R.I.D.G.E. framework
A five-letter delegation framework: Role, Inputs, Decisions, Guardrails, Escalations.
R.I.D.G.E. is a delegation framework for AI agents, used by RidgeHQ to scope and ship every agent. The five letters stand for:
• R — Role. What the agent owns. One job, one job description, one set of outputs. If you can’t write a job description for it, it’s probably two roles.
• I — Inputs. What the agent reads. Named tools and data sources, with credentials in a vault, never in the model itself. New inputs are change requests, not silent expansions.
• D — Decisions. What the agent can decide on its own. Listed as positive permissions, narrow on purpose. “Tone” is a decision; “refunds under $50” is a decision; “whether to send” usually isn’t (the human owns send until the agent earns it).
• G — Guardrails. Where the agent can’t go. Listed as negative permissions. “No legal advice.” “No PII echo.” “No off-policy refunds.” Most production AI failures happen because someone forgot to write down a guardrail.
• E — Escalations. When the agent has to ask. Specific triggers, not vibes. “If refund > $200, route to manager.” “If customer message contains ‘attorney’, escalate immediately.” Every escalation has a destination and a SLA.
R.I.D.G.E. is meant to be written down, reviewed at intake, and updated as part of the weekly review loop. It’s the contract between customer, agent, and vendor. When something changes, the card changes; the agent’s behavior follows the card.
The framework was designed for managed AI agents on a retainer, but the structure is general — teams running AI internally have used the same five-letter checklist to write specs for their own agents. The full long-form treatment lives at /r-i-d-g-e.