Key takeaway
- Steel shaped the Gilded Age. AI is the structural material of this one. Nine lessons for builders and buyers on context, coordination and governance.
What does it mean to call AI the "miracle material" of our era? Steel shaped the Gilded Age and semiconductors built the digital age; AI is the structural material of this one. It is not a better chatbot. It reshapes how work and organisations function, moving from tools that help people work to systems that become the fabric of work itself. That shift carries nine lessons for builders and buyers.
The 60-second brief
- The metaphor: Notion's Ivan Zhao frames AI as "infinite minds", a material that reshapes how work and organisations function, not a better chatbot.
- The shift: from tools that help people work to systems that become the fabric of work itself.
- Where value moves: from UX polish to context depth, from intelligence to coordination, from data moats to context moats.
- The warning: bolting AI onto old processes plateaus at 10–30%; re-architecting work around it yields an order of magnitude.
- The builder's job: think like a city planner, designing for traffic, zoning, utilities and law.
The Miracle Material Argument
In Steam, Steel and Infinite Minds, Notion's Ivan Zhao uses historical analogies to argue that the transformation is not incremental tool improvement but rethinking work around AI that operates like persistent minds. After three decades of building enterprise systems, I agree, and I think the implications for builders and buyers are sharper than most roadmaps admit. Here are nine.
1. AI Is a Structural Material, Not a Feature
Stop treating AI as an add-on: a copilot, a chatbot, a plugin. Tools that assist users will be commoditised. Systems that carry organisational load, memory, coordination and decision-making, will win. Design AI that owns workflows, not just augments steps, and invest in long-lived context and reasoning continuity.
2. Org Charts Will Absorb Non-Human Labour
The unit of work is no longer the human role but the human-and-agent ensemble. Teams become hybrid systems, managers supervise outcomes, and orchestration speed becomes the bottleneck. Build agent governance (scope, permissions, accountability) and track agent throughput, error rates and decision latency rather than headcount. (See The Human-AI Handoff.)
3. Context Is the New Operating System
Whoever controls context flow controls intelligence. AI fails not because models are weak, but because context is fragmented, memory is lossy and decisions are stateless. Build or adopt a Context OS. (The full argument is in Context Is the New Compute.)
4. Productivity Plateaus Without Workflow Redesign
Steam engines inside watermills did not make factories; AI inside ticket systems does not make agentic operations. Redesign workflows assuming agents can run continuously, remember everything and coordinate instantly, and retire the rituals that exist only because of human coordination limits: status meetings, handoffs and reports.
| Step | From bolt-on AI to a re-architected workflow | What happens |
|---|---|---|
| 1 | Map the work | Where decisions, handoffs and waiting actually happen |
| 2 | Remove coordination rituals | Status meetings, reports and handoffs that exist for human limits |
| 3 | Agents own workflow segments | Not just assist individual steps |
| 4 | Govern execution | Scope, permissions and accountability built in |
| 5 | Measure outcomes | Throughput, error rate, decision latency, not headcount |
| 6 | Price on outcomes | Customers pay for work done, not seats |
Bolt-on AI gives 10–30%. Re-architected work gives an order of magnitude.
5. Scale Comes From Coordination, Not Intelligence
Raw model intelligence will converge. The constraint shifts from "can the model reason?" to "can thousands of reasoning threads stay coherent?" Observability becomes strategic infrastructure, not operational hygiene. (See AgentOps and Multi-Agent Coordination.)
6. Data Moats Become Context Moats
Static data advantages erode; lived context compounds: decision histories, outcome feedback, policy and exception memory, organisational preferences. Capture why decisions were made, treat every agent action as a learning event, and build reinforcement loops into workflows. (See The Flywheel.)
7. Governance Moves From Compliance to Control
In a world of infinite minds, uncontrolled intelligence is a liability. Regulated industries will demand determinism boundaries, auditability, reasoning transparency and kill switches. Bake governance into execution, with policy checked before actions run, and position it as what enables scale. (See Bounded Autonomy.)
