Context – and is "context engineering" the new black?

Last time, the philosopher put down his pen with a map in his lap and one word left hanging in the air: context. Time to sort it out, once and for all – before he finds a new map to draw and forgets about us entirely.

The map doesn't tell you where you're standing

An ontology gives you the categories: what things exist, how they relate. Semantics gives you the meaning of the words you use. But neither tells you anything about where in the landscape you actually are right now – which situation, which conversation, which task.

That's context's job. And that's why the same question to an AI can produce entirely different, both reasonable, answers depending on what was said just before, which documents are open, what role the AI was given, and what goal you're in the middle of. The words are the same. The map is the same. But your position on the map has moved.

Is "context engineering" the new black?

A couple of years ago everyone talked about "prompt engineering" – the art of writing the perfect instruction. These days the conversation has shifted a notch: it's not enough to write a good sentence, you also have to give the model the right context to interpret that sentence in. Hence "context engineering."

Is it just a new buzzword for the same old thing? Partly. But it also points at something genuinely true: the better AI systems get at language, the less the bottleneck is "does it understand what I'm saying" and the more it becomes "does it have what it needs to answer correctly." The philosopher would probably say this was always true – that a well-phrased question without context is just a well-phrased guess.

So: new name, old truth, but a truth that's become more practically urgent as we lean ever more heavily on AI for real decisions.

What "context" actually consists of

Context isn't one single thing, but a whole cast of characters that together shape how an answer takes form:

The instruction itself – what you actually wrote.
The history – what's been said earlier in the same conversation.
The material – documents, data, search results the model has access to.
The role – which "hat" the model has been asked to wear (expert, editor, philosopher with a quill pen).
The tools – what the model can actually do, not just say.
The goal – what the whole interaction is actually meant to achieve.

Miss a single one of these and the result can look entirely correct while still being wrong for your specific situation – the same way an AI agent without a shared ontology could update the wrong "address" in the previous article. The difference is that the ontology problem was about not knowing what things were. The context problem is about not knowing where in the story you are.

The question is the answer

Here's the twist the philosopher has been waiting for this whole article. Because if context is what determines a correct answer, the next question becomes: how do you know which context is actually needed?

Hal Gregersen spent years studying what makes breakthrough ideas emerge – among entrepreneurs, researchers, leaders. The conclusion in Questions Are the Answer is as simple as it is uncomfortable: what separates a breakthrough from a standstill is rarely a better answer. It's a better question. Gregersen shows how people stuck on a problem often solve it, not by working harder on the answer, but by simply reframing the question – changing the angle, the assumption, what they're even asking for.

That's the exact same principle that makes context engineering hard. Building the right context for an AI requires you to already know what you're actually trying to solve – and that insight rarely arrives directly. It arrives by trying the wrong question, feeling the answer doesn't quite fit, and asking again, until the question itself becomes sharp enough that the context falls into place on its own.

The philosopher with the quill pen, in other words, hasn't just been sitting around writing poetry this whole series. Without saying so directly, he's reframed the same question four times: what do we actually mean by the words we use? Each article has been a new answer to that question, approached from a different angle, in a different context.

The philosopher closes the book (for now)

So, the short version of four articles: language isn't just words. It's meaning (semantics), it's the structure behind meaning (ontology), and it's the context that determines which meaning actually applies right now (context). And behind all three sits the same old truth Gregersen reminds us of: whoever asks better questions rarely needs to worry about worse answers.

The philosopher leans back, sets down his pen for real this time, and nods toward a new, blank page on the desk. What gets written there next is still open – but the questions waiting there are about how data actually tells a story, how it becomes worth something, and what happens when we let agents loose with their own keys and a ticking clock.