Field guide · six original resources
Build the boundary, not just the wrapper.
Technical decisions for making existing software legible, useful, and accountable to agents. Each resource begins with a direct answer, uses examples from the owned synthetic lab, cites primary sources, and states its claim boundary.
Agent interfaces · 7 min
API vs. MCP vs. an Agent Interface
Understand the distinct jobs of a domain API, an MCP adapter, and the complete product boundary an AI agent can safely use.Tool design · 8 min
Designing Tools Agents Select Correctly
Design agent tools around distinct intent, constrained inputs, useful errors, and evaluations that measure both correct selection and avoidance.Agent security · 9 min
OAuth, Scopes, and Confirmed Agent Writes
Design agent authorization so identity, audience, scope, preview, human confirmation, version checks, and idempotency reinforce one another.Interface evaluation · 8 min
Testing Agent Interfaces with Golden Workflows
Build an interface release gate that tests contracts, tool choices, permissions, confirmation, idempotency, latency, and completed product state.MCP productionization · 7 min
Migrating to MCP 2026-07-28: An Interface Checklist
Plan an MCP protocol migration as a versioned interface release across transport, authorization, schemas, clients, tests, and operational evidence.Reference implementation · 10 min
Teardown: The Synthetic Customer Request Interface
Inspect the architecture, tool contracts, scoped authorization, confirmation capability, data isolation, and tests behind Permadyn’s owned demo.Reference proof
Use the lab while you read.
Every article points back to inspectable contracts and a live synthetic workflow. No client identity or result is implied.
Open the operations lab