My board needs a diary

I looked at my AI board today.

I hadn’t looked at it for a while.

This sounds grander than it is. I have three simulated advisors in ChatGPT for our jewellery business: a leverage strategist, a jewellery operator and a sceptical capital allocator.

They look at the same situation through different lenses. I asked for their opinions on how things were going, and got three perspectives back.

Useful.

Then I thought: why am I waiting until I remember to ask?

My ChatGPT already has quite a lot of context about the business, and connections to places where the work lives. Could it put together a standard report, have the simulated board review it, and bring it back on a schedule?

What has changed? What is progressing? What needs attention? What decisions do we need to make?

And why should that only be for me?

The team could have the report too. We could be looking at the same account of what is happening, with the questions and disagreements alongside it.

That is what we’re exploring. The plan is to try it weekly. Every other week might make more sense. We’ll see.

I found myself thinking about how much work goes into preparing for an actual board. Gathering information, writing the report, anticipating the questions.

You could have a simulated board help you prepare for your real board.

A board for the board.

Hysterical.

If I wanted to prepare for actual advisors, I could gather their public writing, interviews and conversations we’ve already had, and use that as reference material. Their priorities, communication style, recurring concerns.

That might help anticipate their questions. It could reproduce their biases too.

Or choose a different lens altogether. Warren Buffett-inspired, for example. Useful for some questions. Probably beside the point for others.

Being famous doesn’t automatically make someone the right person to ask.

The interesting bit is that the simulated advisor can become part of a recurring process. The information comes in, it gets examined, a report comes back, and we decide what to do.

That possibility feels quite different from opening a chat and asking a question.

It also brought me back to a conversation at dinner yesterday. My father-in-law was talking about human judgment and trust.

I found myself thinking about the phrase “AI is just a prediction engine.”

My brain went: hang on. Isn’t prediction quite a lot of what our brains do too? Drawing on experience, recognising patterns, anticipating what happens next?

I don’t know how far that comparison holds. Calling both things prediction engines doesn’t make them equivalent, or tell me which gives better advice.

But it leaves me curious about what I’m actually trusting when a recommendation feels right.

I can also ask for pros and cons, ratings out of ten across several dimensions, and weighted scores. Change the lens, change the criteria, change what matters most.

Even for design.

How much of taste and judgment can be made explicit enough to give to a machine?

For something I can test quickly, perhaps it helps me make a hypothesis. Put the design in front of customers, get feedback, adjust. The score gives me assumptions to examine; the customer response gives me something else to go on.

Choosing the dimensions and weights is itself a judgment.

For a big decision that is difficult to reverse, that loop may be much slower. Maybe the board is useful there. I don’t know yet.

Which brings me back to the board.

It can produce a convincing report. Three perspectives can sound thoughtful. They can disagree in ways that feel useful.

How do I tell whether the judgment is good?

That question applies when I listen to a human too. Confidence and reputation can influence how advice lands. Three simulated advisors could also share the same blind spot.

I don’t yet know how to fully distinguish useful judgment from a convincing performance of judgment.

That’s something to work on.

In the meantime, I’d quite like the board to turn up next week.