One of the first things I did when I got back to my laptop at the end of my summer break this week was open a client summary document I’d put together when I was shutting up shop in mid-July.

I do these documents as a little routine to help me capture all the context and nuance for the work I’m doing with each client before I go away. It helps smooth the path back into the engagement for me when I restart work.

And this document was good – it was clear, well structured and the recommendations that July Simon made for September Simon all sounded sensible.

And yet there was one section where I couldn’t tell you why any of it said what it said.

I knew I’d written it. I could see my written fingerprints all over it, but the reasoning underneath – the bit where you weigh one option against another and land somewhere – I just couldn’t reconstruct it in my head.

To be honest that unsettled me more than I expected it to.

I mentioned in last week’s Build that time away each summer gives me a distance from the day-to-day that I don’t get otherwise. And this realisation was one thing that this distance showed me this summer.

Where I actually stand on this

Before I go further I should say where I stand on all things AI, because I think it changes how the rest of this reads (and as I enter my 50’s, I’m increasingly conscious of not becoming “that” person).

I’m not an AI sceptic. I use AI tools every single day. Overall my work is better and faster for it and I’d struggle to go back now.

But I’m also not in the camp that thinks it’s about to run your business for you. I’ve watched enough founders spend real money against that promise to be wary of it. I believe in the value of humans and organisations in work now and for the long term.

So in practice that means I’m somewhere in the middle – an AI pragmatist I suppose. That mostly means I’m as interested in what it costs as in what it saves.

What I’ve noticed over the last six months

Nearly every founder-led business I’ve worked with this year has gone from dabbling with AI to genuinely running on it. That means moving from seeing “AI” as a project with a steering group to being just the way work happens nowadays.

That’s mostly been a good thing. Things that used to take a fortnight take an afternoon.

But in almost every client business I work with I’ve also noticed some tricky stuff building up quietly underneath the speed.

I’ve written before about culture debt – the stuff you borrow against every time you make the fast people decision rather than the right one. What I’ve been seeing this year has similar vibes to it.

You get the benefit now. But the real cost to you turns up later when it’s too late to think about it or do anything about it.

I think there are broadly four kinds of these issues that I keep seeing (including in my own work):

1. An “understanding” debt

This is the one I found from my pre-holiday document.

We can all produce something now in a fraction of the time it used to take. You skim it, it looks about right and you move on to the next thing. Job done!

But you didn’t do any of the intermediate deeper work. You didn’t weigh the option you rejected or write the sentence that forced you to be honest about what you actually meant.

So you end up owning an output you don’t fully understand and, more importantly, don’t really believe in.

I notice this one most when someone challenges the work. If you did the hard thinking, you can defend it from first principles and change your mind in the room. If you didn’t, all you can really defend is what the document happens to say. That doesn’t really hold up in any meaningful debate.

2. A “confidence” debt

So knowing about the “understanding debt”, you slow down a touch. You read everything properly before you use it and you put in the effort to stay on top of what the AI is doing on your behalf.

That’s great so far, but then a different problem turns up.

Are you sure this is right? Could it be better? Would a better model have given you something subtly sharper? Did the bit of context you didn’t think to include send you somewhere you shouldn’t have gone?

Often there’s no way to know. And increasingly I’ve noticed low-level doubt creeping into work that people would once have been perfectly happy to put their name to.

Nothing has gone wrong and they’re producing more than ever, but it feels like people are quietly less proud of what they’re producing. I do wonder where this growing lack of belief in your own work output might show up in future.

3. “Skill fade”

Pilots and surgeons use this term. It recognises that a competency you don’t use regularly degrades, whether or not you notice it happening.

I think a version of this is going on for a lot of us.

There are whole areas of judgement I’ve built up over the years (a fair bit of it as an MD running a digital agency and a lot working with multiple founder-led businesses since I went independent).

I worry that I now rely on that judgement far less often, because I lean on the AI to take at least a first pass at something first.

And yes, I’ll admit I’ve caught myself opening a Claude window for things I’d simply have done myself back when doing it myself was the only option available.

I don’t think the skill vanishes overnight. What bothers me more is that you probably won’t find out it’s faded until the point you need it without any assistance.

4. A loss of continuity

AI has no sense of how your business got to where it is.

Every session starts from a lower level of long-term continuity than you or your team have.

It probably doesn’t know you tried this in 2024 and it didn’t work out, or why that person is cautious about that particular client or how your last restructure actually landed with the team.

It can’t hold long-term care for anything it touches because it isn’t around long enough to. It only knows what it holds in memory, which is restricted by the time it’s been deployed as well as restricted to what can be captured in writing, transcripts or analysis.

Which means someone has to think about how to incorporate the relevant aspects of that wider history instead – writing it down, keeping it current and feeding it back in every time.

And that’s real work and arguably impossible to do to the extent that you can as a human, eating into exactly the time you were hoping to save.

What’s underneath all four things

I’ve spent a lot of hours thinking about this but keep arriving at the same place: this is all about judgement.

Each of those four is a small withdrawal from the same “pot”: your ability to know what good looks like here, in your business, with these people, right now.

AI tools amplify whatever you bring them. If you’re clear about what you’re building, they’ll get you there faster than anything.

If you’re not clear, I wonder whether they just help you produce a great deal of very convincing material whilst you drift off somewhere you never meant to go, or they help create a load of unnecessary “busy work”.

The tools don’t supply the direction. They help you think, but they don’t think for you as well as you can.

So what do you actually do about it?

I want to be clear – I’m honestly not arguing for less use of AI by founders.

It might be useful to share a few things I’ve started doing myself to try to consciously address these four things:

• I try to make a provisional decision or take an initial view on an issue before I ask for the output rather than after. It’s interesting how this changes the way I prompt the AI and the nature of the interaction I have with it. It helps AI act as a “thought partner” rather than as a “Simon substitute”.
• I am trying to pick a few things each month to do the hard way on purpose, just to keep those muscles working. I continue to write the majority of my content (and client output) manually, although I lean on AI heavily for copy editing and post-production quality control. For me, the act of writing is intrinsically bound up with my thinking processes.
• And I’ve got much better at writing down why we decided something, not just what we decided. That one is a big win for clients who I’m helping with things like decision-making in their businesses.

So as usual I leave you with a few questions worth pondering as you open up ChatGPT this morning:

• Which document in your business right now could nobody on your team defend from first principles if a board member pushed hard on it?
• What’s the thing you were once really known for as a leader that you’ve quietly stopped doing yourself?
• Where in your business are these four issues I’ve spotted building up fastest, and would you actually notice them before they really hurt your business?

This is what I’ve spent much of this summer chewing over and I know we’re nowhere near understanding the impact of AI on how humans and organisations work.