Table of contents
Open Table of contents
Everyone has the same model
Here’s an uncomfortable truth for anyone betting their competitive advantage on AI: you and your competitor are probably calling the same model. The frontier labs ship to everyone at once. The weights that power your engineering team are an API call away for the team across the street.
So if the model isn’t the differentiator, what is?
It’s context. The depth, relevance, and freshness of what you feed the model is what separates a generic chatbot answer from a response that feels like it came from your most senior engineer. The model is the engine. Context is the fuel — and the map, and the destination.
I’ve started calling the thing that holds this context the organisational brain: a living knowledge layer that captures your team’s decisions, patterns, and institutional memory, and gets injected into every AI session.
Why context compounds and models don’t
A better model is a step function — it lands, everyone gets it, the playing field re-levels. A better context layer is a flywheel. Every design review, every incident, every “we tried that in 2024 and here’s why it failed” makes the next session sharper. It compounds, and it’s yours.
This is the part competitors can’t copy with a bigger budget. They can buy the same tokens. They can’t buy the three years of decisions your team made, the dead ends you already explored, or the unwritten reasons your architecture looks the way it does.
The quality of an AI session is determined by the depth and relevance of the context it receives — not just the model powering it.
What goes into the brain
Context isn’t “paste the whole repo into the prompt.” That’s the lazy version, and it gets worse as you scale. Good context engineering is curation:
- Decisions, not just docs. The why behind a choice is worth more than the what. ADRs and post-mortems carry the reasoning a fresh model can’t reconstruct.
- Patterns, not just code. How your team names things, structures modules, handles errors. The conventions that make output feel native.
- Memory, not just retrieval. What was tried before and how it went. The brain should remember its own past sessions.
- Live signal, not just snapshots. Production data, real usage patterns, the operational context that tells the agent what actually matters right now.
The practical takeaway
If you’re investing in AI for your engineering org, the highest-leverage work is rarely “switch to the newest model.” It’s building the pipeline that captures your context and surfaces it at the right moment.
Spend your effort there. The model will keep getting better on its own — that’s the labs’ job. The brain only gets better if you build it. That’s yours.
This is the foundation everything else in my AI Engineering practice is built on. The model is a commodity. The brain is the moat.