---
title: "What is context engineering? A guide for teams using agents"
description: "Context engineering is deciding what information an AI model sees at each step. How it differs from prompt engineering and how teams use it for agent memory."
image: "https://bombos.ai/assets/og-en-v2.png"
---

# What is context engineering, and what does it mean for a team using AI agents?

Context engineering is deciding what information an AI model sees at each step. How it differs from prompt engineering and how teams use it for agent memory.

*Piet Baudoin, Bombos · published November 16, 2026 · November 16, 2026 · updated November 16, 2026 · November 16, 2026 · 8 min read*

**Does your AI run into this too?** Together we map what your AI forgets. Free, and the plan is yours, with or without Bombos.

[Book your half hour](https://bombos.ai/en/contact/)

**Context engineering is the work of deciding what information an AI model has in front of it at each step: the instructions, the tools, the retrieved knowledge, the memory and the conversation so far. Anthropic defines it as the “set of strategies for curating and maintaining the optimal set of tokens (information) during LLM inference”, and calls it “the natural progression of prompt engineering”. For one developer it is a design discipline. For a team using several AI tools, it becomes an organisational question: who maintains the context, and does every agent get the same one?**

By Bombos. Published November 16, 2026. Sources checked November 16, 2026.

## The short answer

-   **Definition:** curating “the smallest possible set of high-signal tokens” that lets a model do the task ([Anthropic, Effective context engineering for AI agents](https://www.anthropic.com/engineering/effective-context-engineering-for-ai-agents), September 29, 2025).
-   **Why it matters:** models have an “attention budget”; more context is not better, and quality degrades as the window fills, a degradation known as context rot that Anthropic cites.
-   **The main techniques:** “just in time” retrieval, compaction, structured note-taking (agentic memory) and sub-agents (same source).
-   **For teams:** the hard part is not the technique but keeping one maintained context that every agent and tool uses, instead of one per person per tool.

## How is context engineering different from prompt engineering?

Prompt engineering is about writing the instruction. Context engineering covers everything the model sees: the instructions too, plus tools, memory, retrieved knowledge and the conversation history, and how all of that changes over a long task.

| Aspect | Prompt engineering | Context engineering |
| --- | --- | --- |
| Unit | One instruction | The whole context window, every turn |
| Question | How do I phrase this? | What should the model see now, and what should it leave out? |
| Typical work | Wording, examples, format | Retrieval, memory, tool selection, summarising history, splitting work over agents |
| Time scale | One call | A session, or many sessions |
| Who owns it | The person writing | The system, and whoever maintains its knowledge |

Anthropic notes that agents running in a loop generate more and more data that could be relevant, and that this has to be refined cyclically. Tool results, retrieved documents and earlier turns have to be chosen on purpose.

## What does context engineering consist of?

Following Anthropic’s article, and the documentation of the tools that implement it:

1.  **System instructions “at the right altitude”.** Specific enough to guide behaviour, flexible enough to give the model strong heuristics. In Claude Code these live in CLAUDE.md files; in a Copilot Studio agent, in its instructions; in ChatGPT, in a project.
2.  **Tools that are clear and few.** Each tool definition costs attention. Overlapping tools make the model choose badly.
3.  **Retrieval, ideally just in time.** Instead of loading everything up front, the agent keeps references (file paths, links, queries) and loads what it needs when it needs it.
4.  **Compaction.** “Compaction is the practice of taking a conversation nearing the context window limit” and summarising it, so the task can continue with the essentials.
5.  **Structured note-taking, or agentic memory.** “Structured note-taking, or agentic memory, is a technique where the agent regularly writes notes” that persist outside the context window and are read back later.
6.  **Sub-agents.** “Sub-agent architectures provide another way around context limitations”: a sub-agent explores in its own window and returns a short result.

