Over the last few weeks, I’ve spent a lot more time building with AI instead of simply writing or talking about it.

I’ve been developing an adaptive AI assistant, continuing to improve the Decision Debt app, working through AI-enabled product decisions, and dealing with all of the small questions that appear when an idea becomes something people might actually use.

And it has reinforced something I’ve believed for a while:

The hardest part of AI isn’t the AI. It’s deciding what the AI should be allowed to do.

It’s relatively easy to imagine an assistant that can summarize your inbox, prepare you for a meeting, remind you about something important, or surface a decision you’ve been avoiding.

The harder questions come immediately afterward.

Should it act automatically?

Should it ask first?

How much should it remember?

When should it interrupt you?

What information should appear in a notification?

When is personalization helpful, and when does it become uncomfortable?

What happens when the AI is technically capable of doing something that the user probably doesn’t want it doing?

Those aren’t really engineering questions.

They’re decision-rights questions.

And the same problem shows up inside organizations adopting AI.

Companies spend a lot of time asking:

What can AI automate?

A better question is:

What decisions should we delegate, what decisions should AI support, and what decisions must remain explicitly human?

That distinction matters.

If we automate everything simply because we can, we risk removing the moments where judgment actually develops.

If we require humans to approve everything, we lose much of the value AI can provide.

The goal isn’t maximum automation.

It’s intentional delegation.

That has become one of the principles guiding what I’m building:

AI should reduce the cost of making good decisions without quietly taking ownership of the decisions that matter.

I think that principle applies far beyond personal assistants or productivity tools.

It applies to AI governance.

It applies to leadership.

And it applies to almost every organization trying to figure out what “AI transformation” actually means in practice.

There’s a big difference between adding AI to a process and redesigning how decisions get made.

The second one is where things get interesting.

— Matthew

Question for you: Where would you draw the line? What decisions are you comfortable delegating completely to AI—and which ones would you never hand over?