Glossary

Glossary

Context engineering

Context engineering is the work of deciding what information an AI model sees each time it runs: instructions, retrieved documents, tool results and memory. It shapes an AI feature's quality more than the wording of any single prompt.

Context engineering grew out of prompt engineering. Once AI features moved from one-off answers to agents that work through many steps, the hard problem stopped being how to phrase a request. It became what to put in front of the model, out of everything available, at each step. Anthropic describes it as curating the smallest set of high-signal information that gets the job done.

What goes into the context

  • Instructions. What the system is for, its rules and its tone.
  • Retrieved knowledge. Documents, records or code pulled in for this request, often through retrieval-augmented generation (RAG).
  • Tools. Descriptions of what the model can call, and the results it gets back.
  • Memory. What carries over from earlier in the conversation or from past sessions.
  • Examples. A few cases that show what good output looks like.

Context engineering vs. prompt engineering

Prompt engineering is about wording a request. Context engineering is about everything the model sees, of which the prompt is one part.

What the work involves

  • Retrieval design. How documents are split, ranked and kept current, so the right facts arrive and the wrong ones don’t.
  • Trimming. Summarizing or dropping old history so long tasks don’t drown the model in detail.
  • Tool design. A few clear, well-described tools instead of many overlapping ones.
  • Memory. Deciding what should persist across sessions, and what should be forgotten.
  • Evals. Tests that show whether a change to the context made answers better or worse.

LangChain’s overview of context engineering for agents covers these techniques in more depth, including the cost of long contexts and how to test changes.

Where it goes wrong

  • Stuffing everything in. More context is not better context. Irrelevant material distracts the model and raises the cost of every call.
  • Stale or wrong retrieval. The model answers confidently from an outdated document.
  • No evals. A change that fixes one case quietly breaks others.
  • Context in one person’s head. The engineer who knows why the retrieval works the way it does leaves, and nobody can safely change it. Keeping that engineer is an employee retention problem as much as a technical one.

How Remotely approaches it

Context engineering sits across several of the skill areas Remotely screens engineers for in a live interview: prompting and retrieval, evals, and the LLM stack. Our AI fluency entry explains how we look at them.

Hire an AI engineer who knows what the model should see, and what it shouldn’t.

Ready to elevate your software company with top-tier global talent?

Hire developers through Remotely Works, where transparency and value maximization are at the heart of every connection we facilitate. Start the journey to enriching your team with senior software development talent today.

Last updated

Find your dream hires with Remotely.