Divya van Mahajan

Work in the Age of Infinite Agents

ai-daily-brief-podcast

Work in the Age of Infinite Agents

Overview

This episode of the AI Daily Brief (recorded January 4, 2026) presents and contextualizes two full-length essays on AI, knowledge work, and economic transformation. The host (NLW) frames both essays as early contributions to an emerging canon that articulates a future beyond simple productivity gains or job displacement.

The two essays are:

  1. “Steam, Steel, and Infinite Minds” by Ivan Zhao, CEO of Notion
  2. “Jevons’ Paradox for Knowledge Work” by Aaron Levy, CEO of Box

Both were originally published publicly on X (formerly Twitter).

Source video URL: Not provided

Prerequisites

Readers will benefit from familiarity with the following:

  • Basic history of industrial revolutions (steam power, electrification, steel construction)
  • The concept of knowledge work and its role in modern economies
  • Foundational AI concepts: agents, large language models, reinforcement learning
  • Jevons’ Paradox (19th-century economics): efficiency gains in resource use lead to increased total consumption rather than decreased consumption
  • Marshall McLuhan’s “rearview mirror” concept: new technologies are initially understood and used through the lens of the old technologies they replace
  • General familiarity with enterprise software history (mainframes, minicomputers, PCs, cloud SaaS)

Main Points

The “Rearview Mirror” Problem in AI Adoption

  • Marshall McLuhan observed that new technologies are first understood as improved versions of old ones: early phone calls mimicked telegrams; early films looked like stage plays.
  • Today’s AI tools (chatbots modelled on search boxes, agents trained to replicate human workflows) repeat this pattern.
  • The host adds that startups currently deploying AI to watch and copy human worker behaviour represent a necessary but short-lived transition phase.
  • The deeper opportunity lies in reimagining workflows from first principles, not optimising old ones.

Individuals: From Bicycles to Cars (Ivan Zhao)

  • Steve Jobs’ 1980s metaphor of computers as “bicycles for the mind” is now outdated; knowledge workers have been “pedalling bicycles on the Autobahn.”
  • AI coding agents illustrate the upgrade: Zhao’s co-founder Simon went from being a 10x engineer to a 30–40x engineer by orchestrating multiple agents simultaneously, queuing tasks before sleep, and acting as a manager of minds rather than a direct producer.
  • Two barriers prevent this shift from generalising to all knowledge work:
    • Context fragmentation: Coding tools consolidate context in one place (IDE, repo, terminal); general knowledge work is scattered across dozens of tools and resides partly in people’s heads.
    • Verifiability: Code can be tested objectively; there is no equivalent signal for whether a strategy memo or project plan is good, limiting reinforcement-learning-based model improvement for general tasks.
  • Once these two problems are solved, the progression will move from “pedalling → driving → self-driving.”

Organizations: Steel and Steam (Ivan Zhao)

  • Steel metaphor: Pre-steel architecture was limited to ~6–7 floors; steel’s strength and malleability enabled skyscrapers. Similarly, AI can be the structural material that allows organisations to scale without the communication overhead that currently causes them to buckle.
    • Weekly alignment meetings could become five-minute async reviews.
    • Multi-level approval chains could compress to minutes.
  • Steam engine metaphor: Early factory owners swapped water wheels for steam engines but kept everything else the same — modest gains. Real gains came when they relocated factories and redesigned layouts around steam. Electrification later decentralised power further, triggering the Second Industrial Revolution.
    • Current AI adoption (chatbots bolted onto existing tools) mirrors the “swap the water wheel” phase.
    • True gains require reimagining organisational structure itself.
  • Notion’s own experiment: alongside ~1,000 employees, 700+ agents handle meeting notes, IT requests, customer feedback logging, employee onboarding Q&A, and status reports.

Economies: From Florence to Megacities (Ivan Zhao)

  • Pre-industrial cities (e.g., Florence) were human-scaled — walkable, voice-carried, comprehensible.
  • Steel frames and steam-powered railways enabled megacities (Tokyo, Chongqing, Dallas) — not just larger Florences, but fundamentally different ways of living offering more opportunity at the cost of some legibility.
  • Knowledge work (~50% of US GDP) still operates at human scale: small teams, meeting-paced workflows, organisations that break down past a few hundred people.
  • AI agents at scale will produce the equivalent of megacities in the knowledge economy: continuous cross-timezone workflows, thousands of agents and humans operating in concert, new decision rhythms replacing quarterly cycles and annual reviews.

