By Bryan Subijano

Anatomy of
Work

The Intelligence Revolution & How AI Will Rewire the $60 Trillion Knowledge Economy

"Software encoded our workflows. AI will execute them."
32→71
Life expectancy transformed by technology over a century
10×
Population growth in 250 years; prosperity grew 14×
$60T
Knowledge economy — the next frontier for AI
$1.5T Global AI Spend 2025$690B Hyperscaler Capex 2026$3.5T AI Market by 203343% Knowledge Workers use ChatGPT5 Days to 1M Users92% of US GDP Growth from AI Infra50 Zettabytes of Healthcare Data70% of FinServ Work Automatable $1.5T Global AI Spend 2025$690B Hyperscaler Capex 2026$3.5T AI Market by 203343% Knowledge Workers use ChatGPT5 Days to 1M Users92% of US GDP Growth from AI Infra50 Zettabytes of Healthcare Data70% of FinServ Work Automatable
The Macro Opportunity

Technology has always been
the lever of civilization

32Life Expectancy (1900)Now 71 years globally — technology drove this transformation
10×Population GrowthIn 250 years, while economic prosperity grew 14× — powered by technology
$60TKnowledge EconomyThe next frontier. AI is the lever — just as steam, electricity, and the internet were before it
Three Productivity Revolutions

We are living through the third —
and it is moving faster than anything before it

1875

Industrial Revolution

Electricity, steel, and machinery led to mass production (1908). Physical labor mechanized at scale for the first time in history.

1970 – 1994

Digital Revolution

Computers (1970) and the internet (1994) digitized information and codified business logic into software systems.

2022+

Intelligence Revolution

Generative AI — the most powerful lever yet. It doesn't just encode workflows; it executes them. Faster, better, cheaper.

5Days to 1M UsersChatGPT's record-breaking launch
2 moTo 100M UsersFastest consumer app in history
43%US Knowledge WorkersUse ChatGPT today (up from 10% in late 2022)

Years to Mass Adoption — Technology Comparison

Electricity (10%→99%)
50 yrs
Social Media (5%→80%)
15 yrs
ChatGPT (0→100M)
2 mo
The $60 Trillion Knowledge Economy

The gap between services and software
is the opportunity

$100TGlobal GDPKnowledge work = 60% ($60T) — largely untouched by AI
70%+US Services Economy~$23T of GDP. Service industries generate $6T+ annually
$6T vs $370BThe GapService economy vs. software market — services remain largely untouched by AI
Why AI Is Uniquely Suited for Knowledge Work
📄

LLMs

Process documents, search, retrieve, and reason through semantic information

👁

Vision Models

Read, understand, and extract information from any visual format

🎙

Voice AI

Make phone calls and conduct conversations autonomously

🌐

Web Agents

Navigate websites, portals, and digital systems end-to-end

🧠

Reasoning Models

Reason toward goals and outcomes with advanced planning

Three Ways AI Creates Economic Value

Increase Profit

Automate tasks → higher efficiency → lower costs. Direct margin improvement through labor reduction.

Drive Revenue

Augment workers → higher throughput → incremental revenue without incremental cost.

Create New Business

Reinvent service delivery & business models → entirely new value creation at scale.

Anatomy of Work — The Framework

AI acts on atomic units of work

Task-level automations compose into full workflow automations — the foundation of intelligent transformation.

Units of Labor

Workers → Teams → Organization. The human organizational structure that AI augments and transforms.

Units of Work (AI)

Tasks → Workflows → Outputs → Outcomes. The atomic building blocks that AI automates, from individual tasks to measurable business outcomes.

Units of Intelligence

Knowledge → Skills → Reasoning. The cognitive capabilities that AI brings — retrieval, application, and complex reasoning over domain data.

From Digital to Intelligent Transformation

Digital Transformation (Software)

  • Data digitized in databases
  • Workflows encoded in software
  • Business logic codified in systems

Intelligent Transformation (AI)

  • AI uses the digital substrate to automate work
  • Executes workflows — not just encodes them
  • Delivers measurable economic outcomes
AI Delivers Better, Faster, Cheaper

Faster

Higher throughput — more outputs per unit time per unit labor

Better

Higher quality — equal or better than human benchmark, with consistency

Cheaper

Higher efficiency — less cost for same or better outcome

The AI Capability Curve — 7 Chapters

Each chapter unlocks new economic utility

We are entering the most commercially powerful phase. The pace of advancement is accelerating.

01

Chat 2022

Generate, draft, summarize. ChatGPT's accessible design created a cultural moment. ~10% of world population uses ChatGPT weekly. Character.ai: 20M MAU, avg 2 hrs daily. OpenEvidence: 40% of US doctors use it daily.

