Exponential Operator is a 100% human-written newsletter by Sawyer Middeleer, founder of Revi Systems. If you were forwarded this email, join some of the sharpest AI operators by subscribing here:

Hi all,

I’m excited to share that I’ve published a comprehensive guide to building organizational context systems for teams.

A context system is basically knowledge management, but for AI. As anyone who builds or uses AI tools at work knows, context is the most important driver of AI performance, far more than which LLM you use.

The entire piece lives on my site, but I also wanted to share a preview here. So without further ado:

This is a playbook for building team AI context systems, written for operators who want to make AI perform better for their organizations.

Just about everything that we developed and deployed ourselves internally and with clients in a dozen different industries is in here.

I wrote this guide because I talk to dozens of strategic operators every week who need to do more with less using AI, but are still only seeing limited benefits. They’re either stuck on the starting line of adoption, or are drowning under slop. If you’re an operator who has deployed AI tools like Claude, ChatGPT, Gemini, or even OpenClaw/Hermes across your teams, and any of the following issues resonate, then you’re in the right place.

You’re afraid to end AI chats because you don’t want to re-explain everything

Your AI confidently cites a fact that hasn’t been true in months

You don’t trust your AI to work autonomously on important projects

Your gut instinct might be to blame this on the LLM, but even “dumb” models today are really good at complex operational tasks. When they fail it’s usually because of the environment they’re working in - important data isn’t accessible, quality standards are unclear, or important knowledge sources are not organized and up to date.

In other words, most AI problems are really context problems.

This playbook will teach you how to build and operate a durable AI context system that will take your team’s AI from disorganized intern to core contributor. We start from the foundations of how AI uses context, and then explore which systems should hold what, how the knowledge will be used, who owns it, and what good looks like. You’ll walk away with complete roadmap to providing your AI with the knowledge and data it needs to own real business processes.

Ultimately, we are going to equip your AI with:

I’m sure you’ve seen AI fail in many different ways - it forgets things, creates generic slop, doesn’t know your processes... The list goes on. When AI performs badly, it’s usually because the information that your AI uses to understand its project is out of date, incorrect or just plain missing.

What about giving AI all the data?

If your team is small enough, you might be tempted to feed your AI with all the data you have. (Surely that will fix the problem!)

Unfortunately it won’t because agents have finite context windows. They start to forget things after a few hundred thousand words. And even if your AI could handle all that data, it would be prohibitively expensive and potentially expose sensitive information.

You can get around these issues sometimes with better prompts but that’s not efficient for an entire team. The essay-length mega prompt died for all practical purposes in 2024. Don’t think you have to hire an expensive developer to build custom tooling, either. The best way to make your AI perform well is with a context system that anyone can put together without code.

An organizational context system provides AI with the foundation of knowledge that it needs to work in a dynamic team environment. This enables AI to produce better quality outputs at lower cost and fewer mistakes. It achieves this by ensuring that every AI in use is pulling from the same source of truth, following the correct processes when taking actions.

Working with AI inside a whole team presents some challenging problems.

I’m going to address all of these issues in this playbook. But before diving into the deep end, let’s start with the basics: what is AI context, anyway?

Context is what an AI knows about your business and the task it’s working on. It’s specific to you, rather than the generic knowledge it learned when it was trained in the lab.

A unit of context could be a step in your onboarding process, a definition in your wiki, a SKU price, or the fact that a customer subscribed in March.

Context isn’t all one thing. If you look at the context your AI uses, you can break it all down into a few categories. Every context system should have information that falls into five types: declarative, procedural, historical, relational, evaluative. Each is useful in different situations.

Any real piece of work draws on several types of context at once. To write a single outbound email, for example, your AI needs to know who the company is, how you prospect, what you’ve already sent them, who you know there, and what a good email looks like.

If you are only building a personal context system, you could stop here. But for organizations there’s another important dimension to consider for context: scope.

Scope describes the level of operation at which a particular piece of context is owned. There are three levels of scope that need to be taken into account: individual, team and organization.

Keeping context organized by scope makes it easy for your AI to search through and maintain. When team project files get mixed together with company strategy docs and meeting notes, AIs start to get really confused and will make mistakes.

Individual

Individual scope is context for your own productivity, the things you wouldn’t expect to share with teammates.

Team

Team scope is the functional context for operating one piece of the business.

Organization

Organizational scope is company-wide context that everyone and every AI should be working from.

The table below illustrates how common AI context can be broken down by type and scope.

In reality, your context probably can’t be broken down so cleanly, and that’s totally fine. Think of these types and scopes as a framework to assist with organization rather than ironclad rules. In practice there are many ways to organize this context for your org, and how you should do it depends on where your team’s information lives today.

There’s too much to fit in an email newsletter, so you’ll have to go to my website to get parts 3-5!

That’s all for today. Thanks for reading!

江湖再见🫡

-Sawyer