Treat AI like a teammate with a memory problem

Here's a thought experiment I use with business audiences. How would you run a company whose CEO loses their memory every night?

It sounds like a dealbreaker. But sit with it: if that CEO were talented enough to warrant it, you'd figure it out. You'd build the morning briefing. The standing documents. The decision log. The catch-up ritual. You'd engineer around the memory problem, because the talent justifies the engineering.

That's the most useful mental model I know for working with AI.

AI is like a teammate with amazing talent and a goldfish memory. The talent is real. So is the memory problem. Plan for both.

Why smart people bounce off AI

Most of our instincts about computers are precision instincts. Computers are exact: same input, same output, no coaching required. People walk into AI with those expectations and meet something that behaves less like a calculator and more like a brilliant new hire with amnesia — capable of nearly anything, but holding none of your context unless you hand it over, every time.

The mismatch is why people bounce. They expect computer-precision, get person-with-a-memory-problem, and conclude the thing is broken or overhyped. It's neither. It's a different kind of collaborator, and it rewards a different kind of management.

What the mental model tells you to do

• Brief like you'd brief a talented stranger. Context, goals, constraints, examples of good output. The quality of what you get tracks the quality of what you gave.
• Write things down once, reuse them forever. The documents you'd hand the amnesiac CEO — what we do, how we talk, what good looks like — are exactly the documents that make AI useful. Most businesses don't have them. Writing them pays off twice: your AI gets better, and so does your next human hire.
• Consider its perspective. When the output is wrong, ask what it couldn't have known. Most "hallucinations" I get called in on trace back to ambiguity the system was never equipped to resolve. That's a debugging signal, not a verdict.
• Let the relationship compound. One-off prompting is like hiring a new team every morning. The gains show up when context carries forward — shared language, accumulated examples, standing instructions that get sharper over time.

The punchline

The teams getting real value from AI aren't the ones with secret prompts. They're the ones who took the memory problem seriously and built the briefing systems — the same way they would for any talented collaborator who needed them. The capability arrived years ago. The relationship is the unlock.