02

RAG 1.0 2023–24

Grounded answers on private data. Architecture: embedding models + vector databases + LLMs + enterprise apps. Horizontal (Glean) and vertical (Harvey for legal) implementations.

03

RAG² Research 2025

Research-Assisted Generation + Retrieval-Augmented Generation. Complex multi-step reasoning combining private internal data + public web. OpenAI Deep Research, Perplexity, Exa.

04

Copilots 2025

Workflow assistance connected to systems of record. Harvey (Legal), Vultron (Gov't contracting), Penguin (Healthcare). Measurable efficiency, throughput, and quality improvements.

05

Agents NOW

Deterministic workflows + probabilistic models working in concert. The "sweet spot" — AI intelligent decision-making with predictability of automations. Zapier, Vultron (30+ pre-built agentic workflows).

06

Autonomous Systems

Goal-driven agents: perceive → reason → act → learn. Value delivered as outputs ($2K+ savings per automated RFP) and outcomes — measurable business impact with high attribution.

07

Invention

Scientific superintelligence. Weco: ~20% cost reduction algorithms. Vinci: 1,000× faster semiconductor simulations. Chai Discovery: 100× improvement, near-20% hit rate in drug design.

The Hierarchy of Agentic Capabilities

Surge AI tested 9 frontier models in realistic company environments. Key finding: even GPT-5 and Claude Sonnet 4.5 fail 40%+ of tasks in realistic workflow environments.

1 · Tool UseGPT-4o can't reliably break multi-step tasks into subgoals
2 · PlanningBasic planning and goal formation — still brittle
3 · GroundednessModels hallucinate IDs, dates, facts; drift from context
4 · ReasoningEven GPT-5 misses obvious common-sense inferences

Source: Surge AI, 2025. "2025 isn't the year of human-level agents. It's the year we can start seriously diagnosing what's missing."

METR — Measuring AI Progress

Exponential progress on
long-horizon tasks

The length of coding tasks frontier systems can complete is growing exponentially — with recent acceleration.

METR Time Horizon — Task Completion Doubling Rate

~7 mo
Doubling time (2019–2025)
~4.3 mo
Post-2023 doubling time (METR v1.1, Jan 2026) — 20% faster
~100 min
Current frontier 50% time horizon (intellectual domains)
Tasks < 4 min
~95%
Tasks ~30 min
~50%
Tasks 1–4 hrs
~15%
Tasks > 4 hrs
<10%

Source: METR, "Measuring AI Ability to Complete Long Tasks" (Mar 2025); Time Horizon v1.1 update (Jan 2026); latest models incl. Gemini 3 Pro & GPT-5.1 Codex (Feb 2026). metr.org/time-horizons

GDPval — Real-World Economic Task Performance

1,320
Tasks evaluated across 44 occupations, 9 industries
100×
Faster & cheaper — frontier models vs. industry experts
70.9%
GPT-5.2 beats or ties top professionals (Dec 2025)

Performance has more than tripled from GPT-4o to GPT-5 in just one year. GPT-5.2 Thinking produces outputs at 11× speed and less than 1% the cost of human expert professionals. Claude Opus 4.1 excels in aesthetics and formatting; GPT-5 excels in accuracy and domain knowledge. The latest GPT-5.4 (Mar 2026) now achieves state-of-the-art on GDPval with 75% OSWorld success rate — surpassing human performance.

Source: OpenAI, "Measuring the performance of our models on real-world tasks" (Sep 2025); "Introducing GPT-5.2" (Dec 2025); Artificial Analysis GDPval-AA Leaderboard. openai.com/index/gdpval

Massive Capital Flowing Into AI

The largest technology investment
cycle in history

$1.5TGlobal AI Spend (2025)Forecast to top $2T in 2026 (Gartner)
$690BHyperscaler Capex (2026)Big 5 nearly doubling 2025 levels
$3.5TAI Market by 2033From ~$390B in 2025. 30.6% CAGR

AI Infrastructure Is Driving US GDP

Without data centers, US GDP growth was just 0.1% in H1 2025. AI infrastructure was responsible for 92% of GDP growth.

Tech Capex 2025
~1.8%
Broadband Buildout
~0.8%
Electricity Expansion
~0.7%
Apollo Moon Landing
~0.5%
Interstate Highway
~0.4%

% of GDP · Source: Slide data, comparable to Goldman Sachs and Morgan Stanley estimates

AI's Rising Concentration in Markets
34%Mag 7
Magnificent 7 — 34% of S&P 500
Rest of S&P 500 — 66%

Mag 7 delivered 42% of S&P's total return in first three quarters of 2025. AI-related firms: 14% of investment-grade bond index, $1.2T total debt.

The Risk: $500B/year capex needed, $2T/year revenue needed to justify, $800B funding gap remains.

Vertical Deep Dives

Massive markets ready for
AI transformation

Labor-intensive, document-heavy, policy-bounded, high-volume — the perfect beachhead for Applied AI.

