You're Using the Most Powerful Tool in History to Almost-Finish
You have a million free consultants on call. They write code, draft the schema, scaffold the auth, explain the error. The single most capable building tool humans have ever made is open in another tab right now.
And the side project is still 80% done. Like it was last month.
This isn't a motivation problem. It's a usage problem — and a former Google executive named it better than I can.
"The Biggest Waste of Compute"
Mo Gawdat — the former Chief Business Officer of Google X, author of Solve for Happy — put it bluntly in a recent interview: the biggest waste of compute today is that we hand people the ultimate form of intelligence and they use it to write a text message to their girlfriend.
For developers the version is more specific. You have a tool that can take a half-built project to the finish line. You use it to generate one more plan. To refactor a file you'll abandon. To ask a question whose answer you already half-knew.
The capability is enormous. The use is shallow. And shallow use of an enormous tool feels productive — that's the trap. You closed the tab feeling like you worked. Nothing shipped.
"Don't Use AI as a Lazy Person"
Gawdat's actual prescription is the part worth tattooing on the monitor:
"You have to learn to use AI ... not as a lazy person. So don't have them do things for you. Have them make you smarter. So instead of trying to get the same task done with one prompt, try to get a much more interesting and demanding and intelligent task done with more work."
Read that as two different developers.
Developer A prompts "build me the dashboard," pastes the output, hits an error, and quietly stops. One prompt. Lazy use. The tool did a thing for them and they were no further along.
Developer B uses the same model to interrogate the actual blocker — why does the deploy fail, what's the real data model, what's the smallest thing that gets a working flow in front of one user. More demanding task. More work. They're closer to shipping.
Same tool. Same tokens. Completely different outcome. The difference isn't the AI. It's whether a human is driving it hard or coasting on it.
The Tool Is Just the Tool
Alex Bouzari, CEO of the AI-infrastructure company DDN, said the quiet part:
"AI tools will code for you. The value you bring is that you think of the problem, you formulate the problem. The tool does not know. It's like a car — the car doesn't tell you where to go. You tell the car."
A car is the most freeing tool imaginable and also completely inert without a destination and someone willing to drive the whole way. Most stalled side projects have a brilliant car idling in the driveway. Nobody decided where it's going, and nobody's committed to staying behind the wheel past Day 4.
More AI Was Never the Missing Half
Here's the uncomfortable empirical note. METR measured experienced developers on real tasks in 2025 and found they were 19% slower with AI tools (arxiv.org/abs/2507.09089). Not faster. Slower — because shallow, unstructured use adds context-switching and false confidence in "done."
So the fix can't be more AI. Gawdat's own frame is the right one: we're in the era of augmented intelligence — humans and machines together producing what neither produces alone. The machine half is solved. It's abundant, cheap, and getting better weekly.
The half that's missing is the human structure that makes you use the machine deeply and actually finish: a destination, a deadline, and someone who notices when you coast.
What This Looks Like When It's Built In
I run a 30-day sprint for developers with full-time jobs. The daily prompt doesn't say "keep going." It demands one concrete action that day — and the check-in where you report it gets read by a person, not parsed by a bot.
That single constraint kills lazy use. You can't coast through "build me the thing" when tonight you have to write, to an actual human, what you actually moved. And when you go quiet, you get one sentence back: what happened yesterday?
The AI is still doing 90% of the work. The 10% that was missing was never more compute. It was a human in the loop making you spend the compute well.
The Question Worth Asking
Open your repo. Count the AI-generated plans, scaffolds, and half-built features sitting in there. Now count how many became something a real person used.
That gap is the waste of compute. Not the tokens. The shipped project that's still imaginary because the most powerful tool in history got used to almost-finish.
The tool isn't the bottleneck. It hasn't been for a while. You are — and that's actually good news, because it's the part you can change this week.
If your project has been "almost done" for three months, more AI won't move it. A deadline and a human reading your daily check-in might. Cohort #2 is open — no upsell, just a link: mvpbuilder.io/pipeline
Building in public. Day 120.
Top comments (0)