Context engineering
Deciding what an agent should be able to see. Increasingly the real skill: a model with the approved spec, the review notes and the relevant code will beat a better model working from a paragraph.
Prompt engineering was about phrasing. Context engineering is about supply: which documents, which files, which past decisions land in the window, and in what order.
The counter-intuitive part is that more is not better. A window stuffed with everything tangentially relevant produces worse output than a window holding the approved spec, the two files that matter, and the open review notes. Retrieval is a curation problem, and curation means leaving things out.
This is where an approved spec earns its keep for a solo developer with no team to coordinate with. It is the highest-signal context that exists: what we decided, why, and which lines implement it — written down once, reusable in every session afterwards, and immune to the model forgetting.
- MCP
An open protocol that lets an AI agent read and write real systems — files, APIs, trackers — instead of being told the state of the world in every prompt.
- Context window
How much text a model can consider at once. Treat it as a budget you spend deliberately, not a bucket to fill — a window packed with marginal material produces worse answers than a curated one.
- Vibe coding
Building by describing what you want and accepting what comes back without reading it closely. Genuinely fine for a prototype. The failure mode is shipping it, then discovering nobody — human or model — knows why the code does what it does.