The 60% Problem: How AI Coding Tools Are Training Us to Abandon Everything

Every fleeting idea is now just one terminal window away from existing. That's incredible. It's also quietly destroying our ability to finish anything.

There's a specific kind of shame that comes from opening a folder of projects. The names mock you. “invoice-parser-v2.” “recipe-thing.” “actually-going-to-finish-this-one.” Each one exists because an AI made the starting feel free — a few prompts, some scaffolding, the dopamine of watching a thing appear. And each one stops somewhere around 60%. Not abandoned, exactly. Just… paused.

AI coding tools have made starting trivially easy. They have done almost nothing to make finishing easier. And the gap between those two facts is where projects go to die.

Speed without direction is just faster spinning

The promise was always productivity. Build more, ship more, do more. And in a narrow sense, the tools deliver. You can now get from idea to scaffolded application in the time it used to take to set up a project folder. The latency between what if I built... and here's the thing you described has collapsed to almost nothing.

But latency was doing useful work. That friction — the setup, the boilerplate, the initial struggle — was a filter. It separated the ideas worth pursuing from the ones that seemed interesting for about forty minutes on a Wednesday evening. When you had to actually fight to start something, you thought harder before starting.

Now there's no filter. You have a thought, you open a window, you start building. The thought gets real fast enough that your brain registers it as real. The build is alive in your mind before it's alive in any useful sense in the world. And then — because there are five other windows open, each with their own small heartbeat — your attention drifts. The new thing pulls.

The build is alive in your mind before it's alive in any useful sense in the world. That gap — between thinking it and shipping it — is where everything goes wrong.

The compounding cost of context switching

Coming back to a paused project doesn't feel like resuming. It feels like starting over, but worse — because now you have to excavate what past-you was thinking, reconstruct decisions that seemed obvious at the time, and re-enter a mental state that exists nowhere in the codebase. Flow isn't just efficient. It's load-bearing. And each time you drop it, you pay a re-entry tax that's much higher than you expect.

Interrupted progress hamstrings future progress 

The AI can regenerate code. It cannot regenerate your understanding of why the code is shaped the way it is. It cannot reproduce the twelve decisions you made and then implicitly unmade. It cannot reconstruct the narrative you were building in your head. When you leave a project and return, the AI starts fresh. You start slower than fresh — you start in debt.

Moving faster doesn’t make finishing easier

The current wave of AI advice runs something like this: do everything in parallel. Spawn sub-agents. Run multiple workstreams. Let the AI's ability to multitask become your ability to multitask, by proxy. Move fast on everything simultaneously.

This is seductive, and it fundamentally misunderstands the problem. The bottleneck was never computing. It was never raw generation speed. The bottleneck is your ability to make good decisions, maintain coherent vision, and push a thing through the last 40% — which is where all the judgment calls live. The last 40% is almost entirely not scaffolding. It's handling the edge cases. It's the UX you didn't think about until you used it. It's the thing that breaks at scale. It's the part that requires you to care.

The last 40% of any project is almost entirely not code generation. It's judgment. It's the part that requires you to care. No amount of parallelism makes that faster.

Spreading that attention across eight parallel workstreams doesn't amplify it. It dilutes it until you have eight projects, each cared for at one-eighth capacity, all of them hovering in the high fifties percentages, indefinitely.

Swimming upstream: the case for one thing

There's a different way to use these tools. Not as a “yes, and” machine that scaffolds every idea the moment it arrives, but as a force multiplier for singular focus. Not many things faster; but one thing, much better.

What would it look like to apply the full leverage of AI coding assistance to a single project, from start to shipped? Not ten tabs open. One tab. One mission. The AI drafts, then you direct, and you push through the friction of the last 40% with all your attention intact and all your context loaded. You hold the project in your head the whole way through, not in pieces scattered across sessions that grow cold between visits.

You'd finish things. That sounds modest. It isn't. Finished things compound. They ship. They get used. They teach you something complete. A graveyard of 60%-done projects teaches you how to start, which you already knew.

The tools are extraordinary. The speed is real. But speed applied in too many directions at once is just friction wearing a different costume. The most powerful thing you can do with an AI that makes starting free is to still choose carefully what you start — and then, once you've chosen, to see it all the way to the end.

The author, David Kang (DK), is the Chief Technology Officer at Holokai, which helps organizations navigate their AI Journey at a time of unprecedented technical innovation, sky-high expectations, and uncertain outcomes.