Ivan Zhao calls AI a “miracle material” like steel. I like that framing. It forces you to zoom out. Steel didn’t just make old buildings stronger; it made skyscrapers possible. AI is the same. It’s not just about faster workflows. It’s about different workflows.
The part that really lands is his warning about the “rearview mirror.” We keep asking AI to look like the last tool we used. Search became chat. Chat became copilots. But the real shift is changing how work gets stitched together when “thinking” is cheap, always on, and happens in parallel.
He names two blockers for knowledge work: fragmented context and weak verifiability.
In the real world, especially in logistics or healthcare operations, these are the exact friction points. Data lives in emails, portals, feeds, and inside someone’s head. And when outcomes are fuzzy, even the best model needs a human to define “good.” Until those two problems are solved, you only get local wins, not system changes.
This is where his idea of “managing infinite minds” gets interesting. It’s a shift in posture. You stop being a worker and start being a system designer. The best operators will learn to delegate to agents the same way they delegate to a team: queue the work, set the guardrails, and check the outcomes later.
The steam engine metaphor he uses is strong. He argues that the “load-bearing walls” of a company are changing. If AI can hold context across workflows, you don’t need humans to carry that context from meeting to meeting. That is a real design shift, not just a productivity hack.
I also like the reminder that we are in the “waterwheel phase.” We are still bolting chatbots onto old, clunky workflows, a meaningful first step. The second step is the real one: redesigning the system around the new material. That is where most companies will hesitate.
What I would add is this: these shifts require a new view on accountability.
If AI generates a weekly report, who signs it? If a system handles customer feedback, who owns the follow-up? In my world, dispatch only works when someone owns the turn. That human obligation doesn’t disappear just because the suggestion was automated. The winners will be the teams that put clear ownership on the output of these “infinite minds.”
What I’m taking away
- Context is culture. People hoard information. We have to make sharing it cheaper than hiding it.
- Feedback loops are gold. If you can’t measure “good,” you can’t scale. Simple, ground-truth data is worth more than a fancy model.
- Legibility matters. Bigger systems get harder to understand. We need dashboards that act as a nervous system, not just vanity metrics.
I appreciate that he anchors this in reality, like his cofounder running coding agents or Notion using 700 internal agents. It proves the “cars” are already on the road.
If I map this back to supply chain, I see the same shape. We are still mostly adding AI chatbots to old software. We haven’t redesigned dispatch around real-time signals yet. The leap isn’t just “better ETAs.” It’s a workflow where the system reallocates capacity and flags risk before a human even asks.
This essay reads like a design brief and a push to rebuild the workflow. It treats AI as a new material with different physics, and that’s the lens worth keeping.
Steel and steam built the modern world. Infinite minds will build the next one. The people who win won’t be the ones adding a chatbot to the old process. It will be the ones who reimagine the process so the old constraints don’t matter anymore.