The Physical Work Iceberg Index: Measuring the $1.08 Trillion Gap in AI Workforce Exposure
MIT's Iceberg Index measured $1.1 trillion in US wage value exposed to cognitive AI automation. They explicitly excluded physical work as 'future work.' This essay fills that gap. Using a three-filter framework - voice-deliverable, vision-verifiable, knowledge-capturable - applied to O*NET and ESCO skill taxonomies across 16 skilled trade occupations in four regions, the analysis identifies $1.08 trillion in global wage value that is coachable via real-time AI voice+vision guidance through AR smart glasses. Across 33.5 million target workers, 64% of occupational skills pass the AR-coachability threshold. This translates to a SaaS TAM of $11B-$32B at 1-3% of wage value, with a realistic five-year obtainable market of $30M-$60M. The physical iceberg is as large as the cognitive one. Coffee AI is building directly into the gap MIT left open.
Introduction
In October 2025, MIT published the Iceberg Index. They mapped 13,000+ AI tools to 32,000+ skills across 923 occupations. The finding: 11.7% of the $9.4 trillion US labour market - $1.1 trillion - is technically exposed to cognitive AI automation. The "surface" (visible tech adoption) covers 2.2%. The mass beneath it is five times larger.
Three sentences from the paper matter here:
"Physical automation through robotics is excluded here but will become increasingly relevant as capabilities mature."
"Digital AI focus: Analysis covers cognitive and administrative automation where deployment is observable. Physical robotics excluded due to immature adoption data."
"Future work will model adoption dynamics, extend to physical automation as robotics mature, and integrate task-level quality benchmarks."
MIT measured the cognitive iceberg. They left the physical one unmeasured. This essay measures it.
The key insight: the right frame isn't physical automation (replacing workers with robots). It's physical augmentation (coaching workers through AR glasses in real time). The human stays. The AI guides. The technology stack for this - voice synthesis, computer vision, expert knowledge capture - already exists. Affordable, lightweight smart glasses arrived in 2025. The physical iceberg is ready to surface.
The ACE Framework
MIT asked: "Can an AI tool perform this skill?" We ask a different question: "Can an AI coach a human through this skill via voice and vision?"
We call the answer AR Coaching Exposure (ACE). It's a three-dimensional score applied to every skill in a target occupation, using the same O*NET taxonomy MIT used.
Three dimensions, each scored 0-2:
| Dimension | 0 (No) | 1 (Partial) | 2 (Full) |
|---|---|---|---|
| Voice (V) | Needs dense diagrams, multi-sensory input | Partially audio-guidable | Fully step-by-step audio |
| Vision (Vis) | Invisible to camera | Partially visible | Fully verifiable |
| Knowledge (K) | Decades of inarticulate pattern-matching | Partially capturable via demo | Fully capturable via walkthrough |
Threshold: A skill scoring 4+ out of 6 is classified as AR-coachable. This requires at least two dimensions at maximum or all three at partial-or-above. The intent is to capture skills where AR coaching is meaningfully useful, not merely technically possible.
Aggregation follows MIT's approach. Per-occupation ACE is the importance-weighted fraction of AR-coachable skills. AR-coachable wage value is employment multiplied by median wage multiplied by ACE.
Skills that consistently fail the threshold: interpersonal (customer negotiation, conflict resolution), abstract cognitive (creative problem-solving without observable context), and proprioceptive (tactile feedback that can't be visually verified - feeling pressure in a pipe fitting, sensing vibration frequency by touch).
Results: The Physical Iceberg
Global Summary
| Region | Workers | Wage Value | ACE | Coachable Value | Data |
|---|---|---|---|---|---|
| United States | 7.16M | $402B | 64% | $254B | High |
| United Kingdom | 1.19M | $52B | 64% | $33B | High |
| EU-27 | ~15M | $540B | 64% | $346B | Medium |
| APAC key 4 | ~10.1M | $383B | 64% | $245B | Medium |
| Broader APAC | ~62M | $315B | 64% | $202B | Low |
| Global Total | ~95.5M | $1.69T | 64% | $1.08T | Mixed |
APAC key 4: Japan (5.5M workers, $187B), Australia (1.97M, $116B), South Korea (2.2M, $69B), Singapore (450K, $11B).