Jevons’ Paradox Applied to Knowledge Work (Aaron Levy)

  • Historical pattern: Each order-of-magnitude reduction in computing cost produced an order-of-magnitude increase in adoption — mainframes (hundreds of units) → minicomputers (tens of thousands) → PCs (millions). Cloud eliminated the procurement/maintenance advantages of large enterprises for deterministic software work.
  • The new frontier: AI agents extend this democratisation to non-deterministic knowledge work (contract review, code generation, market research, 24/7 customer support) — previously impossible to automate at scale.
  • The investment cost argument: The mistake in ROI thinking is treating the return as the key variable; the real leverage is reducing the cost of investment. Small teams previously had to choose between building a marketing page, improving the product, or handling support — AI collapses many of those trade-offs.
    • Example: A 10-person firm that could never justify custom software can now have a team member prototype a full app in days.
  • On jobs: AI takes over tasks, not jobs. Incorporating task outputs into full workflows still requires human judgment. “Today’s jobs become tomorrow’s tasks.”
    • Marketing jobs in the US grew ~5x (from hundreds of thousands to low millions) between the 1970s and today, because technology made marketing cheaper and accessible to more companies — not despite it.
  • The demand expansion conclusion: The vast majority of future AI compute will be used on work that does not exist today — software projects not yet started, contracts not yet reviewed, medical research not yet attempted, campaigns not yet launched.

The Host’s Synthesis

  • The two essays are complementary: Zhao explains how the structure of work and organisations will change; Levy explains why total demand for knowledge work will expand rather than collapse.
  • The destruction side of creative destruction is visible first; the creation side lags.
  • The intended audience for the AI Daily Brief is those who choose to actively shape how AI changes their role, rather than waiting passively.

Key Concepts

  • Infinite minds: Ivan Zhao’s term for AI agents — cognitive resources that do not sleep, can be parallelised, and can be deployed at organisational scale.
  • Context fragmentation: The dispersion of information relevant to a knowledge work task across many tools, making it difficult for AI agents to operate autonomously outside narrow domains.
  • Verifiability gap: The absence, in general knowledge work, of objective test signals analogous to code execution, which limits reinforcement-learning-based AI improvement for those domains.
  • Waterwheel phase: The current stage of AI adoption, in which AI is inserted into existing workflows without redesigning those workflows — analogous to early factory owners replacing water wheels with steam engines but otherwise changing nothing.
  • Jevons’ Paradox: The economic principle (William Stanley Jevons, 19th century) that efficiency improvements in resource use lead to increased total consumption of that resource by unlocking new use cases.
  • McLuhan’s rearview mirror: Marshall McLuhan’s observation that societies interpret new technologies through the lens of prior technologies, causing them to misread the new medium’s true potential.
  • AI agents: Software systems that can take sequences of actions, make decisions, and complete multi-step tasks with varying degrees of autonomy, as distinct from single-turn chatbots.
  • Non-deterministic knowledge work: Tasks whose outputs cannot be fully specified in advance by rules (e.g., drafting strategy, reviewing contracts, creative work), contrasted with deterministic software tasks.
  • 10x / 30–40x engineer: A shorthand for the multiplicative productivity leverage that AI agent orchestration provides to skilled individual contributors.

Summary

Both essays, read together, argue that AI is not primarily a tool for doing existing work faster, but a “miracle material” — analogous to steel and steam — that will enable qualitatively new structures at the individual, organisational, and economic levels. Ivan Zhao contends that knowledge work is still in the “waterwheel phase,” with AI bolted onto human-designed workflows, and that the real transformation will come when organisations are rebuilt around AI the way factories were rebuilt around steam and electricity. Aaron Levy extends the argument through Jevons’ Paradox: because AI dramatically lowers the cost of performing non-deterministic knowledge work, demand for that work will expand by an order of magnitude or more, producing net job and task growth rather than net elimination — just as cheaper marketing technology grew the marketing workforce fivefold over fifty years. The host frames these essays as important early articulations of a constructive, expansionist vision of AI’s economic role, and positions 2026 as the year in which practitioners need to move beyond optimising old processes and begin imagining the fundamentally different workflows that the new capabilities make possible.