Who Should Run Your AI Project? Not the Person Picking the Tool.

Shrimp farmers will argue for an hour about which genetics line to buy. Ask those same farmers who made sure the animals actually arrived alive, and most of them couldn’t give you a name. On my farms it was a specialist veterinarian, and almost nobody on the operation understood how much of the business quietly rested on him.

Moving live broodstock across international borders is brutal. Health certifications, import permits, oxygen and temperature in transit, a customs officer having a bad morning, a flight that slips six hours and turns a healthy shipment into a dead one. Every single shipment invented a new problem in real time. No catalog solves that. No supplier solves that. A person who has already lived through forty versions of it solves that.

The tool is the easy decision

I get asked which model, which platform, which vendor. It’s the wrong first question. The honest state of things in 2026 is that the mainstream AI tools are all broadly capable, they leapfrog each other every few months, and for most small and mid-sized businesses the difference between the top options is noise next to the difference between a project run well and a project run badly. That isn’t a technology gap. It’s an execution gap, and no purchase order closes it.

What the specialist is actually for

I spent over a million dollars on shrimp genetics and took survival from a stuck 15–20% to over 90%. I want to be precise about how that happened, because it’s the whole point: I did not buy that number. There was no line on a price list that said “90% survival.” That result came out of years of disciplined selection — hundreds of small judgment calls about what to keep, what to cull, and what to ignore when the fashionable advice said otherwise. The genetics were the output. The judgment was the input.

AI work is the same shape. Somebody has to decide which problem is worth pointing the thing at, notice when the data is too dirty to trust, tell you the honest answer when a promising pilot isn’t earning its keep, and kill it before it becomes a line item you defend out of pride. You can buy a tool. You cannot buy the judgment about when to use it, and more importantly, when not to.

The question to ask before you sign anything

Ask it plainly: who, specifically, is doing the work? Not the firm. The person. Will they be in the room when it breaks, or is the senior name on the proposal a closer who hands you off to a rotating bench the week after you sign? That handoff is where most engagements quietly go sideways, and it is exactly why I work as a fractional advisor you actually get rather than a brand you get billed by.

There’s a real tradeoff here and I’ll name it: one operator is not a bench. If you need eight workstreams running in parallel across a global org, hire a firm — that’s what firms are for. But that is not most companies. Most companies need one thing to work, once, and then to understand why it worked. For that, the deciding factor was never the tool. It was who was holding it.