"We are always driving into the future via the rearview mirror." —— Marshall McLuhan
This line describes how most organizations use AI today: bolting a jet engine onto a horse-drawn carriage, then complaining it doesn't go fast enough.
1. Starting with "Miracle Materials": Ivan Zhao's Diagnosis
In late 2025, Notion co-founder and CEO Ivan Zhao published an essay titled Steam, Steel, and Infinite Minds. His core argument can be condensed into a single sentence:
Every era has its "miracle material." Steel forged the Gilded Age, semiconductors opened the digital revolution, and AI now arrives as the next miracle material — in the form of infinite minds.
This isn't rhetoric. Steel wasn't just "harder iron" — it redefined the scale of architecture, railroads, and cities; cold-rolled steel made skyscrapers possible. Semiconductors weren't just "faster circuits" — they redefined the cost of information, computation, and connection, taking the internet from the lab into our pockets.
By the same logic, AI isn't just "a smarter search box." But Zhao observes that most organizations today are making the same mistake:
- Early telephones were used as "telegrams you could speak into," with calls lasting only seconds
- Early films were shot as "filmed stage plays," with cameras locked in place
- Today's AI is treated as "a smarter Google search box" — drop in a question, get back an answer
The problem isn't that AI isn't good enough. It's that we haven't yet invented the "skyscraper" appropriate to this new material.
2. McKinsey's Sobering Data: 89% of Organizations Still Live in the Industrial Age
If Ivan Zhao gave us a metaphor, McKinsey's September 2025 report The Agentic Organization: Contours of the Next Paradigm for the AI Era delivered the cold diagnostic numbers:
| Organizational Paradigm | Share | Defining Traits |
|---|---|---|
| Industrial-era pyramid | 89% | Functional silos, hierarchical reporting, high cross-team friction |
| Digital-era agile / platform | 9% | Squads, iteration, product-driven |
| Decentralized network | 1% | Multi-point decisions, high autonomy, real-time coordination |
In other words, nearly every company is loading 21st-century technology onto a 20th-century organizational chassis. McKinsey gives this emerging paradigm a name: the Agentic Organization — a new structure where humans work side by side with virtual and physical AI agents at near-zero marginal cost to create value together.
Notion Itself Is the Sample
This isn't theoretical. Ivan Zhao offers a hard data point from inside Notion:
1,000 employees, working alongside more than 700 AI agents.
They handle meeting notes, IT tickets, customer feedback synthesis, employee onboarding — the kind of work that used to require dedicated teams.
3. Why It Must Be "Symbiosis," Not "Replacement"
Seeing 700 AI agents, the first reaction many people have is: will the 1,000 employees be cut down to 300?
McKinsey's data points to a different answer. Across the companies they studied, the biggest gap between high performers and everyone else wasn't "scale of AI deployment." It was this:
Use of "human-in-the-loop" mechanisms
- High performers: 65%
- Everyone else: 23%
This ratio reveals something important: the companies actually getting value from AI are the ones thoughtfully designing where humans step in. AI can generate anything but can't be accountable. It can optimize metrics but can't choose values. It can recognize patterns but can't decide which patterns deserve reinforcement.
This is the heart of symbiosis: not humans stepping aside while AI takes over, and not humans merely supervising AI, but the two trading the lead role at different moments in the work.
4. How the Human Role Shifts: From "In the Loop" to "Above the Loop"
This is one of the most clear-eyed and incisive observations in the McKinsey report. In the traditional setting, humans are "in the loop" — reviewing step by step, confirming item by item, approving one at a time. In the agentic organization, the human position moves up a level.
From "Completing Tasks" to "Orchestrating Outcomes"
- Before: an employee = the executor of a task; value = what they personally produced
- Now: an employee = the orchestrator of a cluster of AI agents; value = how much output they can organize
Three New Roles Are Emerging
- Agent Supervisor: sets agents' goals, boundaries, and exception-handling rules
- Workflow Specialist: redesigns the human-machine collaboration path and handles edge cases that AI can't catch
- AI-augmented Frontline Worker: a single person commanding 5–10 specialized agents to do the work that used to require an entire team
Performance Evaluation Shifts Along With It
Talent assessment moves away from "what did this person personally deliver" to three new questions:
- How high is their leverage ratio when working with AI? How many times their own output can they move?
