Context Management
AI doesn't work well just because the model is powerful. It works better when it receives the right context.
Context management answers a simple question: what does the AI need to know to do this task well?
What Is Part of the Context
Context may include:
- product vision;
- users and segments;
- user stories;
- specs;
- acceptance criteria;
- architecture;
- API documentation;
- database schema;
- technical standards;
- previous decisions;
- good and bad examples;
- legal or security constraints;
- editorial tone;
- glossaries;
- boundaries of action.
Not all context should always be provided. Good context management means supplying what is necessary, not dumping all the documentation.
Difference from prompt engineering
| Practice | Central question |
|---|---|
| Prompt engineering | What instructions do we give? |
| Context management | What information does the AI need to see to carry out those instructions well? |
A well-written prompt can fail if the context is poor. And extensive context can go to waste if the instructions are ambiguous.
Why it matters in agile teams
In agile teams, many decisions depend on shared knowledge: how "done" is defined, what criteria the Product Owner uses, what technical constraints exist, what patterns are followed, and what was learned in discovery.
If the AI does not receive that context, it will tend to produce generic answers. They may sound useful, but they will not be aligned with the real product.
Risks of poor context management
- Plausible but irrelevant responses.
- Implicit requirements invented by the AI.
- Contradictions with previous decisions.
- Loss of traceability.
- Duplication of work.
- Unnecessary exposure of sensitive data.
- Overload from feeding in too much information.
- Context degradation in long conversations.
Best practices
- Define which documents are the source of truth.
- Use specs and verifiable criteria.
- Maintain context templates for each type of task.
- Include examples of acceptable outputs.
- Remove outdated information.
- Separate instructions from external data.
- Provide the minimum sufficient context.
- Periodically review project instructions, prompt libraries, and documentation used by agents.
AI and context management
Context management becomes critical with Spec-Driven Development (SDD), coding agents, and work on existing products. The more autonomous the agent, the more important it is that it has clear boundaries, sources, and criteria.
It's not about making AI "know everything," but about reducing the decisions it shouldn't have to make up.
Common mistake
Confusing more context with better context. Adding documents without judgement can worsen the result: it introduces noise, contradictions, and outdated material. Context management is about selecting relevant, up-to-date information for the specific task.
Resources
🏦 Harness EngineeringSkill Arena · Scrum Manager
🏦 Scrum en equipos con IA [Scrum in Teams with AI, in Spanish]Skill Arena · Scrum Manager
📊 Guía didáctica Scrum en equipos con IA [Teaching Guide: Scrum in Teams with AI, in Spanish]Resources · Scrum Manager
References
- Anthropic. (2024). "Introducing the Model Context Protocol", Anthropic.
- Model Context Protocol. (2025). "Specification", modelcontextprotocol.io.
See also
Want to advance in agility? You can search for course and exam calls or go at your own pace by becoming a member of the Agile Club. This membership includes exclusive resources, e-learning classrooms, and access to Skill Arena: a space to practice and measure your agile skills at your own pace.