AI made starting free. Finishing is still the job.
AI dropped the cost of starting a project to almost zero. That didn't shrink my backlog. It made it worse.
AI has made starting a project almost free. That’s the part nobody warns you about. Going from nothing to a working prototype used to cost enough time and effort that it killed most of my bad ideas before they ever got a folder. The cost was a filter. Now the filter is gone, and my backlog has never been longer.
Nearly two decades in, I’ve collected ideas the way other people collect open tabs. Most have lived in my head for years, half-formed, waiting for a free weekend that never quite arrives. For most of my career that was fine, because starting was hard and so most of them stayed ideas. Then I folded AI into how I work, and starting got easy. Too easy.
The 80% wall
Here’s the pattern I keep repeating. A new idea grabs me, I start, and the attention holds beautifully. Right up to about the 80% mark. The thing works. You can see what it’s for. And then, almost on schedule, a newer idea walks past and I’m gone, the last one left sitting at 80% forever.
I used to think this was a focus problem. It isn’t, or not only. It’s that the first 80% is the fun part. That’s where you get the rush of a blank canvas, the quick wins, the feeling of a thing coming alive under your hands. The last 20% is a different job entirely: edge cases, the boring copy, the deploy that fights back, the small ugly decisions you postponed. None of it feels like creating. All of it feels like chores.
AI is brilliant at the first 80%. It will get you off zero faster than you can believe. What it quietly does, though, is lower the price of starting without lowering the price of finishing. So the ratio gets worse. You can stand up five new prototypes in the time the old you would have spent grinding one across the line.
Finishing is a separate skill
The thing I had to admit is that finishing isn’t just “more of starting.” It’s its own discipline, and it’s the one that almost nobody practices, because the fun ran out three steps ago. A pile of 80%-done projects isn’t a portfolio. It’s a graveyard with good intentions for headstones.
This is also, not coincidentally, exactly what clients pay me for. Nobody hires me to feel the rush of a fresh idea. They hire me for the last 20%: the unglamorous closing work that turns a promising draft into a thing that actually ships, holds up, and does its job on a quiet Tuesday six months later. The demo is easy. Shipping is the trade.
So I’ve stopped treating the backlog as a list of things to start. I treat it as a list of things to finish, one at a time, in public if that’s what it takes to keep me honest. The next idea can wait its turn. It always thinks it’s special. It rarely is.
AI made it cheap to begin. It didn’t make it cheap to finish. Begin less, finish more, and the backlog stops being a graveyard and starts being a body of work.