You know that feeling. You open your laptop, fire up ChatGPT, and within thirty seconds you’ve got an outline for a newsletter, a draft of a business plan, and the skeleton of a side project you’re already excited about. You feel like a god. You feel productive.
Then it’s 11 PM and you’ve started four new things and finished zero.
This isn’t a discipline problem. It’s a design problem. And AI made it ten times worse.
The tool was never the bottleneck. You were. And now you’re 10x faster at proving it.
Here’s what’s actually happening: LLMs have created what Brent Fitzgerald calls a “productivity ouroboros” — a snake eating its own tail. The easier it becomes to generate output, the more output you generate. The more output you generate, the more overwhelmed you feel. The more overwhelmed you feel, the more you reach for the tool to “help you manage it all.” And the cycle repeats.
We were promised liberation. We got acceleration without destination.
Think about your own workflow right now. How many documents, drafts, prototypes, outlines, and project plans are sitting in various states of incompletion? Each one started with a burst of dopamine — “I can do this now!” — and each one now sits there, a small monument to your scattered attention.
Every half-finished project is a small tombstone for your focus.
The uncomfortable truth nobody’s talking about is this: AI didn’t create your inability to finish things. It exposed it. Before LLMs, the friction of starting was high enough that you could only begin a few projects at a time. That friction was doing you a favor. It forced prioritization. It made starting expensive enough that you’d only spend that currency on things you actually cared about.
Now starting is free. And when starting is free, everything gets started. Nothing gets finished.
One commenter on Fitzgerald’s post described a strategy that cuts right to the heart of it: instead of allowing the AI to spawn dozens of half-finished projects, they commit to one big ambitious project. Not because it’s more efficient. Because it removes the accusation. “There is always one big thing,” they wrote, “not a bunch of half-finished projects staring at me, each accusing me of abandoning them.”
That word — “accusing” — is doing a lot of work. Because that’s what those open tabs and unfinished drafts feel like, don’t they? They feel like evidence. Evidence that you can’t commit. Evidence that your excitement exceeds your follow-through. Evidence that you’re all ignition and no engine.
We optimized for starting because finishing requires something no model can generate: the willingness to be done.
Here’s the twist nobody saw coming: the real promise of AI was never speed. It was supposed to be freedom. Freedom from drudgery, freedom from the blank page, freedom from the slow grind of turning ideas into reality. But freedom from friction isn’t freedom — it’s just a faster way to lose control.
The people who will actually benefit from AI in the long run aren’t the ones using it to start more things. They’re the ones using it to finish what matters. They’ve figured out that the scarce resource isn’t ideas, isn’t output, isn’t even intelligence. The scarce resource is commitment. The willingness to look at a dozen exciting possibilities and say: not now. Not this. I’m finishing what I started.
That’s a human decision. No model makes it for you. No prompt engineering can substitute for it. And the sooner you stop using AI as a starting gun and start using it as a finishing tool, the sooner you’ll stop feeling that hollow ache of “doing more but achieving less.”
So here’s the challenge: pick one thing. One project. One draft. One idea from the graveyard of half-started ambitions. Don’t open a new chat. Don’t generate a new outline. Go finish the one that’s already staring at you.
The real productivity hack isn’t doing more faster. It’s doing less, completely.
Your AI can write a thousand beginnings. Only you can write the ending.
FAQ
Q: Isn't this just a discipline problem that existed before AI?
A: Yes, exactly — that's the point. The discipline gap was always there. AI just removed the friction that was accidentally compensating for it. Before, starting cost effort, so you self-filtered. Now starting is free, so the filter is gone and the gap is exposed at 10x scale.
Q: So should I stop using LLMs for brainstorming and starting new projects?
A: No. Use them for everything. But build a hard rule: no new project starts until the current one ships. The LLM should be your finishing tool, not your starting engine. Flip the ratio from 90% starting / 10% finishing to the reverse.
Q: What about people who genuinely use AI to finish things faster?
A: They exist, and they're the exception that proves the rule. The people winning with AI right now aren't the power users with 50 active chats — they're the ones who use it to close loops, not open them. The tool rewards finishers and punishes starters. That's the uncomfortable filter nobody's talking about.