Field Notes
Written from inside the work.
Notes on governed AI inside companies, from a platform running in production. What actually blocks adoption, how we build with agents, and what we have got wrong.
6 notes published.
Governance and delivery
How ownership, permissions and evidence make AI delivery safe enough to move beyond a pilot.
- 8 min read
How does Miiwa decide who can run a Miiwa Agent?
Miiwa resolves agent access from a workspace baseline and explicit user overrides, then enforces the result again when a run starts.
- 8 min readStart here
Who should build in Miiwa—and who should run the finished tools?
Miiwa gives system design to admins and builders, then delivers published, permissioned Miiwa Agents to users.
Building with agents
Patterns, mistakes and practical methods for building dependable work with AI agents.
- 9 min read
How should a Miiwa Builder design inputs for a Miiwa Agent?
A strong Miiwa Agent asks only for the business context its graph needs, using explicit field keys that become dependable workflow variables.
- 9 min readStart here
How does Miiwa turn a workflow into a reliable AI agent?
Miiwa makes agents reliable by publishing validated native workflow versions, then running every trigger through the same stateful execution core.
Operating AI inside a company
What changes when AI becomes part of real workflows, teams and operating responsibilities.
- 9 min read
What does Miiwa record when a Miiwa Agent run fails?
A failed Miiwa Agent leaves a workspace-scoped run record and event trail, while retry and replay depend on the kind of failure and operator authority.
- 8 min readStart here
What changes when a Miiwa Agent moves into real work?
In Miiwa, production starts when a validated workflow version becomes a deployed, assigned and observable capability inside a workspace.