## Where does AI agent memory fit in?

Memory is the part of context engineering that outlives a session. Every major tool now has some form of it, and they share a pattern: personal memory stays per person, per tool; what is shared stays inside one project or agent of one tool.

| Tool | What persists | Shared with the team? |
| --- | --- | --- |
| ChatGPT Business, personal | Saved memories | No: “Not transferable to other users, even within the same Business workspace” ([OpenAI](https://help.openai.com/en/articles/9295112-memory-faq-business-version)) |
| ChatGPT Business, shared project | Project-only memory, files, chats | Yes, within the project: ChatGPT can draw on the project’s chats from all participants ([OpenAI, Projects](https://help.openai.com/en/articles/10169521-projects-in-chatgpt)) |
| Microsoft 365 Copilot | Memory in a hidden folder of the user’s mailbox (preview) | No: “A user’s memory profile is not shared with other users” ([MC1539004](https://mc.merill.net/message/MC1539004)) |
| Claude, personal | Account memory | No ([Claude Help Center](https://support.claude.com/en/articles/11817273-use-claude-s-chat-search-and-memory-to-build-on-previous-context)) |
| Claude, shared project | Project knowledge, instructions, project memory | Knowledge and instructions yes; chats private by default; whether project memory is per member is not stated (“Each project has its own memory”) |
| Claude Code | CLAUDE.md (shared via git), auto memory (machine-local) | Only the files you commit ([Claude Code docs](https://code.claude.com/docs/en/memory)) |
| Copilot Studio agent | Knowledge sources you configure | The sources yes; what the agent learns in conversation no ([Microsoft Learn](https://learn.microsoft.com/en-us/microsoft-copilot-studio/knowledge-copilot-studio)) |

For developers building their own agents, memory layers such as Mem0, Zep, Letta and Supermemory provide the storage and retrieval; we compare them in [Mem0 alternatives for team memory](https://bombos.ai/en/knowledge-base/mem0-alternatives-for-team-memory/). For Claude Code specifically, see [Claude Code memory for teams](https://bombos.ai/en/knowledge-base/claude-code-memory-for-teams/).

## What does context engineering look like for a team, not one agent?

For one developer, context engineering is a design choice inside one system. A team of ten using ChatGPT, Claude Code, Copilot and Cursor has a different problem: ten people each engineering their own context, in four tools, with no shared source.

What that looks like:

1.  **The same correction, many times.** One engineer teaches the agent a rule; the next person’s agent repeats the mistake.
2.  **Knowledge tied to a person.** The colleague who set up the good prompts and projects leaves, and the context leaves with them.
3.  **Context that drifts.** Four copies of the pricing rules in four tools, each updated at a different moment.
4.  **No way to check.** When an agent answers, nobody can see which piece of context it used.

Team-level context engineering means treating context as shared infrastructure:

-   **One source of truth** for knowledge and working rules, owned by the company, not by one tool vendor.
-   **Every agent reads it** through an open connection such as the Model Context Protocol (MCP), which, as of December 2025, the Linux Foundation said had “more than 10,000 published MCP servers” ([Linux Foundation, December 9, 2025](https://www.linuxfoundation.org/press/linux-foundation-announces-the-formation-of-the-agentic-ai-foundation)).
-   **Corrections become rules** that apply to everyone, with a review step.
-   **Retrieval is traceable**, so a wrong answer can be traced to the context that caused it.
-   **Permissions follow the source**, so an agent does not show a document to someone who may not see it.

How to decide what goes into that shared context is covered in [How do you give AI agents company context?](https://bombos.ai/en/knowledge-base/how-to-give-ai-agents-your-company-context/). For Microsoft teams, the Dutch article on [giving a Copilot Studio agent your team’s knowledge](https://bombos.ai/kennisbank/copilot-studio-kennis-van-je-team/) shows where the built-in options stop.

Read also: [How to keep AI-written messaging consistent across a GTM team](https://bombos.ai/en/knowledge-base/how-to-keep-ai-written-messaging-consistent-across-a-gtm-team/).

## What are common mistakes in context engineering?

1.  **Loading everything.** A 300-page handbook in the context tends to make answers worse, not better. Retrieve the relevant part.
2.  **Rules written as wishes.** “Be careful with discounts” is not a rule a model can apply. “Above ten percent discount, refer to the account manager” is.
3.  **Treating instructions as enforcement.** Anthropic is explicit for Claude Code: memory files are “context, not enforced configuration”. What must never happen needs a hook or a permission, not a sentence.
4.  **No owner.** Context that nobody maintains goes stale.
5.  **Measuring on vendor benchmarks.** Vendors usually publish their own memory benchmarks. Test with your own tasks.

## How does Bombos do this?

Bombos is context engineering for a whole team, as a product: the memory layer under the AI tools a team already uses, from Claude Code and Cursor to ChatGPT and Copilot.