- How well do they orchestrate human-AI collaboration? Are the results repeatable and transferable?
- Are they continuously training the AI to better understand the business? Or does the knowledge live only in their own head?
5. From Florence to Tokyo: Ivan Zhao's Economic Metaphor
In Infinite Minds, Zhao offers a larger prediction: future economies will shift from Florence-style cities to Tokyo-style mega-metropolises.
| Dimension | Florence (past) | Tokyo (future) |
|---|---|---|
| Scale | Walkable, comprehensible | Vast, hard to see in one glance |
| Rhythm | Follows human waking hours | Runs 24/7 |
| Nodes | Person-to-person connections | Thousands of humans + AI woven into a network |
| Legibility | High; obvious at a glance | Low — but extremely efficient |
Hidden in this metaphor is a thorny leadership question: when the organization becomes more like Tokyo — opaque and illegible — can executives still "see the whole picture"? McKinsey's answer: executives stop doing line-by-line reviews and instead set policy, monitor anomalies, and tune the depth of human involvement. "Decision-making" itself becomes layered.
6. Five Pillars for Building a Symbiotic Organization (McKinsey's Framework)
McKinsey lays out a clear restructuring checklist. Any company seriously moving toward a symbiotic organization needs to act on all five pillars at once:
1. Business model: AI-native delivery
Not "bolting AI onto the existing product," but asking again from scratch: if this business were designed from zero, what would it look like? Hyper-personalization, proactive service, and proprietary data assets become the moat.
2. Operating model: AI-first workflows
Not "human-led, AI-assisted," but by default letting AI agent teams push tasks across stages, with humans intervening only at the right nodes.
3. Governance: real-time decisions and an auditable chain
Who can spin up AI agents? What data can they touch? Who bears legal responsibility when something goes wrong? These questions need full-lifecycle governance — from agent onboarding to retirement.
4. Workforce, people, and culture: from "one person, many tasks" to "one person, many outcomes"
Hiring, evaluation, and training all need to be rewritten. The emphasis is no longer functional expertise but judgment, systems thinking, and the ability to decide alongside AI.
5. Technology and data: the substrate that lets agents run
A unified agent runtime, composable enterprise data, and a serious permissions/security layer — this is the invisible infrastructure of the symbiotic organization.
7. Three Action Recommendations for Leaders
Drawing from both sources, here are three things you can start on today:
Action 1: Stop asking — through the rearview mirror — "how do we use AI to speed up our current process?"
Ask instead: if this work were designed from zero, would it be a completely different process? The "Florence-wants-to-become-Tokyo" mindset is a different thing from "using AI as a tool."
Action 2: Set up an "onboarding and retirement" mechanism for AI agents
Decide now: who can create agents, who validates them, under what conditions they're retired. This is one of McKinsey's most repeated points.
Action 3: Redesign "human contribution" inside performance evaluation
Fold leverage ratio, orchestration quality, and AI-training quality into reviews. Otherwise, the old KPIs will push people back onto the lone-wolf track.
8. A Closing Note
Ivan Zhao closes his essay with: "Steel. Steam. Infinite minds. The next skyline is right there, waiting for us to build."
McKinsey gives that optimism a numerical floor: 89% are still standing in place, which means the opportunity is still there. But it also delivers a reality check — this isn't a technology upgrade; it's an organizational restructuring. Whoever completes the restructure first owns the shape of the next generation of enterprise.
The moat of future organizations isn't who deploys the strongest model first. It's who first learns to let humans and infinite minds sit at the same table.
References:
- Ivan Zhao, "Steam, Steel, and Infinite Minds," Notion Blog, December 2025
- McKinsey, "The Agentic Organization: Contours of the Next Paradigm for the AI Era," September 26, 2025
- McKinsey, "The State of AI in 2025: Agents, Innovation, and Transformation," November 2025
Open in small cohorts. If how the next team is built is your question too, come try it with us.