8. Winners Will Sell Outcomes, Not Software
When AI does the work, customers will not pay for seats or features. Pricing shifts to outcomes: tickets resolved, claims processed, decisions made within policy. Shallow SaaS will not survive; deep post-sales integration will.
9. Builders Must Think Like City Planners
You are not building an app. You are building a cognitive city, and you must design for traffic (task bursts), zoning (permissions), utilities (memory, compute, observability) and law (audit and enforcement):
| AI as structural material: designing the cognitive city (top to bottom) | What it holds |
|---|---|
| Hybrid teams | People and agents as one ensemble; managers supervise outcomes |
| Governance as law | Determinism boundaries, audit, reasoning transparency, kill switches |
| Coordination and observability | Thousands of reasoning threads kept coherent and visible |
| Context OS | Long-lived memory, decision history, policy and exceptions |
| Infrastructure | Compute, data, integration: the roads and utilities |
Builders plan for: traffic (task bursts), zoning (permissions), utilities (memory and compute), laws (audits).
From tools that help people work, to systems that become the fabric of work.
Design for failure, overload and abuse. Assume agents will interact in unexpected ways. Build systems that stay legible at scale.
The Synthesis
From tools that help people work, to systems that become the fabric of work itself.
The builders who understand this will define the era, just as those who mastered steel defined the modern world. For the operating-system view of that fabric, see Why Enterprise AI Needs an Operating System.
Your turn: which workflow in your organisation would change most if you designed it for agents from scratch?
Design One Workflow for Agents From Scratch
ElixirData Context OS provides the long-lived context, decision history and governance layer of that cognitive city, with governed agents running on ElixirClaw. In a Context Workshop, XenonStack works with your business and technology leaders to take one workflow, map where decisions and handoffs happen, and redesign it for agents with governance built in. Related reading: Context Is the New Compute: Why Enterprise AI Fails Without It, AI Native Is Not a Feature: It's a Full-Stack Re-architecture, and 2026 Will Be the Year of Context and Ontology.
Book a Context Workshop | Explore ElixirClaw Agentic OS
Frequently Asked Questions
Why is AI described as a "miracle material"?
Because, like steel or semiconductors, it is a structural material that reshapes how work and organisations function, rather than a feature added to existing tools.
Why does bolting AI onto existing processes plateau?
The old workflow still carries rituals built for human coordination limits. Gains stay at 10–30% until the work is re-architected around agents that run continuously, remember and coordinate.
What is a context moat?
The advantage that comes from lived context that compounds: decision histories, outcome feedback, policy and exception memory, and organisational preferences.
How does governance change in a world of AI agents?
It moves from compliance to control: policy checked before actions run, with determinism boundaries, auditability, reasoning transparency and kill switches.
What does it mean for builders to think like city planners?
Designing for traffic (task bursts), zoning (permissions), utilities (memory, compute, observability) and law (audit and enforcement), so the system stays legible at scale.
Sources
This article first appeared in Navdeep Singh Gill's LinkedIn newsletter, AI + Human = Human Squared.
Explore the XenonStack Platform and Agent Network
Platforms: ElixirData · ElixirClaw · ElixirHub · NexaStack · MetaSecure · Neural AI · AutonomousOps · CohortIQ · AI Aviator · SAIF Aviator
AI agents: AgentAnalyst · AgentEvaluation · AgentForce · AgentGRC · AgentHR · AgentInstruct · AgentLabel · AgentPixel · AgentQA · AgentRAI · AgentScrum · AgentSearch · AgentSketch · AgentSRE
XenonStack: xenonstack.com · xenonstack.ai · xenonstack.ae · xenonstack.jobs
Explore the XenonStack platform family: ElixirData · ElixirClaw · ElixirHub · NexaStack · MetaSecure · Neural AI · AgentGRC · AgentSRE · AgentAnalyst · AgentInstruct · AgentRAI · AgentScrum · AgentSearch · AgentForce · AI Aviator · SAIF Aviator · CohortIQ · XenonStack AI · XenonStack UAE
