First, let’s be clear: what follows is our synthesis of a trend, not the recital of a settled verdict. But we believe that in working with language models, the centre of gravity is shifting. There was a time when the main art came down to writing a good prompt; today, in our view, the larger problem is designing the whole context the model works in.
From an instruction to an environment
Prompt engineering, in its classic form, is about one thing: how to write a good instruction so the model does the right thing. This still matters. But as systems grow more complex, it becomes clear that the model’s output depends not just on the instruction, but on everything that sits in its context window: what information was retrieved, what memory is available, what tools were called, and what state remains from the past. This whole environment is the “context.”
The principle of least knowledge
One principle that shapes this view is what you might call least knowledge: an agent should know only what it needs for its job — no more. This principle, which we believe in, turns context engineering from a “the more the better” job into a careful design job. The question is no longer “what can I give the model,” but “what should I actually give it, and what should I keep out.”
Why this shift matters
Why does this change of view matter? Because the behaviour of a complex system can’t be controlled by working a single prompt alone. If the wrong information is in the context, even the best instruction gives a bad result. If the context is crowded with irrelevant things, the model’s focus drops. And if the boundaries aren’t clear, the system becomes fragile and even unsafe. For this reason, designing the context becomes as important as writing the instruction — and sometimes more.
The instruction still matters
This doesn’t mean abandoning prompt engineering. A clear instruction is still the backbone of the work. The point is that the instruction is only one part of a larger picture. Context engineering encompasses prompt engineering rather than replacing it: writing a good instruction, alongside the deliberate management of everything else that sits in the context.
Putting it together
We believe that as intelligent systems grow more complex, the key skill moves from “writing a good prompt” to “designing a good context.” This is a synthesis, not a prophecy; but it’s a trend we believe in and apply in our own design. The model reflects what it sees — and our job is to decide carefully what it sees.