VERTICAL DEEP DIVE

Healthcare — $4.9T, 18% of US GDP

21.7MEmployees12% of US labor force
$14K+Cost Per PersonUp from $353 in 1970 — 6× increase, inflation-adjusted
1 in 3Adults in Medical Debt$220B total medical debt outstanding
50 ZBHealthcare Data⅓ of world's data; 97% unused; 80% unstructured
$1TAdmin Mkt
Payers & Insurance Claims — 49%
Revenue Cycle & Billing — 26%
Provider Admin Costs — 25%

9 processes make up 70% of admin cost. $265B savings opportunity identified. Healthcare BPO: $486B → projected $1T in 10 years.

Penguin — Admin AI

Utilization management review time reduced by 80%. Prior authorization automation. RCM denial prediction + AI copilots. Goal: 85–90% collection rate (from 70–75%). >50% of RCM workflows automated in 3–5 years.

Abridge — Clinical Documentation

Saves physicians 2 hrs/day. 78% improved job satisfaction at Sutter Health with 49% reduced cognitive load. Lee Health: 86% less after-hours work, 57% completing notes same day.

Hoppr — Radiology AI

Captures ~$30M in lost revenue from patient leakage per customer. ~75% cost savings on radiologist labor. Radiology Partners: 72% of radiologists using AI daily, 20M+ patient exams processed.

VERTICAL DEEP DIVE

Insurance — $8T Global Market

$8TGlobal Market→ $11T by 2029
$3.5TUS MarketHealth $1.5T, P&C $1T, Life $1T
$300BBPO SpendAnnually on back-office document processing

Pace — AI-Native BPO

Intake complex docs → Reason over business logic → Act via APIs/portals. Use cases: submission intake, policy servicing, first notice of loss.

Reserv — Tech-Enabled TPA

Full-service claims engine: modern software + AI + in-house adjusters. 1.6–2.6× faster cycle times. 94+ CSAT score for 6 months straight. Migration: 9 months → 2 weeks. 350+ employees, 80+ MGA clients, 20 carriers.

Business Ontology

The secret weapon. Map objects → Enable integration → Drive automation → Identify leverage. An ontology = the syntax and grammar of a company. Those who build them first compound fastest.

VERTICAL DEEP DIVE

Accounting — $643B Global Market

$146BUS Market~45,000 CPA firms
Throughput GainFor every $1M revenue, 10 people → $2M with AI

Black Ore (Tax)

Cost per return: $300 → <$25 (90% decrease). Margins: 40% → ~95% (2.4× uplift). Time: 2.5 hours → 30 min (5× productivity increase).

Kick (Bookkeeping)

One bookkeeper manages 311 businesses vs. industry norm of 30 — a 10× improvement. AI-native firms: 40–60% margins vs. traditional 20–30%.

Crete PA (Full-Service CPA)

$400M revenue, 20 acquired firms. Reduced billable hours by 30% using AI. Massive labor shortage means AI-enabled firms can take 2–3× more clients.

VERTICAL DEEP DIVE

Financial Services — $7.4T US Market

70%AutomatableOf financial services work can be automated or augmented
$35B→$97BAI Spend2023 → 2027 in financial services
6.7MEmployees7% of US GDP, $410T global banking assets
70%AI-Ready
Full Automation — 35%
Augmentation — 35%
Human-Only — 30%

Savvy Wealth

$0 → $2.2B AUM in under 3 years. Saves advisors 19 hrs/week. Wealth management: $1.8T → $3.5T by 2033 at 12% CAGR. Fragmented — top 10 control just 13%.

Range

10–20× faster advice. 75–90% fee savings vs. traditional advisors. $3B assets, 1,000+ HNW members. Fifteenth tax advisory: avg savings $10–20K, some $100K+.

VERTICAL DEEP DIVE

IT Services — $300B Untouched Market

Titan

Acquires exceptional IT firms + deploys AI tools. Onboarding: weeks → minutes. 40% workflow automation achieved. 2–3× improvements in net margins.

Shield Technology

AI-enabled managed IT platform. M&A + innovation approach. Local businesses gain national-scale resources + world-class tech expertise at a fraction of traditional cost.

Horizontal Opportunities

Customer Service: Crescendo 50–80% automation, 3× resolved calls, ~60% gross margins. IT: Moveworks 50–90%+ autonomous resolution. Finance back-office: payables, reconciliations, close checklists.

Identifying AI Transformation Opportunities

Near-term upside concentrates where units of work are frequent, time-consuming, and valuable — and where labor is costly or scarce.

