck moved upstream. You can build almost anything in a weekend if you know what to build. The hard part isn’t the code. It’s choosing the right problem, scoping it tightly, and actually sitting down to start.
What Actually Takes Time Now
AI didn’t eliminate all the time sinks. It just reshuffled them. The stuff that still takes long is the stuff AI can’t do for you.
Talking to customers still takes time. Understanding what people actually need versus what you think they need requires real conversations, not generated personas. Strategy takes time. Deciding what not to build is harder than building it. Distribution takes time. Getting your work in front of the right people is still a grind, no matter how fast you shipped it.
And the meta-work, what I call “work about work,” is still the silent killer. Status update meetings. Writing reports about progress instead of making progress. A UCI study found that every interruption costs about 23 minutes of recovery time. One Slack thread, one “quick question,” and you’ve lost half an hour. AI made the real work faster, but it didn’t touch the fake work. If anything, the gap between the two is more obvious now.
Stop Adding 40%
I’m not saying throw out estimation entirely. But the old “add 40% to everything” rule needs an update. Here’s what I actually do now.
I estimate how long something would take me without AI. Then I divide by three or four for execution work, things like coding, writing first drafts, data analysis, and anything that AI can accelerate. For everything else, customer conversations, strategic decisions, design taste calls, I keep the original estimate or even add time.
The result is that I ship way more, way faster, but the thinking work takes just as long as it always did. That’s fine. The thinking is the valuable part.
The builders who are winning right now aren’t the ones with the best AI tools. They’re the ones who recalibrated their sense of what’s possible. They stopped planning like it’s 2023 and started asking a better question: if I can build anything in a weekend, what’s actually worth building?
That’s the question the planning fallacy never prepared us for.