Writing
Part 6: Work-as-Code: The Micro-MCP Revolution
A lot of engineering work is trapped in human procedure: check this dashboard, update that config, file this ticket, run this script, paste the result somewhere. Work-as-code means turning those repeated procedures into explicit callable interfaces.
The core idea
Micro-MCPs are small task surfaces for agents. They do not need to be grand platforms. A good micro-MCP exposes one useful workflow with clear inputs, outputs, permissions, and failure modes. That is enough to turn an ambiguous chore into an automatable unit of work.
Why it matters
This matters because AI agents are only as useful as the actions available to them. If the company leaves work hidden in tribal knowledge and browser clicking, agents stall. If the company exposes work as code, agents can compose it.
How to use it
- Start with repetitive engineering tasks that already have clear human playbooks.
- Make every tool return structured results and actionable error messages.
- Keep ownership explicit so each micro-MCP has a team accountable for correctness and access control.
The interface pattern
Work-as-code means turning repeated human workflows into narrow, typed interfaces that agents can call. The right abstraction is not "let the agent use the UI." It is a small tool with a schema, validation, authorization, dry-run behavior, and an owner. That turns tribal operating knowledge into something testable.
Micro-MCP tools should be boring. They should wrap one workflow boundary: open a rollout ticket, fetch a service dependency graph, run a migration precheck, request a policy review, create a feature-flag cleanup plan, or stage a report. The agent supplies intent and context; the tool supplies guardrails and a consistent operational contract.
Design constraints
- Prefer many narrow tools over one broad admin tool with ambiguous authority.
- Make side effects explicit: read, dry-run, stage, execute, notify, or delete.
- Return structured results that can be composed by another agent or reviewed by a human.
- Track owner, version, and deprecation path for every workflow interface.
Bottom line
The future of enterprise automation is not one giant agent. It is a growing library of small, trustworthy work interfaces.