Context Engineering: when common sense gets a fancy name

Lately I keep seeing the term "context engineering" everywhere. LinkedIn posts, AI newsletters, conference slides. Supposedly the hot upgrade from prompt engineering.

"My reaction? It's just common sense dressed up with a fancy name."

If you have used AI seriously in your business, you already know the deal: the quality of what you feed into the system determines the quality of what you get out. When I set up AI to handle student inquiries or birthday bookings at FU Café, I do not just throw prompts at it. I load it with the right information: course availability, pricing differences, enrollment deadlines, visa rules. In other words — context.

What It Really Means

Andrej Karpathy @karpathy

Karpathy explained it neatly: the LLM is the CPU, the context window is the RAM. The skill is deciding what to load into memory at each step.

Follow on X →
LLM
= the CPU
=
Context Window
= the RAM

Fair enough as an analogy. But is that really news? Every LLM already has a "projects" feature or equivalent — a way to bundle relevant context so you do not keep pasting the same stuff in. That is not rocket science, it is just good workflow.

Why It Does Matter

Context does matter. Think of an AI-driven marketing campaign system. It does not just produce ad copy. It pulls in audience demographics, recent buying behaviour, seasonal trends, campaign budgets, and past performance data. Without that context, the output would be generic and ineffective.

The principle is simple: give the system the right information, keep it current, and do not overload it with junk. Business owners already get this instinctively. They have been applying it without knowing it had a name.

The So-Called Strategies

The big "discoveries" of context engineering usually boil down to four things. Which are just... organisation:

Write
Keep important info in structured notes or databases — or just use the built-in project features already available.
Select
Only bring in what is relevant to the task at hand. Nothing extra.
Compress
Summarise long material. Do not paste entire documents when a summary does the job.
Isolate
Do not mix unrelated tasks. Keep conversations focused.

These are the same principles you would apply when training a new employee. Nothing revolutionary.

What Actually Matters

So do not worry about learning a new buzzword. Worry about these four questions instead:

The real craft

What information does your AI actually need to do its job well?

How do you keep that information up to date as things change?

Can you automate context preparation so it happens without manual effort?

Are you wasting space with irrelevant details that dilute the output?

Yes, context management becomes more important as AI gets more complex — with agents, long conversations, and multi-tool workflows. But calling it "context engineering" makes it sound scarier than it is.

You have probably been doing this already. Without ever calling it engineering.