2 min lesson
The FDE technical layer: configuration is the product
Imagine this comes up at work: "A platform lead says 'we already pay for Cursor, why do we need an FDE engagement?' Frame the answer in one sentence using the three planes." Start with the practical move.
Step 1 of 2
At enterprise scale the model is a commodity.
The configuration is the differentiator. Anyone can install Cursor in an afternoon. The FDE earns their seat by turning a sharp tool into a governed, repeatable system: standards encoded once, context wired in safely, execution power clamped to least privilege.
Think in three planes that every agent interaction passes through:
- Knowledge is what the agent knows about your standards.
- Context is what the agent can see outside the repo.
- Capability is what the agent is allowed to do.
Get those three right and autonomy becomes a dial you can turn with confidence instead of a leap of faith.
An agent's behavior is the product of three configurable planes: knowledge, context, capability. The FDE's job is to make each one explicit, version-controlled and auditable. Then 'the agent did something dangerous' stops being a vibe. It becomes a config gap you can point to and close.
We don't 'trust the model.' We trust the controls around the model. Knowledge, context and capability are each configured, scoped and logged. Autonomy is something we earn with evidence, not something we hope for.
- Plane
- Knowledge
- 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. / AGENTS.md
- Failure if you skip it
- Agent reinvents your conventions every prompt; PRs drift from house style
- Plane
- Context
- 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 + allowlists
- Failure if you skip it
- Agent guesses about Jira and observability it can't see or pulls from an ungoverned source
- Plane
- Capability
- Mechanism
- Allow/blocklists, hooks, sandboxing, VM isolation
- Failure if you skip it
- Agent runs a destructive command or exfiltrates data with no blast-radius control
| Plane | Mechanism | Failure if you skip it |
|---|---|---|
| Knowledge | Project RulesVersion-controlled instructions in the repo that every Cursor agent interaction inherits, so standards are encoded once. Press Enter for the full definition. / AGENTS.md | Agent reinvents your conventions every prompt; PRs drift from house style |
| Context | 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 + allowlists | Agent guesses about Jira and observability it can't see or pulls from an ungoverned source |
| Capability | Allow/blocklists, hooks, sandboxing, VM isolation | Agent runs a destructive command or exfiltrates data with no blast-radius control |
The three planes an FDE configures before anyone talks about autonomy.