Skip to content
The Future Explorerby Benjamin Talin
The Optimisation of EverythingIn the book · Ch. 5Megatrend

Cognitive automation

Automation of Cognitive Work

'Cognitive work' is thinking work: writing, summarising, coding, answering customers. This megatrend is machines taking on more of any task that can be described in words. When customer-support staff were given an AI assistant, they solved about 15% more problems an hour on average, and beginners improved by about 30%. The software keeps getting better and cheaper; what slows it is people and organisations learning to trust it and work with it.

Tasks that can be specified in language trend toward machine execution.

2.7/ 5

Gradient 2.7 / 5 - weighed down most by the weakest enabler.

The average of these enablers is 3.1. The ramp runs at 2.7.

Shared weakest link at 2: Data & Datafication, Digital Transformation, Managerial Capability, Neuromorphic & Alternative Compute, Regulatory Clarity, Workforce Trust. The ramp is held back by all of them at once, and deepening one alone lifts it only a little.

6 enablers have to mature together before this ramp steepens.

In the network

Its enablers on the left, least mature first and drawn largest; the megatrends it is coupled with on the right. A thicker line is a heavier weight; the arrow shows which way a coupling pushes.weakest linkamplifiesgates

Also shaped by The Closing of the Planetary Envelope (1 shared enabler) · The Collapse of Shared Reality (4 shared enablers, 3 couplings) · The Decoupling of Physical and Digital Life (4 shared enablers) · The Inversion of the Population Pyramid (3 shared enablers, 1 coupling) · The Rebuild of the Physical Base (10 shared enablers, 1 coupling) · The Redistribution of Power (9 shared enablers, 1 coupling)

Enablers

EnablerFamilyWeightMaturityRole
Data & Dataficationtechnologicalmoderate
2/5
shared weakest link (with 5 others)
Digital Transformationtechnologicallarge
2/5
shared weakest link (with 5 others)
Managerial Capabilityhumanlarge
2/5
shared weakest link (with 5 others)
Neuromorphic & Alternative Computetechnologicalmoderate
2/5
shared weakest link (with 5 others)
Regulatory Claritystructurallarge
2/5
shared weakest link (with 5 others)
Workforce Trusthumanlarge
2/5
shared weakest link (with 5 others)
Adoption Appetitehumanlarge
3/5
-
AI & Machine Learningtechnologicallarge
3/5
-
Capital Availabilitystructurallarge
3/5
-
Energy Price & Availabilitystructuralmoderate
3/5
-
Deep Learning & Neural Networkstechnologicallarge
4/5
-
Large Language Modelstechnologicallarge
4/5
-
Advanced Chips & Semiconductorstechnologicallarge
5/5
-
Automationtechnologicallarge
5/5
-
Cloud & Datacentre Infrastructuretechnologicallarge
5/5
-
Brain-Computer Interfacestechnologicalmarginal
1/5
marginal: not counted as a weakest link
Quantum Computingtechnologicalmarginal
1/5
marginal: not counted as a weakest link
Autonomous Vehiclestechnologicalmarginal
3/5
marginal: not counted as a weakest link

The gradient behaves like a sticky rubber band: it is pulled hardest toward the weakest enabler. Marginal links are shown but do not pull. Everything above 2 is strength this ramp can use only in part until the weakest link moves.

What this pushes on, and what pushes on it

Bound by

Workforce trust, managerial capability, regulation

NOT compute. Human and institutional absorption.

Momentum. Capability improves continuously and cost per unit falls; neither depends on any institution agreeing to it.

The four clocks

Four systems absorb this at four speeds. The distance between them is the adaptation gap.

Individuals

months

Weeks. The fastest adoption of any technology measured, because it rides infrastructure people already own.

Companies

years

One to five years to adopt, five to forty to reorganise around it. The gap between those is the chapter's subject.

Governments

decades

Three to thirty years. Labour, liability and tax law all assume a human performing the task.

The biosphere

not ours to set

Indirect, through the energy and water cost of the infrastructure build.

Impacts by domain

Social & demographic

Freelance writers, coders and other online knowledge workers saw monthly job postings fall 2.0% and earnings fall 5.2% in the months after ChatGPT's public release, relative to freelancers in unexposed trades (Hui, Reshef & Zhou, Organization Science, November 2024).

Environmental

Generative AI's energy draw depends heavily on the task: 1,000 text queries to text-generation models of at most 6.7B parameters burn about 0.047 kWh, but 1,000 image generations burn 2.9 kWh (Luccioni, Jernite & Strubell, ACM FAccT, June 2024).

Economic

Give customer-support staff a generative AI assistant and resolved issues per hour rise about 15% on average, with the gain concentrated among less experienced and lower-skilled workers, while the most experienced see small gains in speed and small declines in quality (Brynjolfsson, Li & Raymond, Quarterly Journal of Economics, 2025).

Geopolitical

Washington has banned the export to China of logic chips with interconnect bandwidth of 600 GB/s or more and the advanced lithography tools used to make them (US Bureau of Industry and Security, Federal Register, October 2022 and October 2023).

What would prove this wrong

Sustained total factor productivity growth attributable to AI above the pre-AI trend, which would show absorption is no longer the constraint. Or the reverse: capability plateauing while the application gap stays open.

Talin, B. (2026). Automation of Cognitive Work. The Future Explorer, dataset v0.4.0 (2026-11-16). https://explorer.talin.digital/m/automation-of-cognitive-work/