AgentBack: Agent-Native Framework Loop

Coding agents build the API through framework constraints; the API becomes an interface other agents can discover and call.

AgentBack connects human intent, coding agents, framework contracts, and agent-facing API surfaces. prompt + product intent generate + verify runtime feedback loops back into the next agent edit Human intent, review product judgment Coding Agent reads repo + docs + tests edits services and contracts framework conventions become the agent's rails AgentBack framework boundary One Boundary Contract Zod schema on route/tool decorators validation + types + OpenAPI + MCP schemas Agent-Native Development Rails DI container, components, bindings auth policy, errors, idempotency, tests /llms.txt, generated docs, inspectable runtime agent writes code with less ambiguity Agentic Interface Surface REST API, OpenAPI, MCP tools, typed clients machine-readable capabilities and errors usable by humans, coding agents, and runtime agents the API becomes a tool surface agents can trust Legend Agent/developer surface Framework runtime rails Contract artifact Human judgment

Agent development loop

  • • Human sets intent and reviews judgment calls
  • • Coding agent edits within framework conventions
  • • Tests, validation, and errors provide feedback

Single contract

  • • Zod schema describes the boundary once
  • • REST, OpenAPI, MCP, and clients derive from it
  • • Contract drift becomes structurally harder

Double win

  • • Agents can build APIs with less ambiguity
  • • APIs become surfaces other agents can call
  • • Frameworks become leverage for both sides

AgentBack • agent-native API and MCP framework • generated for blog sharing