Who’s keeping your company’s AI up to date?

Hello fellow humans - before we dive in, here’s a reminder of the resources I’ve put together for you:

IN THIS EDITION

Google started rolling out skills to Workspace, which lets teams save and reuse instructions. For now, though, skills don’t sync between the Gemini app and Workspace apps. If your team uses both, check which version of your instructions each one is working from.

GOOGLE · OCT 19

OpenAI retired admin-managed sync connectors and is moving Enterprise teams to real-time Google Drive, SharePoint, and Teams apps. Existing permissions still apply. If your team relied on synced company knowledge, ask it a few of your usual questions to make sure it still finds the right sources.
OPENAI · OCT 16

Microsoft’s Project Perception is built to monitor agents working inside companies, with their own identities, permissions, and controls. Its point is that an internal agent can cause problems without ever being hacked. So next time you give an agent more access, be clear about what it can change and who can stop it.
MICROSOFT · OCT 19

I’m teaching a free, 30-minute Lightning Lesson with Yusuf Ahmed, founder of Zevenue, today at 2 pm ET.

We’ll cover what GTM engineering is, the five stages of a GTM workflow, and how to map a first workflow for your team. If you work in sales, growth, or RevOps, join us to see how AI can help your team research accounts, prioritize prospects, and run more useful plays.

Last week, an operations leader told me about a new employee who asked their internal AI assistant how many people worked in one of their offices. The assistant pulled a number from a presentation from several years ago.

That’s a pretty inconsequential mistake, but it gets at a problem I’ve been spending a lot of time thinking about: how do you keep a company’s knowledge in good enough shape for people and AI to work from it?

In last week’s issue, I wrote about people getting better at AI individually while struggling to work together. A big part of that comes down to the information everyone is using.

In practice, this looks like everyone having their own stash of company knowledge. One employee has a Claude project full of docs, another has a set of saved prompts, and someone else just built an agent. They’re all working from slightly different versions of the truth. So two people can ask their AI the same question and get two different answers, and neither of them knows which one is right.

I hear versions of this across a lot of the companies I talk to. And I think it’s going to become a much bigger problem as AI takes on more work.

Keeping company information organized has always been difficult. Anyone who’s searched through five versions of the same deck knows that.

But people used to fill in a lot of the gaps themselves. They knew the presentation was old, or that the customer had changed their mind since. An AI assistant needs a reliable way to get that context too.

At the same time, we’re creating much more material for it to sort through. One person can now generate research, meeting notes, plans, and little apps throughout their day. Each might contain useful information, but each can also become another version of the company’s knowledge that someone needs to maintain.

Here’s an example. Your team agrees to change a customer’s delivery date. You update the project plan, but the old date is still sitting in a meeting summary, a sales deck, and an agent’s saved instructions. Then someone asks the agent to draft a customer update.

Which date does it use?

Even if it shows you a source, you still need to know whether that source should be trusted.

And once agents start sending those updates or changing records themselves, the cost of getting this wrong goes up.

Some people call the solution to this a “company brain.” If you’ve come across the idea of a “second brain” in personal knowledge management, this will sound familiar.

I’ve been in that space for a long time. A lot of its principles become even more useful when you work with AI. Save what matters, organize it around how you’ll use it, connect related ideas, and keep it current.

AI can help with all of that. But capturing everything also gives you a lot more to sort through. You still need judgment about what deserves to become part of the knowledge you rely on.

That gets more complicated when a whole company is involved. People disagree. Some information is private. A suggestion from one person might conflict with a decision someone else has already approved.

The company brain needs to preserve those differences. Otherwise, you’re giving everyone easier access to a mess.

I find it helpful to break company knowledge into three things both people and AI need:

That last one is easy to lose. A document might tell you what the team decided without explaining why. The reason can matter a lot when someone tries to apply that decision to a new situation.

I’ve been exploring this as a “context layer”: a way to maintain useful, shared knowledge across the places where work happens.

People should be able to keep creating information naturally through calls, messages, documents, and their AI tools. I don’t think we’ll get very far by asking everyone to put every thought into the same template.

But we do need a shared way to decide what becomes official, and to make sure everyone (people and agents) is working from that version.

Imagine a living view of what’s happening with a customer, pulling from your CRM and project plan. Every detail links back to where it came from, and anything that’s still undecided is clearly flagged, so nobody mistakes an open question for a done deal.

The CRM would still own the official account details. The project plan would still own the schedule. The brief would bring the relevant pieces together so people don’t have to reconstruct the whole story every time they need to do something.

When new information comes in, the system could propose an update. Routine changes might happen automatically. A conflict about what we promised the customer should go to someone who can settle it.

And that maintained context should be available to the different AI tools the team uses, within the same access rules.

There are already useful pieces of this in existing products. For example, Notion lets teams assign owners and verify pages, with reminders to review them again. Tools like that help, but somebody still needs to decide whether the information is right.

There could be efficiency benefits, too. If several agents need the same customer context, maintaining it once may save repeated searches and processing. That could reduce token costs, although keeping the context updated has a cost of its own.

I think context management will become a much more important field, with specialist roles of its own. I also think the basic skills will become necessary for anyone responsible for a team’s knowledge.

A sales manager should understand how a new customer commitment gets recorded and reaches everyone who needs it. A team lead should know how an approved process replaces the old one. People building agents need to understand which sources those agents should use and what happens when the sources disagree.

You can have a central team providing the tools and common rules. But the people who understand the work need to help maintain its meaning.

This is why I don’t expect technology to solve the whole problem automatically. Software can tell you there are two different deadlines floating around. But it takes a person who actually knows the customer to say which one we promised.

First, figure out where the important facts about it actually live, and who’s allowed to change them.

Then try a quick test: think of something that changed recently (a new deadline, a pricing update) and ask your AI about it. Check the answer, and check where it got that answer from.

If it comes back with the old info, trace where the update got stuck. Maybe it’s an outdated doc, or a decision that never made it out of a Slack thread. Either way, you now know exactly where to start fixing things.

New section alert. I’ve spent over a decade in AI, helped more than 300 teams put it to work, and taught 70,000+ people through AI For Anyone. Along the way, I’ve picked up a few lessons, and I want to use them to help you where you’re stuck.

So hit reply and ask me anything. Each week, I’ll pick one question and answer it here. I’ll keep your name and company out of it unless you tell me otherwise :)

That’s it for this week’s issue of Not A Bot (formerly known as AI Ready, formerly known as Not A Bot). Thanks for reading.

If today’s problem sounded relatable, come work through it with me at the free workshop I’m hosting this Friday.

Over two hours, we’ll take one recurring task from your organization, map where the work gets stuck, and decide on a practical first fix.

There are only 30 seats, and requests are open now.

Apply for your free seat

Until next time,

Haroon