Automate First

Low frequency, high value — Healthcare admin, insurance, legal volume

Agentic Copilots

High frequency, high value — Tax advisory, M&A, government RFPs

Agents at Scale

Low frequency, low value — Logistics, customer ops, IT desks, back-office

Low Priority

High frequency, low value — Rare, low-value tasks

← Low FrequencyHigh Frequency →
Business Model Innovation

From seats to outcomes

Top AI companies charge 25–50% of value delivered vs. SaaS at 10–15%. Only 5% use outcome pricing today; 25% forecast in 3 years.

① Low / Low

Seat-Based

Low attribution, low autonomy — traditional SaaS. Pays per user, not per outcome.

② High / Low

Hybrid

High attribution, low autonomy — seat + consumption. Copilot with measurable attribution.

③ Low / High

Usage-Based

High autonomy, low attribution — scales with volume but misses value alignment.

Playbook for Founders & Builders

Six compounding principles

Start at the Task Level

Inventory top tasks by time × business value. Target document-heavy, policy-bounded, high-volume beachheads where ROI is immediate.

Wrap the Workflow

Automate end-to-end, not isolated steps. Intake → understanding → action → verification → logging.

Instrument Quality

Ship with evals, not dashboards alone. Define accuracy thresholds. Continuous evaluation catches model drift before it reaches customers.

Price to Outcomes

Charge on resolution/output, not seats. Captures 25–50% of delivered value — 2–3× SaaS economics.

Build the Compounding Stack

Data → Ontologies → Compute → Governance. Each layer compounds advantage through network effects over time.

Invest in People

Reskill toward judgment + machine throughput. Not replacement — elevation. 30–80% of tasks automated, freeing humans for higher-value work.

Macro Context

The US economy needs AI-driven
productivity gains

Growth Stagnant

0.1% GDP growth without AI infrastructure in H1 2025. Slowing to 1.5% (2025) and 1% (2026) per Morgan Stanley.

Dollar Weakness

Depreciated 11% in H1 2025 — largest 6-month decline in 50+ years. Inflation rising to 2.9%.

Consumer Crisis

70% of consumer spending drives US GDP ($19T/year). Top 10% accounts for 50%. Middle class income share down 33% over 5 decades. Homes 7× income (vs 4.4× in 1985).

Healthcare Last

Highest spend. Ranks LAST among top 10 high-income nations. College costs 3× higher: $10K (1989) → $29K (2024). Student debt: $1.7T.

The Cognitive Stack

What we have vs. what AGI needs

We are pre-AGI generalists — impressive breadth, but without stateful cognition: remember → learn → adapt.

✅ Strong Today

Production-Grade

Knowledge, reading/writing, math, speed, vision/audio — all at or above human level for well-specified tasks.

⚠️ Advancing but Brittle

Scaffolded

Reasoning with scaffolds, cross-modal working memory, basic tool-use planning. Works with guardrails; fails gracefully without them.

❌ Not Yet Adequate

Missing for AGI

Working memory at scale, long-term storage, long-horizon planning, autonomous learning, portable memory.

The Bridge to AGI — 6 Missing Capabilities

Long-Term Memory

Stable storage, precise retrieval, intentional updating

Working Memory at Scale

Reliable very long multimodal contexts without context rot

Advanced Perception

Hierarchical cross-modal abstractions from raw data

Advanced Reasoning

Long-horizon planning, self-critique, metacognition

Continual Learning

Learn while working, preserve prior skills without forgetting

World Models

Predictive world models with uncertainty quantification

The future is capability + compression: hierarchy over tokens (H-Nets), state over attention (Mamba-3), memory over logs (ReasoningBank + ACE), learning over labeling. Trade "more parameters" for "more actual intelligence."

AGI Levels — Where We Are Today
1ChatbotsConversational AI — fully achieved ✅
2ReasonersHuman-level problem solving — achieved ✅
3AgentsSystems that take actions — we are here ← EARLY STAGE
4InnovatorsAid in invention — emerging
5OrganizationsCan do the work of an entire organization

We are at early Level 3. The economic opportunity is massive even at current capabilities.

The Investment Thesis

The question is not whether AI will transform knowledge work.
It is who will capture the value.

$60T Knowledge Economy

Largely untouched by AI vs. $370B software market — the gap is the opportunity

AI Adoption Accelerating

Faster than any prior technology wave in history

Applied AI in Services

Healthcare, insurance, legal, accounting, finance = immediate, massive TAM

Outcome-Based Pricing

Captures 25–50% of value delivered vs. 10–15% for SaaS

Business Model + AI Alignment

Creates compounding moats that widen over time

METR Doubling Law

Task capability doubling every 4–7 months — exponential unlock ahead

"For a century, America's growth machine converted human calories into output. The AI era lets us convert compute into output, at scale and on demand. That does not sideline people — it elevates them to where humans are scarce and priceless: complex judgment, creativity, connection."

Applied AI · Knowledge Work · Intelligence Revolution