Context engineering: questions and answers

What a model gets to see, and why that matters more than the prompt.

  • What is a context graph?

    In Foundation Capital's definition, a context graph is a structure of decision traces linked across entities and time. A decision trace records what happened in one concrete case: the inputs from several systems, the policy applied, the exception granted and who approved it. Linked together, those traces make precedent searchable, so an AI agent sees not just the current state but why it is that way.

  • What is agentic context engineering (ACE)?

    Agentic Context Engineering (ACE) is a method from a 2025 paper by researchers including Qizheng Zhang, James Zou and Kunle Olukotun. It improves an AI agent by letting its context evolve into a playbook through three roles, generation, reflection and curation, with small incremental updates instead of rewriting. The abstract reports gains of about 10.6% on agent benchmarks and 8.6% on finance tasks, without changing model weights.

  • What is the difference between context engineering and prompt engineering?

    Prompt engineering is writing and organizing the instructions you give a model. Context engineering is curating everything the model sees at inference time, including those instructions but also tools, retrieved data, message history and memory, and keeping that set right over many turns. Prompt engineering is a part of context engineering, not its rival.

  • What is an LLM context window and how big is it?

    An LLM's context window is the input and output token budget for a request, a form of working memory distinct from training data. Current Claude models such as Opus 5.5, Sonnet 5.5 and Haiku 5.5 have a 1M-token window, per November 2026; some older models have 200k. System prompt, messages, tool definitions, tool results, files, images and the output all count toward it.

  • What is the context engineering skill for Claude, and how do you install it?

    "The context engineering skill" usually means Agent Skills for Context Engineering, an open-source GitHub repository by Muratcan Koylan under the MIT license. It is not one skill but a collection, 18 per November 2026, each a SKILL.md folder on topics like context degradation, compression, memory systems and multi-agent patterns. In Claude Code you install it as a plugin through its marketplace; Cursor, Codex and other hosts can load the same folders.

  • What does Anthropic say about context engineering for AI agents?

    Anthropic's engineering post "Effective context engineering for AI agents" (September 29, 2025) treats context as a finite resource with an attention budget. Its core advice: find the smallest set of high-signal tokens that makes the desired outcome likely. In practice that means system prompts at the right altitude, few and clear tools, a handful of canonical examples, just-in-time retrieval, and for long tasks compaction, structured notes and sub-agents.

  • What is context compaction in AI agents?

    Context compaction is replacing the older turns of a conversation with a summary, so a long agent task can continue inside the context window. The Claude API offers server-side summarization; context editing separately clears selected tool results or thinking blocks. Compaction keeps the session going, but whatever the summary leaves out is gone, so knowledge that must survive belongs in a file or memory outside the window.

  • How does LangChain approach context engineering?

    LangChain's July 2025 post "Context Engineering" sorts the work into four strategies: write context outside the window (scratchpads, memories), select it back in (retrieval of memories, tools and knowledge), compress it (summarization, trimming) and isolate it (sub-agents, sandboxes, hidden state fields). LangGraph and the LangChain v1 docs map those strategies onto state, a long-term store, summarization and middleware.

  • What did Manus learn about context engineering for AI agents?

    In July 2025 Yichao "Peak" Ji of Manus published six lessons from building the Manus agent: design around the KV-cache, mask tools instead of removing them, use the file system as context, recite the plan in a todo.md, keep failed actions in the context, and vary the context so the agent does not copy its own patterns. The thread through all six: stable, append-only context that the agent can extend with files.

  • What is harness engineering, and how does it differ from prompt and context engineering?

    Harness engineering is designing the environment, feedback loops and controls that keep an AI agent on track, not just what it reads. The term took off in February 2026 with an OpenAI post on building a product with Codex without hand-written code. Prompt engineering words the instruction, context engineering decides everything the model sees, and harness engineering adds the checks and constraints that catch and correct what the agent does.

  • How do you do context engineering for multi-agent systems?

    Context engineering for multi-agent systems is deciding what each agent sees and what it hands back. The working pattern from Anthropic, Cognition and Google: give each sub-agent a scoped brief with objective, output format and boundaries, let it work in its own clean window, return a short result or a reference to a stored artifact, and keep decisions that must agree with each other in one agent.

  • How do you set up context engineering for a team using AI tools?

    Set it up in layers. Organization-wide rules in one managed file, project rules in a short file in the repository, personal preferences in a personal file, and team knowledge in a shared place every tool can search. Keep instruction files short and consistent, assign an owner to each layer, and save approved corrections where colleagues' AI tools can retrieve them.

  • What is Context Engineering 2.0?

    "Context Engineering 2.0: The Context of Context Engineering" is an October 2025 arXiv paper by Qishuo Hua, Pengfei Liu and seven co-authors. It argues context engineering is over twenty years old and splits it into eras: 1.0 (1990s to 2020, sensors and rules), 2.0 (from 2020, LLM agents that read natural language), and a future 3.0 and 4.0. Its core idea: context engineering is reducing entropy.

  • What is Google's context engineering whitepaper on sessions and memory?

    "Context Engineering: Sessions, Memory" is a 72-page Google whitepaper from November 2025 by Kimberly Milam and Antonio Gulli, published on Kaggle for the 5-Day AI Agents Intensive. It defines context engineering as dynamically assembling and managing what goes into an LLM's context window, and splits the work into sessions (one conversation) and memory (what persists across conversations).

  • What is advanced context engineering for coding agents?

    "Advanced Context Engineering for Coding Agents" is an August 2025 essay by Dex Horthy of HumanLayer, based on a talk at Y Combinator. Its method, frequent intentional compaction, builds the whole workflow around context: work in three phases (research, plan, implement), write each phase to a markdown file, start fresh windows often, keep context use around 40 to 60 percent, and spend human review on the research and the plan.

  • RAG vs context engineering: is RAG still needed?

    RAG (retrieval-augmented generation) fetches passages from a knowledge source and puts them in the prompt. Context engineering is the wider job of deciding everything the model sees on each step: instructions, tools, examples, memory, history and retrieved documents. RAG is one technique inside it. What changed is how retrieval is used: agents increasingly search when they need something, instead of getting one batch of chunks before they start.

  • What is context rot and how do you avoid it?

    Context rot is the decline in an AI model's accuracy and recall as the amount of text in its context window grows. Chroma's 2025 study of 18 models found performance gets less reliable with longer input, even on simple tasks. You avoid it by keeping the context small and relevant: retrieve only what the task needs, clear old tool output, compact long sessions and start fresh when the task changes.

All topics →

The first half hour · free

What does your AI forget?

  1. 01You tell us which AI you use and where it keeps starting over.
  2. 02We show what a shared memory changes there, on your own examples.
  3. 03You keep the plan, with or without Bombos.
Piet Baudoin
Piet BaudoinYou talk to Piet, from the Bombos teamDutch team from Utrecht and Nijmegen · data in Europe
We reply the same working day.