Context Pointer
also called Reference
A cheap reference to something the agent can fetch if needed.
A short mention of where information lives — a file path, a document ID, a URL — placed in context instead of the content itself. The agent dereferences it only when the task requires it.
Pointers trade a little latency for a much smaller, sharper context. They also stay valid when the underlying content changes, which copies do not.
A pointer is a bet that the agent will fetch what it needs and will do so competently. That bet is good when the environment has cheap, reliable retrieval — a filesystem with grep, a search tool — and bad when the information is expensive to find or easy to misread.
Paths instead of file contents
An agent needs to work across a large repository.
Give it the tree and the conventions, and let it read the files it decides matter. Pasting the contents of a fifty-file module costs a fortune and buries the relevant three paragraphs, which is why the working style converges on grepping first.
- blogEffective context engineering for AI agents ↗
Defines context engineering as curating everything supplied at inference, names context rot, and covers just-in-time retrieval, compaction and keeping the window high-signal.
- blogCode execution with MCP: building more efficient AI agents ↗
Presenting tools as a code API so the agent loads only what it needs and filters results in the sandbox instead of pushing large intermediate data through the context window.
Progressive Disclosure
Reveal detail only when it becomes relevant.
Context Engineering
Deciding what goes in the window, in what order, at all.
Context
Everything the model can currently see.
RAG · Retrieval-Augmented Generation
Look things up, then put what you found in the prompt.