This report is written for two kinds of readers: the employee sitting inside a departmental box, and the owner who draws the boxes. AI entering a company is not an event — it's a road with four steps: swap the people, share the state, gate the writes, swap the species. McKinsey's Agents, Robots, and Us maps the first step precisely; this report walks the whole staircase, then comes back to tell you what each step does to your desk — and to your org chart. This report is written for two kinds of readers: the employee sitting inside a departmental box, and the owner who draws the boxes. AI entering a company is not an event — it's a road with four steps: swap the people, share the state, gate the writes, swap the species. McKinsey's Agents, Robots, and Us maps the first step precisely; this report walks the whole staircase, then comes back to tell you what each step does to your desk — and to your org chart.
Agents, Robots, and Us is the best field guide to the Mode A world — 72% of jobs land in the human-machine partnership zone; humans shift from executor to orchestrator; skills split into three paths (disappear / evolve / persist). Read it. Then notice the one variable it holds fixed: the shape of the company. It asks what happens to work inside the org — never what happens to the org itself. That's not an oversight; it's the audience: incumbents don't hire consultants to be told they're the species being replaced. This report starts where that one stops. Agents, Robots, and Us is the best field guide to the Mode A world — 72% of jobs land in the human-machine partnership zone; humans shift from executor to orchestrator; skills split into three paths (disappear / evolve / persist). Read it. Then notice the one variable it holds fixed: the shape of the company. It asks what happens to work inside the org — never what happens to the org itself. That's not an oversight; it's the audience: incumbents don't hire consultants to be told they're the species being replaced. This report starts where that one stops.
Departments exist for two hidden reasons: they are containers of trust (audit trusts "the department signed off", not any single judgment), and they are how a company decomposes an objective nobody can fully state into local KPIs each box can execute. The cost: information → goods → cash runs as a relay, delay compounding at every handoff; coordination happens in meetings — apparently free, actually the most expensive line item. Departments exist for two hidden reasons: they are containers of trust (audit trusts "the department signed off", not any single judgment), and they are how a company decomposes an objective nobody can fully state into local KPIs each box can execute. The cost: information → goods → cash runs as a relay, delay compounding at every handoff; coordination happens in meetings — apparently free, actually the most expensive line item.
The person in the seat becomes an agent; the org chart doesn't move; humans step back from executor to gate — approving POs, dual-signing payments, catching exceptions. This is exactly the world Agents, Robots, and Us describes. Distance: 3–7 engineering years, paced not by model capability but by two things — each seat's real rules can only be explored through logged stumbles (never installed), and trust runs on the organization's clock: every action type promoted to "auto" needs an audit trail behind it. The person in the seat becomes an agent; the org chart doesn't move; humans step back from executor to gate — approving POs, dual-signing payments, catching exceptions. This is exactly the world Agents, Robots, and Us describes. Distance: 3–7 engineering years, paced not by model capability but by two things — each seat's real rules can only be explored through logged stumbles (never installed), and trust runs on the organization's clock: every action type promoted to "auto" needs an audit trail behind it.
Most of the relay's losses aren't bad decisions — they're information dropped at handoffs. The fix is surprisingly cheap: let everyone (and every agent) read one shared world state; then open cross-department writes one gate at a time. Departments remain — still trust containers — but they read one truth and write through one set of gates. This is the realistic ceiling for incumbents, and it's worth the whole climb: most of the relay's losses die right here. Most of the relay's losses aren't bad decisions — they're information dropped at handoffs. The fix is surprisingly cheap: let everyone (and every agent) read one shared world state; then open cross-department writes one gate at a time. Departments remain — still trust containers — but they read one truth and write through one set of gates. This is the realistic ceiling for incumbents, and it's worth the whole climb: most of the relay's losses die right here.
One core (world model + objective function) solves information, goods and cash jointly; agents exist only on edge interfaces; humans only at the two ends. The three walls keeping incumbents out are none of them technical: the objective function is politics (writing the trade-offs down = writing the power structure down); segregation of duties is statutory control (a single core violates it by design); and the ontology takes years of running Mode A to discover. So Mode B arrives by replacement: native firms born with no departments to dissolve — API-edged businesses first; physical goods, multi-currency and regulated QA last, 10–15 years out. One core (world model + objective function) solves information, goods and cash jointly; agents exist only on edge interfaces; humans only at the two ends. The three walls keeping incumbents out are none of them technical: the objective function is politics (writing the trade-offs down = writing the power structure down); segregation of duties is statutory control (a single core violates it by design); and the ontology takes years of running Mode A to discover. So Mode B arrives by replacement: native firms born with no departments to dissolve — API-edged businesses first; physical goods, multi-currency and regulated QA last, 10–15 years out.
Swapping people is engineering; changing shape is politics; swapping species is the market's job. Whichever seat you're in, the moves are the same: write down your box's hidden rules, earn trust one action type at a time, and read the clock of the step you're actually standing on. Swapping people is engineering; changing shape is politics; swapping species is the market's job. Whichever seat you're in, the moves are the same: write down your box's hidden rules, earn trust one action type at a time, and read the clock of the step you're actually standing on.
Diagrams & full reasoning: Agent-Operated vs AI-Native · A Dual-Flow Map of Org Evolution · Run a simulated company yourself: Metis 2.0 Diagrams & full reasoning: Agent-Operated vs AI-Native · A Dual-Flow Map of Org Evolution · Run a simulated company yourself: Metis 2.0