The Bottleneck Was Never the Building

I have a folder on my laptop named "abandoned." Forty-one first chapters live in it, each one a novel I was certain about for exactly as long as it took to write the opening scene. I used to think the folder was evidence of a discipline problem. I’ve come to think it’s the most honest record I have of how ideas actually work, and lately I’ve started seeing the same folder everywhere I look in product development, just filed under a friendlier name: the backlog.

The chapters in that folder are not bad ideas. That’s the part that took me years to accept. Most of them are perfectly good premises, sentences I could still defend in a pitch meeting today. What they share is not a flaw in the concept. It’s that I stopped writing them at almost exactly the same point, somewhere between page twelve and page thirty, right after the opening momentum ran out and right before the story told me what it was actually about. That stretch has no name in writing craft that makes it sound appealing. Everyone just calls it the middle, and everyone who has finished anything long knows it’s where the real work happens and where almost everything dies.

AI changed the economics of the opening scene, in both crafts, and changed nothing about the middle. A model can now produce a competent first chapter, or a working product prototype, in the time it used to take to clear your throat. I want to be honest about how good that is. The opening used to cost real time, and spending real time on an idea before you know if it deserves it is a genuine tax on experimentation. That tax is gone. What I did not expect is how clearly its absence would reveal the thing it had been quietly hiding: the opening was never the hard part. It only felt hard because it took effort. The middle takes something effort alone doesn’t buy.

I’ve watched this play out in rooms that had nothing to do with fiction. A team gets a working demo by Friday, and the room reads that speed as validation, because speed used to be evidence of conviction, back when it was expensive. Nobody in the room talked to a user. Nobody wrote down what failure would actually look like. The demo is real, and it is also just proof that the tool works, which is a different claim entirely from proof that the idea deserves the next six months. The cheaper the prototype gets, the more validation it should buy you, not less, because you have that much more room to spend the savings on the conversation you used to skip for lack of time. Most teams spend the savings on a second demo instead.

The actual failure I keep watching isn’t a bad prototype. It’s a good one, built fast, that a team then mistakes for the whole job. The energy of the first working version is real and it is also temporary, and organisations built around sprints and launches are not naturally built to notice when that energy has quietly become the only thing still moving the project forward. A novel in that state is easy to spot; the writing goes slack and I feel it in my own hands. A product in that state looks like momentum right up until the metrics arrive and everyone is surprised.

What actually gets an idea through the middle, in either craft, is a decision that has nothing to do with capability and everything to do with a kind of stubbornness I don’t think a model can supply. In fiction it’s the willingness to keep writing a scene after you’ve stopped being sure it’s good, because the only way to find out is to finish it and look. In product it’s the willingness to keep talking to the five users who are lukewarm about version one, instead of retreating to build version two on instinct, because the tool makes building version two so cheap that talking to anyone starts to feel like the slow option. Both are the same mistake wearing different clothes: treating the ease of producing the next draft as a reason not to sit with the one you already have.

I don’t have a framework for the middle, and I’ve stopped looking for one, because I think the absence of a shortcut is the actual finding. Every novel I’ve finished was won in that unglamorous stretch, not the opening chapter, which any of us can now generate a dozen versions of before lunch. Every product I’ve watched survive contact with real customers was carried there the same way, by someone who stayed after the first version stopped feeling exciting and kept asking what was actually true, long after asking had stopped being fun. AI made the folder of abandoned first attempts fill up faster than ever. It did nothing to make the walk out of that folder any shorter. That walk was always the whole job, and it still is.