The physical iceberg - $1.08 trillion - matches the scale of MIT's cognitive iceberg ($1.1 trillion). These aren't competing exposures. A worker's cognitive tasks may be automatable while their physical tasks are coachable. The total AI surface area of the labour market is larger than either index alone suggests.
Which Trades Score Highest
| # | Occupation | Workers | Wage | ACE | Coachable Value |
|---|---|---|---|---|---|
| 1 | Aviation MRO | 160,800 | $79,140 | 78% | $9.9B |
| 2 | Industrial Machinery | 538,300 | $63,510 | 72% | $24.6B |
| 3 | HVAC Technicians | 425,200 | $59,810 | 68% | $17.3B |
| 4 | Plumbers/Pipefitters | 504,500 | $62,970 | 66% | $21.0B |
| 5 | Carpenters | 959,000 | $59,310 | 65% | $37.0B |
| 6 | Heavy Equip. Mechanics | 200,000 | $62,740 | 65% | $8.2B |
| 7 | Electricians | 818,700 | $62,350 | 64% | $32.7B |
| 8 | Flooring Installers | 112,300 | $52,000 | 63% | $3.7B |
| 9 | Automotive Technicians | 805,600 | $49,670 | 62% | $24.8B |
| 10 | Insulation Workers | 67,400 | $50,730 | 62% | $2.1B |
| 11 | Masonry Workers | 75,000 | $56,600 | 60% | $2.5B |
| 12 | Drywall/Ceiling Tile | 118,700 | $58,800 | 60% | $4.2B |
| 13 | Gen. Maintenance | 1,500,000 | $48,620 | 58% | $42.3B |
| 14 | Roofers | 166,700 | $50,970 | 55% | $4.7B |
| 15 | Welders/Cutters | 457,300 | $51,000 | 55% | $12.8B |
| 16 | Painters (Construction) | 250,000 | $48,660 | 52% | $6.3B |
Aviation MRO tops the list because the work is intensely procedural, FAA/EASA-regulated, and inspection is visual by nature. Painters score lowest because quality assessment is subjective and finish quality is hard to verify via camera.
The highest ACE score isn't the highest value. Electricians generate $32.7B in US AR-coachable wage value despite ranking 7th in ACE, because the occupation employs 818,700 workers at $62,350 median. The three largest pools: general maintenance ($42.3B), carpenters ($37.0B), electricians ($32.7B). These are the volume plays. Aviation MRO is a premium niche with high willingness to pay.
Framing Against MIT
| Index | Wage Value | % US Market | Method |
|---|---|---|---|
| MIT Surface Index | $211B | 2.2% | AI tool deployment |
| MIT Iceberg Index | $1.1T | 11.7% | Skills-AI overlap |
| Physical Iceberg (US) | $254B | 2.7% | ACE scoring |
| Physical Iceberg (global) | $1.08T | - | ACE scoring |
The Physical Work Iceberg adds 2.7 percentage points to MIT's 11.7%, bringing total AI-addressable US wage value to ~14.4%.
MIT measured what digital AI tools can automate. This index measures what AI can coach. These are complementary. MIT's methodology structurally cannot reach these occupations - they mapped 13,000+ production-ready digital AI tools, none of which perform physical work.
The Market
TAM, SAM, SOM
| Metric | Wage Value | SaaS (at 2%) | Range (1-3%) |
|---|---|---|---|
| TAM (global) | $1.08T | $21.6B | $10.8B-$32.4B |
| SAM (US+UK+EU+APAC4) | $878B | $17.6B | $8.8B-$26.3B |
| SOM (5yr, US+UK) | $290B | $30M-$60M | Conservative |
The 2% wage-to-SaaS midpoint is validated by ServiceTitan's pricing: $250-500/month per technician against median wages of $56,000-$63,000 (BLS OEWS May 2024) [2]. Construction IT spend sits below 1% of revenue today versus a 3-5% cross-industry average (McKinsey, 2023; Gartner) [3]. An AI coaching tool at $100/month ($1,200/year) represents 2% of a $60,000 wage - comparable to Housecall Pro ($49-109/month), well below ServiceTitan.
Five-year SOM assumes 0.2-0.3% penetration of SAM SaaS spend. ServiceTitan reached ~1% of 900,000 target businesses in roughly 10 years (Bessemer, 2022) [4]. Toast reached 10% of 1.4 million restaurant locations in ~12 years. The SOM of $30M-$60M implies 25,000-50,000 paid seats at $1,200/year - 0.3-0.6% of 8.35 million US+UK target workers.
