Grounding
What it is for
Grounding helps to:
- reduce hallucinations;
- avoid generic responses;
- work with up-to-date information;
- maintain consistency with team decisions;
- cite sources;
- verify claims;
- produce outputs aligned with specs or internal criteria;
- limit the model's freedom when it must not make things up.
Examples
| Without grounding | With grounding |
|---|---|
| "Propose acceptance criteria for this feature." | "Propose acceptance criteria based only on this story, this spec, and this Definition of Done." |
| "Summarize the status of the product." | "Summarize the status of the product using these tickets, this roadmap, and the decisions from the last review." |
| "Review this code." | "Review this diff against the approved spec, the repository conventions, and the existing tests." |
Grounding Does Not Guarantee Truth
Grounding improves quality, but it does not make AI infallible. The model may misinterpret a source, omit nuances, miscite, or combine information incorrectly.
That is why grounding must be accompanied by verification: human review, traceability, tests, citations, cross-checking, or automated sensors.
Relationship with RAG and MCP
Grounding can be implemented in different ways:
- manually pasting the context into the prompt;
- using retrieval-augmented generation (RAG);
- connecting the AI to documents or databases;
- using search tools;
- integrating systems through Model Context Protocol;
- working with repositories, specs, or knowledge bases.
MCP is not grounding in itself. It is a technical way of connecting AI applications to external tools and sources. Grounding happens when that connection is used to anchor a response to relevant, verifiable information.
AI and Scrum
In AI-assisted Scrum, grounding is especially useful for:
- story refinement;
- writing specs;
- reviewing the Definition of Ready for AI;
- planning tasks with agents;
- reviewing generated code;
- analyzing user feedback;
- preparing Sprint Reviews;
- updating documentation.
The team must know which sources are valid. Not every document is a source of truth.
Common mistake
Believing that providing a source removes the need for review. An AI can work with correct documents and still misinterpret them, oversummarize, or mix in assumptions. Grounding reduces the risk, but it does not replace verification.
Resources
🏦 IA aplicada al trabajo ágil [AI Applied to Agile Work, in Spanish]Skill 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 [Study Guide: Scrum in Teams with AI, in Spanish]Resources · Scrum Manager
References
- Model Context Protocol. (2025). “Specification,” modelcontextprotocol.io.
- NIST. (2023). Artificial Intelligence Risk Management Framework (AI RMF 1.0). National Institute of Standards and Technology.
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.