the model was ten percent of it

every founder building on ai right now is fighting over the model. a guy i spoke with this month thinks that is the least interesting tenth of the job, and his client list suggests he might be right.

he started as a data consultant and noticed something structural. in enterprise work, the people who advise and the people who build are two different companies. somebody consults, draws the architecture, writes the deck. then there is a handoff, and builders pick it up, and the handoff is where the months go to die. a week of work becomes a quarter because it changed hands three times and lost its meaning at every exchange.

so his firm does both. that is the whole wedge. not a model, not a wrapper. removing a handoff.

then the line that stuck with me. everyone put the model in the front seat, and the foundation, the operations and the governance in the back. that is backwards, and it is why so many pilots stall before production. the model is maybe a tenth of the problem. the other nine tenths is whether the data underneath is real, whether anybody owns the thing when it misbehaves at three in the morning, and whether the company can explain what it did afterwards.

he also says something that cuts against the entire mood of the moment. emergent behaviour is a liability in the enterprise, not a feature. and for his buyer he is dead right. a bank does not want to be delighted. a bank wants to not be surprised. the demo that makes the room gasp is usually the demo that never ships.

his first question for anybody stuck in pilot purgatory is not which model. it is why do you need this at all, and are you after a business improvement or a process improvement, because those are different problems and one of them does not need ai in it anywhere.

he compared this moment to the fibre laid across the world in the late nineties. enormous capacity, most of it dark for years, and almost none of the value captured by the companies that laid it. that comparison is not bearish. it says the money ends up one layer away from where everybody is currently standing.

the boring layers are where the durable companies get built. plumbing, ownership, accountability. nobody posts about governance.

what are you putting in the front seat?