1 min lesson
What an enterprise agent needs
Put yourself in this case. "A platform lead asks why a Cursor license still requires implementation work. Answer using knowledge, context and capability." Give the clearest next step.
Step 1 of 2
A model is only one part of an enterprise agent.
The same model can behave very differently depending on its instructions, the systems it can reach and the actions it can take. An FDE configures three planes so a team gets consistent results without unnecessary access: Knowledge, Context and Capability.
When an agent produces a poor result, work out which plane failed, because each one has a different fix.
- Plane
- Knowledge: the team's standards and instructions
- Mechanism
- Project RulesVersion-controlled instructions in the repo that every Cursor agent interaction inherits, so standards are encoded once. Press Enter for the full definition. and AGENTS.md
- Failure if you skip it
- The agent misses team conventions and produces inconsistent changes
- Plane
- Context: the information the agent can read inside and outside the repository
- Mechanism
- MCPModel Context Protocol. A standard that lets an AI agent pull in context from outside the repo, like Jira tickets or internal docs. Press Enter for the full definition. servers and admin controls
- Failure if you skip it
- The agent guesses about external systems or reads from an unapproved source
- Plane
- Capability: the actions and systems the agent can use
- Mechanism
- Run Modes, hooks, sandboxing and VM isolation
- Failure if you skip it
- The agent can run an unsafe action or send data where it should not
| Plane | Mechanism | Failure if you skip it |
|---|---|---|
| Knowledge: the team's standards and instructions | Project RulesVersion-controlled instructions in the repo that every Cursor agent interaction inherits, so standards are encoded once. Press Enter for the full definition. and AGENTS.md | The agent misses team conventions and produces inconsistent changes |
| Context: the information the agent can read inside and outside the repository | MCPModel Context Protocol. A standard that lets an AI agent pull in context from outside the repo, like Jira tickets or internal docs. Press Enter for the full definition. servers and admin controls | The agent guesses about external systems or reads from an unapproved source |
| Capability: the actions and systems the agent can use | Run Modes, hooks, sandboxing and VM isolation | The agent can run an unsafe action or send data where it should not |
Review knowledge, context and capability before increasing agent access.
We do not rely on the model alone. We define its instructions, limit the context it can reach and restrict the actions it can take. We add autonomy only after the controls work in practice.