How much direction does AI need?

From Tiny Bottle AI

Hello!

The Drip

How much of your opinion about AI comes from actually using it?

Ethan Mollick’s “The Overhang” gave us a useful starting point: the gap between what AI can do and what people are doing with it. Even if model development stopped, we would have plenty of work ahead figuring out how to use what already exists.

But a demo and a process your business depends on are different things. Getting some firsthand experience helps you see the distance between them.

Inside The Bottle

Justin’s personal habits have changed in two different directions.

When building software, the starting point used to be very detailed: defined epics, small issues, lots of decisions made before the agent got to work. That was familiar from years of building without coding agents.

Now there’s more room for the agent to propose options. Start with the outcome. Ask it to help clarify what you want. React to the alternatives instead of deciding every detail up front.

That changes where the creative work happens. If you’ve already decided the switch must be blue, you’re not really asking what the agent would propose.

With strategy and planning, the change has gone the other way.

A few bullet points could produce something that looked like a strategy. The uncomfortable part came later, when someone asked why a particular choice was in it. If the answer was essentially “I didn’t realize the AI put that in there,” the document had got ahead of the thinking. People are now affectionally referring to this as “slop grenades”.

So the work becomes more methodical. What don’t we know enough about? What research would help? What are the competing points of view?

Use AI to bring in information and options. Form an opinion. Then ask for help articulating it.

Research has its own limits, too. An answer can sound complete while leaving out something important. Asking where the blind spots might be is a useful prompt, not proof that you’ve found them all. Books, your own experience, and ideas outside the answer still have a role.

Try different amounts of direction on work you can assess. Notice where the agent surprises you, where your instructions constrain it, and where you need to stay closely involved.

That experience is useful even when the result disappoints you. You’re learning something more specific than “AI is good” or “AI is bad.”

And as the tools get better, checking the work still matters. Confidence and accountability need to grow together.

If choosing a starting point for your business is the hard part, AI Pathfinder lets you request a free Pathfinder Brief prepared by a real human.

Lab Notes

What Stopped Our Scroll

Unsubscribe