Woven In, Not Bolted On
Why so many AI projects get quietly shelved, and what the durable version looks like.
Everyone is telling you to start weaving AI into your business this year. Do it now or get left behind.
Here's the part they leave out.
Analysts expect more than four in ten of these AI projects to get scrapped within a couple of years. Paid for, launched, then quietly shelved after the money is spent. You can see it already. Some of the companies that went loudest on AI a year ago are quietly walking it back.
I see a version of this up close, from two directions.
The two failure modes
Founders offer me equity and an advisor seat to help build their AI idea. I pass on most of them. What they're building is usually something the next version of ChatGPT or Claude just does natively. A feature waiting to get absorbed. The product isn't durable because it's a thin layer over someone else's model, and the someone else keeps shipping.
Same story one level down, inside operating businesses. Fly-by-night AI consultants are selling owners agent frameworks and automations with no real grasp of how a business actually runs. It works for a month. Then it breaks, needs constant babysitting, or gets ripped out. Money you don't get back, and worse, a team that's now allergic to the next attempt.
Two different failure modes, one root: something bolted on where nothing was defined underneath.
Why the tech isn't the problem
It's almost never the tech that fails. The models are good and getting better on a schedule you can nearly set your watch to.
It's that the wrong solution was bolted onto a broken process.
AI is like having an employee who never sleeps and rarely stops to ask questions. That sounds like a feature until you sit with it. Every good human employee quietly corrects for your bad instructions. They see the process doesn't make sense and fix it without telling you. AI doesn't. Point it at a job that hasn't been defined and it just runs the wrong way faster, at scale, with confidence.
The durable version
Here's how we think about it. Human first, AI native.
Fix the organization first. Get the direction and the strategy clear. Build a framework solid enough that AI can be woven into it, not bolted onto it. Woven means the machine sits inside a defined process with a defined owner and a defined standard. Bolted means it sits on top of a mystery and automates the mystery.
The owners getting real results do the boring part first. They pick one process that matters. They write down how it's supposed to run. Who owns it. What done looks like. Then they explore handing what they can to a machine.
The AI doesn't make them the money. The clarity does. The AI just runs it faster.
That's the game right now. Not who sprinkles the most AI on top. Who has an operation clear enough that AI has something solid to stand on.
You can't automate a process that only lives in your head.
Where have you added a tool, AI or otherwise, when what you really needed was a clearer process?