/07

Transformation

A simple path from today, through copilots and department brains, to bounded autonomous agents.

Leader

11 min

Open +

Do not jump from "we use ChatGPT" to "autonomous company." Each stage needs a different level of context, control, proof, and trust.

Leave knowing / Know what to build next, not someday

The four stages

Move only when the previous stage is boring and reliable.

1. Individual assistance

People use approved AI for drafts, summaries, analysis, research, and learning. The person owns every output.

Output / Copilot

2. Workflow assistance

AI sees the right business context and is embedded inside a repeated process with clear review.

Output / Workflow

3. Department intelligence

Shared tools, knowledge, permissions, and measurement let multiple workflows use the same trusted foundation.

Output / Brain

4. Bounded agents

AI can take multi-step action inside explicit permissions, budgets, stop conditions, logs, and verification.

Output / Agent

Readiness

What has to be true before you move up a stage.

Stage	You need	Do not move on if
Individual	Policy, basic literacy, approved tools	People cannot tell useful output from plausible nonsense
Workflow	Known process, context, owner, metric	Nobody can explain the before state or verify the output
Department	Shared data access, standards, governance	Every team has a different tool and no ownership
Agent	Permissions, state, recovery, audit, hard limits	The system can act without proving what happened

The maturity trap

Autonomy is not automatically maturity. A simple workflow that reliably returns ten hours a week can be more mature and more valuable than an impressive agent nobody trusts.

The goal is not maximum autonomy. The goal is maximum useful work at an acceptable level of risk.

Notes / Subscribe

No schedule. No content calendar. A note goes out only when there is something worth understanding.