Dr. Jeff Wurfel

Neuroscientist · AI practitioner · Writer

I build AI for a living, and write about using it well.

I’m Jeff Wurfel — a PhD neuroscientist, AI Tech Lead in film production, and the author of Before You Build. These are essays on judgment, craft, and closing the distance between what AI promises and what it actually delivers.

Read the essays Subscribe

Recent writing

  • AI Is More Human Than You Realize

    The famous AI bloopers are real, but judging AI by its worst first-month mistake is the wrong standard. Treat it like a capable, fallible new hire, through clear briefs, handoffs, review and feedback, and the work changes.

  • We've Been Using AI for Twenty Years

    ATMs didn't kill the bank teller and spreadsheets didn't kill the analyst. The fear that AI means less work rests on a picture of the economy that's been wrong every time we've tested it. The real question isn't whether AI takes the jobs, but who gets the time it saves.

  • The Last One Percent Is the Whole Job

    A demo is the best two minutes of a hundred hours, and your own working proof of concept is a trailer too, because you were the error-catcher. Why the last one percent is the whole job, and how to climb it in rungs.

  • Colearning

    Everyone is waiting for AI that learns you. A carry-forward doc is the version you can build today: the shared memory between you and the model, readable, correctable, and portable to whatever tool comes next.

  • The Step Before the Step

    The most useful habit with these tools is backing up one step: ask the AI what you should be asking before you ask for the thing. The best prompt is often the one you erase.

  • The Excuses Used to Be True

    The excuses for living with broken things — not technical, not worth the time, wait for IT — were accurate pricing for decades. The price changed. The reflex didn’t.

  • Being the Friction

    The hard part was never knowing the right questions to ask before an AI build. It’s getting a room that was promised easy to run them anyway — and still getting invited back. Three moves, and one for whoever runs the room.

  • Validator, Not Bottleneck

    Two ways to be the human in the loop: validator or bottleneck. Bottlenecks review everything, feel essential, and get designed around. Validators make their judgment legible — and get handed the next loop.

  • Why 95% of AI Projects Fail (and How to Land in the 5%)

    MIT found 95% of enterprise AI projects deliver no value — and winners ran the same models as losers. The failure isn’t the tech; it’s the first conversation — and that’s the part you can fix.

All essays →