Skip to content

AI operating model

What's on this page

  • How humans and cyborgs collaborate at Inkorporated
  • Risk tiers and human-in-the-loop
  • Knowledge sources (this docs monorepo)
  • Anti-patterns for agent sprawl

What this enables

Human-in-the-loop tiers

flowchart TB
  req[Agent proposal] --> tier{Security level}
  tier -->|CRITICAL| hitl[Human approve]
  tier -->|HIGH| review[Approve side effects]
  tier -->|MEDIUM| sample[Sampled review]
  tier -->|LOW| auto[Automated with limits]
  hitl --> act[Execute tools]
  review --> act
  sample --> act
  auto --> act
  • Productive automation without silent high-risk actions
  • Consistent persona quality via job roles + YAML specs

Principles

  1. Role-aligned agents — Every cyborg maps to a role code.
  2. Docs as knowledge — Agents load relevant docs/ paths listed in YAML knowledge_docs.
  3. Tool allowlists — MCP/tools declared explicitly; no ambient superuser.
  4. HITL for CRITICAL — Side effects require human approval.
  5. Auditability — Decisions and tool calls are logged.

Lifecycle

Design role → write job page → author cyborg YAML → wire tools → evaluate → deploy to namespace → monitor.