I set up a Claude subscription for my team
A day later I got sent a blog that said we were quietly innovating. Cringe.
It was time to declare an AI amnesty.
I knew my entire team had been running personal subs for years so it was time to setup a team subscription. A day or two later I got sent some social posts and a blog to review and sign off. One had the phrase seamless payments in it. Which I hate as everyone in payments over uses it and I’ve told this team specifically not to use it. The blog said we were quietly innovating. Yes mga cringe. Every AI model now spits out the word quietly for some reason. It’s up there with load-bearing and em dashes.
My first reaction wasn’t even about the AI. It was you know the brand guidelines so why didn’t you just fix this before sharing it?
I’m not sure exactly how it happened because the team did use the right briefing with the AI. They are aware of the AI guidelines, but I think they just assumed that whatever the team AI came back with had to be the right answer or more likely, I felt that they just didn’t apply the same brain power to the reviewing as they would normally if they’d written posts themselves.
This got me thinking.
We needed some way to centralise the quality, but not centralise every decision within the team. Centralising every decision would drive everyone nuts.
Centralising the decision means one person or one committee approves everything. Nobody wants that level of straight jacket. And every founder’s right to be scared of that as it’s slower than doing nothing. Plus it kills morale and the will to live. What I actually mean is the company owns the truth.
New roles will fall out of this. Nobody had a digital marketing manager or social media manager job title in 2004. They arrived a few years later, once someone had to actually own it and the CEO needed to know who to blame when it went wrong. I think the same happens here with jobs such as an AI content controller appearing or even something like a truth seeker, whose actual job is getting what’s in people’s heads onto a page before it disappears when they leave.
And by truth I mean the stuff that should be written down and not just live in whoever’s head that’s been there the longest. I mean the brand guidelines, product information, customer insights and all other meta data that describes every aspect of the company.
Centralising control means the company owns the truth and if centralise all this company information, then you can create agents that act as quality controllers. But what stays individual is the judgement. You know the call in the moment such as deciding whether a Sam Altman joke lands or understanding what is happening culturally. That’s the human part.
Get it right though, and who knows, maybe you actually fix the thing that’s broken at every company I’ve worked at.
But centralising control at a company level is going to be extremely challenging. Ask anyone who has worked in digital transformation. The technology is the easy part, but dealing with messy humans sends some people to early retirement. That’s why I think it’s best to start this at department level. If each department head owns their own department’s memory first, before anyone tries to solve the whole company, you’d get most of the value. Start small and don’t try to boil the ocean.
P.S. This week’s soundtrack “Everything In Its Right Place” by Radiohead.


this is something that came up in a conversation with a friend of mine that's at a very large organisation. he faced this exact issue that founders can get ai to align with their values but the other employees write things that are plausible but not value aligned.
I'm working on a simple solution using obsidian and a tool i built i would love to have you and your company as a beta tester to see if I managed to solve your problem.