-   **Shared team knowledge.** It reads what already exists: mail, documents, business systems and repositories.
-   **Learned rules.** A correction from one person becomes a rule about how the team works, which every AI applies before anyone asks.
-   **A context graph with retrieval**: similarity, keywords, proximity in the graph, and rules that apply only in a given context. No model of its own; every answer shows which pieces of context it rests on.
-   **Cloud or self-hosted**, on your own server with your own models.

Bombos is young; in the call we tell you which connections work with your tools today. One person can always start: [try the plugin](https://bombos.ai/en/waitlist/) for your own agents.

## Frequently asked questions

**Who coined the term context engineering?** The term spread in 2025 among people building agents; Anthropic’s engineering article of September 29, 2025 gave it a widely cited definition. We do not attribute it to one person.

**Is context engineering the same as RAG?** No. Retrieval-augmented generation is one technique inside context engineering. Context engineering also covers instructions, tools, memory, compaction and how work is split across agents.

**Is AI agent memory part of context engineering?** Yes. Memory decides what from earlier sessions comes back into the context window. It is the part that outlives a single conversation.

**Do small teams need context engineering?** Any team where more than one person uses AI for the same kind of work already does it, usually by accident and in several copies. Doing it on purpose saves the repeated corrections.

[Book your half hour](https://bombos.ai/en/contact/): together we map what your AI forgets. Free, and the plan is yours, with or without Bombos.

## Which sources did we use?

All sources opened and read on November 16, 2026.

1.  **Anthropic, Effective context engineering for AI agents** (September 29, 2025): [https://www.anthropic.com/engineering/effective-context-engineering-for-ai-agents](https://www.anthropic.com/engineering/effective-context-engineering-for-ai-agents)
2.  **OpenAI, Memory FAQ (Business version)**: [https://help.openai.com/en/articles/9295112-memory-faq-business-version](https://help.openai.com/en/articles/9295112-memory-faq-business-version)
3.  **Microsoft 365 Message Center, MC1539004** (November 5, 2026): [https://mc.merill.net/message/MC1539004](https://mc.merill.net/message/MC1539004)
4.  **Claude Help Center, chat search and memory**: [https://support.claude.com/en/articles/11817273-use-claude-s-chat-search-and-memory-to-build-on-previous-context](https://support.claude.com/en/articles/11817273-use-claude-s-chat-search-and-memory-to-build-on-previous-context)
5.  **Claude Code docs, How Claude remembers your project**: [https://code.claude.com/docs/en/memory](https://code.claude.com/docs/en/memory)
6.  **Microsoft Learn, Knowledge sources summary, Copilot Studio** (updated October 29, 2026): [https://learn.microsoft.com/en-us/microsoft-copilot-studio/knowledge-copilot-studio](https://learn.microsoft.com/en-us/microsoft-copilot-studio/knowledge-copilot-studio)
7.  **Linux Foundation, Agentic AI Foundation announcement** (December 9, 2025): [https://www.linuxfoundation.org/press/linux-foundation-announces-the-formation-of-the-agentic-ai-foundation](https://www.linuxfoundation.org/press/linux-foundation-announces-the-formation-of-the-agentic-ai-foundation)

## 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](https://bombos.ai/assets/team/piet-klein.jpg)

**Piet Baudoin**You talk to Piet, from the Bombos teamDutch team from Utrecht and Nijmegen · data in Europe

## More about memory and context

-   [Memory layer for AI agents: what it does, where it sits, and build or buy](https://bombos.ai/en/knowledge-base/memory-layer-for-ai-agents/)
-   [Best MCP memory server for a team: what to compare and which options exist](https://bombos.ai/en/knowledge-base/best-mcp-memory-server-for-a-team/)
-   [What is AI agent memory, and how does it work?](https://bombos.ai/en/knowledge-base/what-is-ai-agent-memory/)
-   [Claude Code memory for teams: how do you share what your agents learn?](https://bombos.ai/en/knowledge-base/claude-code-memory-for-teams/)
-   [How do you keep company context portable across ChatGPT, Claude and Copilot at a post-seed startup?](https://bombos.ai/en/knowledge-base/how-to-keep-company-context-portable-across-ai-tools/)

[All questions in the knowledge base →](https://bombos.ai/en/knowledge-base/)


