This week's newsletter is sponsored by Sana Learn.
Today's piece is about context - specifically, what happens when learning has access to someone's actual situation rather than a generic version of it. Sana is the platform I keep coming back to when I think about what that looks like in practice. Their AI tutor doesn't just deliver content; it works with your organisation's real knowledge to meet each learner where they actually are. If that's a problem you're actively thinking about, their team is worth talking to.
I’ve been thinking quite deeply about Context recently. Specifically context for learning.
I tested an AI Tutor embedded in a learning platform and ran two “people” through it.
One person got what you’d expect: a solid learning experience based on the course content available on a particular topic in the platform.
The other got the AI tutor working through her actual KPIs — her real numbers, her specific project, the exercises became a live working session with the metrics she looks at every Monday morning.
The only difference was a couple of context files sitting behind it. A few hundred words describing who she was, what she was trying to achieve, and what her organisation was measuring.
I have been thinking about why that works, and what it means for how we’ve been building learning. I’m not delving into the tool I was using in this article. You can keep an eye on The Learning Stack Youtube for a walkthrough of what the process looked like. You can see below what some of the conversation looked like.
Adding personal context to Priyas journey
What we’ve always done
I made a comment on a LinkedIn post about Context Architecture and I was challenged to explain more deeply. I was in a campsite in France aat the time so took some time to gather my thoughts, then responded.
My scribblings about context while on holiday
This article is a deeper dive into those early thoughts.
Learning teams over the years have been exceptional at compression.
We take the messy, complicated reality of how something works. Strip it down to the transferable minimum. Keep it clear, keep it focused, get people to the point where they can do something with it. That was the right call, and in many ways it still is.
But clarity is a compromise. When you build for everyone, you make decisions on behalf of everyone. You pick the example that travels furthest. You remove the detail that only applies to some people. You smooth out the edges that would confuse a particular audience.
What you’re left with is something that works reasonably well for most people, most of the time — and alongside it, everything you removed. All the context, the specificity, the situational detail that would have made it land differently for different people. That stuff didn’t disappear, it just didn’t get built.
For most of learning’s history, that was a reasonable trade-off because you couldn’t do anything else with it anyway
What we left on the cutting room floor
A learning platform, or AI, can now work with a person’s specific situation in a way it never could before. It can take a principle and run it through someone’s actual numbers, their actual constraints, the thing they’re working on this week — but only if that information exists somewhere it can draw on.
If the only thing you’ve ever built is the compressed version, and for most of us that’s all we’ve built, the AI repackages what you gave it. It can’t reconstruct what you left on the floor. The context we stripped out for clarity, the examples we cut because they were too specific, the business reality we never captured because it wasn’t our job to capture it — that’s exactly what would make the difference, and most of it doesn’t exist yet.
Why architecture
I’ve been calling this context architecture, and the architecture part matters, so I want to explain it.
We shouldn’t be building more content or bigger courses or longer documentation. Architecture is what sits underneath — the decisions about structure that nobody sees but that determine what’s possible. When you architect a building, you’re not decorating rooms, you’re deciding what loads the walls can carry, where the utilities run, what the whole thing can actually do.
Context architecture works the same way. You’re deciding what information needs to exist, where it lives, and how it connects — so that when someone arrives with their specific situation, there’s something real to meet them with.
The foundation
Before we get to the context that surrounds the learning, there’s a layer underneath it that most of us have never built deliberately.
Call it the subject matter layer. It’s everything that existed in the knowledge before we compressed it.
Think about what happens when a course gets built. A subject matter expert sits down with a designer. Between them they make hundreds of small decisions about what stays and what goes. The edge cases that only apply in certain industries. The “it depends” answers that would confuse a general audience. The worked examples that were too specific to travel. The expert instinct that was too hard to articulate in a learning objective. All of it ends up on the cutting room floor because the job is to get people to the point — and the point, by definition, is the minimum version of the truth.
That compression made sense when everyone was getting the same course. If you’re building for the middle, the middle is what you build.
But if the AI is going to give Priya a different experience from everyone else — one grounded in her specific situation — it needs more than the minimum version. It needs the edge cases, because her situation might be one of them. It needs the “it depends” answers, because the thing it depends on might be exactly her context. It needs the examples that were too specific to include for everyone, because one of them might be precisely right for her.
The subject matter layer is the unexpurgated version of what you know. Not everything — there’s still a job of curation to do — but significantly more than what makes it into the course. It’s the layer that says: here is the fuller picture, held in reserve, available to be drawn on when someone’s specific situation calls for it.
Without it, the other layers can tell the AI everything about Priya, but the knowledge it’s working with is still the compressed version. You’ve personalised the delivery. You haven’t deepened the content.
The floors above it
Built on top of that foundation are the layers that connect the subject matter to the specific person standing in front of it.
The personal layer is who someone is — their role, their goals, what their manager expects of them, how they actually work. This is what turned a generic KPI exercise into a session using Priya’s real numbers. The AI didn’t invent that detail; it was there in the context file, waiting to be used.
The business layer is what the organisation is actually trying to achieve — the metrics that matter, the pressures that are real this quarter, the definition of success that never makes it onto a course outline but shapes every decision the people in that organisation are making.
The environment layer is the world the organisation operates in — the market shifts, the competitive pressures, the forces that make certain decisions urgent and others beside the point.
Most learning is built without any of these layers, not because we didn’t care, but because we never had the capability to use them. The foundation rarely gets built either, for the same reason. The capability existed to compress. It didn’t exist to preserve what compression removed.
It does now.
The question nobody owns yet
Context architecture isn’t just a technology problem — most of the technology exists. It’s also a design problem. Someone has to decide what context needs to exist, where it lives, and how it connects to the learning that sits on top of it.
The question I am asking myself now: if AI is going to meet people in their specific situation, who in your organisation is responsible for making sure that situation is actually known?
Right now, in most places, nobody has a good answer to that.