AI & SAAS · REFERENCE MODEL · V1.0.0

Agent-ready SaaS: capability-centred reference architecture.

Separate AI entry points from the business capabilities that actually own policies, transactions and state.

REFERENCE FIGURE / NR-001CONCEPTUAL VIEW · NOT A DEPLOYMENT TOPOLOGY
Agent-ready SaaS: capability-centred reference architecture. Separate AI entry points from the business capabilities that actually own policies, transactions and state.
nishantr · nishantr.techOpen full-size figure ↗
THE PROBLEM

Why this pattern exists

Teams often expose CRUD endpoints to agents and call the product agent-ready, without delegated identity, semantic contracts or transactional controls.

DESIGN GUIDANCE

Principles to preserve

  1. Agents request scoped business outcomes, not arbitrary data mutations.
  2. Business capabilities own validation, authorization and commit.
  3. Use stable intent/action schemas and versioned domain APIs.
  4. Emit domain events only after durable state transitions.
  5. Every action needs identity, traceability and a recovery path.
THE REAL DESIGN WORK

Trade-offs, not just boxes

  • Semantic capability APIs cost more to design than raw CRUD APIs but provide safer evolution.
  • An agent gateway centralizes controls but must not become a single all-knowing business service.
  • Keeping humans in selected approval loops trades latency for risk reduction.
REVIEW QUESTIONS

Before you implement

  • Who owns each business capability and its invariants?
  • Where is delegated user or service authority checked?
  • What is the idempotency and compensation model?
  • Can the product execute correctly if the AI service is unavailable?
MORE FROM THE COLLECTION
NR-002The reasoning–execution boundaryNR-003Enterprise AI control plane
← Back to all patterns