---

```json
{"@context":"https://schema.org","@graph":[{"@type":"Organization","@id":"https://bombos.ai/#organization","name":"Bombos","url":"https://bombos.ai/","logo":"https://bombos.ai/assets/bombos-bee.png","description":"Bombos is the memory layer under every AI your team uses. One shared memory for Claude, ChatGPT, Copilot, Cursor and your own agents, that learns how you work.","sameAs":["https://www.linkedin.com/company/bombos-ai","https://www.youtube.com/@truebombos","https://www.instagram.com/truebombos/","https://bsky.app/profile/bombostrue.bsky.social","https://www.threads.com/@truebombos","https://mastodon.social/@Bombos","https://www.tiktok.com/@truebombos","https://t.me/truebombos","https://www.pinterest.com/truebombos/","https://github.com/bombos-ai","https://linktr.ee/bombos.ai","https://bombosai.substack.com","https://hashnode.com/@bombosai","https://hackernoon.com/u/bombos","https://dev.to/bombos"],"email":"info@bombos.ai","areaServed":["NL","Europe"],"knowsLanguage":["nl","en"],"contactPoint":{"@type":"ContactPoint","email":"info@bombos.ai","contactType":"sales","url":"https://bombos.ai/en/contact/","availableLanguage":["nl","en"]}},{"@type":"WebSite","@id":"https://bombos.ai/#website","name":"Bombos","url":"https://bombos.ai/","inLanguage":["nl-NL","en"],"publisher":{"@id":"https://bombos.ai/#organization"}},{"@type":"Person","@id":"https://bombos.ai/#piet","name":"Piet Baudoin","url":"https://bombos.ai/over-ons/","worksFor":{"@id":"https://bombos.ai/#organization"},"sameAs":["https://www.linkedin.com/in/piet-baudoin","https://x.com/marketeerpiet","https://dev.to/marketeerpiet","https://medium.com/@piet_7573"]},{"@type":"Article","@id":"https://bombos.ai/en/knowledge-base/what-is-context-engineering/#article","headline":"What is context engineering, and what does it mean for a team using AI agents?","description":"Context engineering is deciding what information an AI model sees at each step. How it differs from prompt engineering and how teams use it for agent memory.","datePublished":"2026-11-16T00:00:00.000Z","dateModified":"2026-11-16T00:00:00.000Z","author":{"@id":"https://bombos.ai/#piet"},"publisher":{"@id":"https://bombos.ai/#organization"},"image":"https://bombos.ai/assets/og-en-v2.png","mainEntityOfPage":"https://bombos.ai/en/knowledge-base/what-is-context-engineering/","isPartOf":{"@id":"https://bombos.ai/#website"},"inLanguage":"en"},{"@type":"BreadcrumbList","itemListElement":[{"@type":"ListItem","position":1,"name":"Bombos","item":"https://bombos.ai/en/"},{"@type":"ListItem","position":2,"name":"Knowledge base","item":"https://bombos.ai/en/knowledge-base/"},{"@type":"ListItem","position":3,"name":"Memory and context","item":"https://bombos.ai/en/knowledge-base/#memory"},{"@type":"ListItem","position":4,"name":"What is context engineering, and what does it mean for a team using AI agents?","item":"https://bombos.ai/en/knowledge-base/what-is-context-engineering/"}]},{"@type":"FAQPage","@id":"https://bombos.ai/en/knowledge-base/what-is-context-engineering/#faq","mainEntity":[{"@type":"Question","name":"How is context engineering different from prompt engineering?","acceptedAnswer":{"@type":"Answer","text":"Prompt engineering is about writing the instruction. Context engineering covers everything the model sees: the instructions too, plus tools, memory, retrieved knowledge and the conversation history, and how all of that changes over a long task."}},{"@type":"Question","name":"What does context engineering consist of?","acceptedAnswer":{"@type":"Answer","text":"Following Anthropic's article, and the documentation of the tools that implement it:"}},{"@type":"Question","name":"Where does AI agent memory fit in?","acceptedAnswer":{"@type":"Answer","text":"Memory is the part of context engineering that outlives a session. Every major tool now has some form of it, and they share a pattern: personal memory stays per person, per tool; what is shared stays inside one project or agent of one tool."}},{"@type":"Question","name":"What does context engineering look like for a team, not one agent?","acceptedAnswer":{"@type":"Answer","text":"For one developer, context engineering is a design choice inside one system. A team of ten using ChatGPT, Claude Code, Copilot and Cursor has a different problem: ten people each engineering their own context, in four tools, with no shared source."}}]}]}
```
