The Human Context Window

Three months ago, I built an AI agent to run the intelligence layer of my company. I called it Clawdbass. It reads our CRM, our HRIS, our ATS, our financials. It builds dashboards, monitors security, and briefs me every morning before I open my laptop.

Early on, it was useless.

Not because the technology was bad. The model was one of the most powerful in the world. But it had no context. It didn't know our business model. It didn't know our org structure, our language, our history. It didn't know that when we say "capacity" we mean something specific. It didn't know which VP owns which region, which clients are expanding, or which metrics actually matter to me versus which ones just look good on a slide.

It had intelligence. But intelligence without context is just noise.

So I started feeding it. Memory files. Identity documents. Decision logs. Lessons learned from incidents. Client context. Team structure. Strategic priorities. Gradually, over weeks, I built what AI engineers call a "context window". The total body of knowledge the agent can access and reason over at any given moment.

The deeper the context got, the more useful the agent became. Not linearly. Exponentially. At some point it stopped being a tool I had to manage and started becoming something that could genuinely anticipate what I needed.

And that's when it hit me: we've completely forgotten how true this is for humans too.

What a Context Window Actually Is

In AI, a context window is the information the model can "see" when making a decision. Think of it as working memory plus reference material. It includes who the agent is working for, what happened recently, what's been decided before, what's in flight, what the priorities are, and what the constraints look like.

Without this, even the most capable model gives generic, surface-level answers. With it, the same model becomes genuinely useful. Sometimes brilliant.

The parallel to people in organisations is almost exact.

Every employee operates with a context window. It's the sum total of what they know about the business, the team, the strategy, the customer, the history. Some people have rich, deep context windows. They understand not just what we're doing, but why. They know the backstory. They know where the landmines are. They can pattern-match across situations because they've seen enough to connect dots that others can't.

These are your highest-performing people. And it's not because they're smarter. It's because they have better context.

Why Context Compounds

When I watch my agent operate with full context, something interesting happens. It doesn't just answer questions better. It starts to anticipate them. It sees that revenue is up but concentrated in three clients and flags the risk before I ask. It notices a hiring velocity problem two weeks before it shows up in the dashboard. It connects a client NPS drop to a specific team change because it has the history to make that link.

This is exactly what your best employees do.

The VP who's been with you for five years doesn't just know the current quarter's numbers. They know that the last time numbers looked like this, it was because of a specific market condition, and they know what we did about it. They don't need a briefing document before every meeting because the briefing is already loaded. It's their context window.

The new hire, no matter how talented, is starting with an empty window. They have raw capability but no context. And just like my agent in week one, they'll produce generic, surface-level work until that window fills up.

This isn't a criticism of new hires. It's a structural reality that most companies completely ignore.

How the Best Leaders Build Context

Working with my agent taught me something specific about how context gets built. It's not passive. You don't build a rich context window by simply existing in an organisation for a long time. You build it through deliberate systems.

Here's what I did for my agent:

Identity documents. Who it is, how it should operate, what its defaults are. Without this, it has no frame of reference for any decision.

Memory systems. Every important interaction gets logged. Decisions, outcomes, lessons, preferences. Not everything. The important things. The things that should inform future decisions.

Active project awareness. It knows what's in flight, who's responsible, what the blockers are. Not because I told it this morning, but because the system maintains this continuously.

Relationship context. It knows who the key stakeholders are, what they care about, how they communicate. This means it can tailor its approach rather than treating every interaction the same way.

Post-mortems. When something goes wrong, we document what happened, why, and what we learned. This becomes permanent context that prevents the same mistake twice.

Now think about how most employees build context in your organisation:

They don't.

They sit in meetings and absorb what they can. They read Slack messages and try to piece together what matters. They ask colleagues who may or may not have accurate information. They learn through trial and error. Often repeating mistakes that the organisation has already made and already learned from, because nobody wrote it down.

The best leaders I've worked with, the ones who seem to have an almost supernatural ability to make the right call, aren't operating on instinct. They're operating on an incredibly deep, well-maintained context window. They've been deliberate about building it.

What Companies Get Wrong

Most organisations invest heavily in capability. Skills training, certifications, tooling, processes. These are the equivalent of giving someone a more powerful model.

Almost nobody invests in context.

Think about onboarding. A new senior hire joins your company. They're brilliant. They have all the skills. You give them a laptop, access to the systems, and maybe a week of introductions. Then you expect them to perform.

You've just deployed a state-of-the-art model with an empty context window and wondered why the output isn't great.

The companies that will win the next decade are the ones that get serious about context engineering. For humans, not just for AI. That means:

Making institutional knowledge accessible, not tribal. If something important happened and the lesson lives only in one person's head, your organisation has a single point of failure. Write it down. Make it findable.

Designing onboarding around context, not compliance. Stop making new hires sit through policy reviews for a week. Instead, immerse them in the why behind the business. The strategic context, the competitive landscape, the history of key decisions and their outcomes. Give them the context window they need to be effective, not a list of rules.

Creating systems for continuous context sharing. The stand-up, the weekly update, the all-hands. These aren't just communication rituals. They're context distribution mechanisms. If they're not actively enriching your team's context window, they're a waste of time.

Protecting context when people leave. When a senior leader departs, you don't just lose a person. You lose a context window that took years to build. Most companies do nothing about this. The smart ones treat it like a knowledge transfer, not just a handover.

The Coming Divide

Here's where this gets strategic.

AI is getting better at capability every month. The raw processing power, the analysis, the synthesis. That's a solved problem, and it's getting cheaper by the day. My article on understanding your relationship with work argued that output alone won't differentiate you. Context is the extension of that argument.

The employees who will be irreplaceable are the ones with context windows that can't be replicated. The person who's been in the room for a hundred client conversations and can read the subtext. The leader who knows that the last three times we tried this approach, it failed for a specific reason that isn't in any document. The operator who understands not just what the data says, but what it means given everything else they know about the business.

You can give AI more data. You can give it more memory. But you can't give it the lived experience of navigating a specific organisation, with specific people, through specific challenges, over years.

That is uniquely human context. And it is the most valuable asset in your company that doesn't appear on any balance sheet.

The Mirror

Building an AI agent has been the most clarifying exercise in understanding what makes humans effective. When you're forced to make context explicit, to write down everything the agent needs to know to be useful, you realise how much of human effectiveness comes from context that's never been written down.

The uncomfortable question this raises: if you had to build a context window for your replacement tomorrow, could you? And if you can't articulate your context, how much of your value is truly in your head versus in the systems around you?

The best leaders I know could answer that question. They've been building and maintaining their context window deliberately for years. They just never called it that.

Now there's a word for it. And it might be the most important concept in leadership for the next decade.