What a Failing Shrimp Farm Taught Me About AI Projects

For two years, I spent more than a million dollars trying to kill a bacteria. It didn’t work. And it turned out to be the most valuable AI lesson I ever learned — years before I was doing any AI work.

A million dollars, fighting the wrong thing

This was my shrimp operation. We had Vibrio — a bacteria that was wiping out our crops, with survival rates stuck at 15 to 20%. So I did what everyone does: I threw money and technology at the visible problem. Monitoring systems. Disinfection. Better filtration and aeration. Specialized feeds. Probiotics. Consultants. Every lotion and potion the industry sells. A million dollars in, survival hadn’t budged.

Then we changed one thing. Not the equipment, not the water, not the protocols — the animal. We switched to a genetically stronger line of shrimp. Same farm, same everything else. Survival jumped to over 90%.

The bacteria was never the cause

Here’s what that taught me: Vibrio was never the cause. It was the result. The bacteria was always in the water — it just preyed on weak animals. I’d spent a million dollars fighting a symptom while the real problem, fragility, sat there untouched.

Failing AI projects are the same story

A company tries AI, it doesn’t deliver, and they blame the tool. Wrong model. Bad vendor. We need a better platform, a bigger dataset, the newest thing. So they spend more — more lotions and potions — and it still doesn’t work.

But the tool is rarely the cause. The failure is the resultof pointing technology at the wrong problem, or bolting it onto a broken process underneath. You can buy the best AI on the market, wire it to a workflow nobody actually uses, and you’ll get yourself a very expensive 15% survival rate.

Find the weak animal

So when a project is failing, I don’t start by asking which model to use. I ask: what’s the weak animal here? What’s the underlying process, incentive, or data problem that would sink anytool you pointed at it? Fix that, and often the “AI problem” turns out to have a boring, cheap solution — sometimes not even AI at all.

That’s the whole difference between chasing tools and solving problems. One drains a budget. The other moves a number — like 15% to over 90%.