Labour Shortages Make This Urgent
The labour market dynamics are inverted relative to cognitive AI. Cognitive AI exposure creates displacement anxiety. Physical AR coaching creates demand from employers who can't find workers.
US: Electricians project 81,000 annual openings. Industrial machinery mechanics project 13% growth over 2024-2034 (BLS) [2].
UK: 140,000 unfilled construction vacancies. 74% of employers cite skill shortages as the primary barrier (CITB CWO 2025-29) [5]. The workforce is ageing: 35% over 50, only 20% under 30. Only 21% of firms employ an apprentice. The UK needs 47,860 additional construction workers per year through 2029. The government committed £600M for construction skills in March 2025 [6].
Japan: Construction employment fell from 6.85M to 4.83M since 1997. Stricter overtime regulations (April 2024) compound the shortage [7].
Australia: Median technician/trades pay is AUD $91,000 - among the highest globally. Strong SaaS adoption precedent (ServiceM8, Tradify) [8].
In this context, AI coaching is not a threat to workers. It's a response to the absence of workers.
Why Now: Three Barriers Just Fell
1. Hardware costs dropped 90%. RealWear Navigator 520: $2,900, 274g. Microsoft HoloLens 2: $3,500, 566g. Sub-$500 lightweight alternatives (2025): under 50g. The deployment friction that killed every previous "AR for field workers" play just disappeared.
2. VLMs crossed the capability threshold. Vision-language models can now process camera frames and generate coaching responses in single-digit seconds. This wasn't possible 18 months ago. Unit economics are viable at scale.
3. Construction IT spend has nowhere to go but up. Less than 1% of revenue today. Cross-industry average is 3-5%. Every other vertical already made this transition. Trades are next [3].
What Coffee AI Is Building
Coffee AI sits directly in the gap MIT left open. The system captures expert knowledge from experienced tradespeople and delivers real-time voice coaching to junior workers through AR smart glasses. An expert records a procedure once. Every worker after that gets guided through it, step by step, with the AI verifying correct execution via the camera.
No competitor ships real-time AI coaching for physical work today. Existing players offer static digital work instructions (Tulip, Vuforia, Taqtile) or remote human expert video calls (Librestream, XMReality). Coffee AI closes the loop: expert knowledge in, real-time voice coaching out, through lightweight AR glasses.
Go-To-Market Considerations
The following are early-stage considerations, not executive decisions. Segment selection and pricing are TBD and will be validated through pilot deployments.
The ACE data highlights several candidate segments - including HVAC, industrial machinery, electricians, and aviation MRO - each with different tradeoffs between coaching intensity, workforce size, and employer concentration.
If you're an investor or potential partner interested in the go-to-market strategy, unit economics, or technical architecture, please reach out to Abin using the contact details at the top of this essay.
Limitations
This is an analyst framework, not a calibrated model.
Judgment-based scoring. The ACE scores are assessments, not outputs of a simulation. MIT ran Large Population Models on the Frontier supercomputer. This extension performs static scoring. Different analysts applying the same rubric might produce ACE scores varying by 5-10 percentage points.
Geographic data asymmetry. US data (BLS/O*NET) is granular to the individual skill level. UK data (ONS/CITB) is solid at occupation level. EU and APAC figures involve interpolation. Broader APAC (China, India, SEA) is rough.
Hardware dependency. Current affordable AR glasses are camera-only - no heads-up display. Voice-only coaching works without a display. AR overlays will require display-capable hardware as the category matures. The ACE scores assume reliable AR hardware exists for harsh field conditions. That's unproven at scale.
Self-employed segment. 35-40% of UK electricians are self-employed (ONS). Per-seat SaaS pricing optimised for enterprise may need adaptation for owner-operators.
The wage-to-SaaS conversion rate is assumed, not derived. The 2% midpoint draws on ServiceTitan pricing and cross-industry SaaS benchmarks, but Coffee AI is a novel category. Actual willingness to pay needs empirical validation.
These limitations are real. The headline number ($1.08T) represents the total addressable opportunity under the framework's assumptions, not a forecast.
Conclusion
MIT found a $1.1 trillion cognitive iceberg. They explicitly excluded physical work. The Physical Work Iceberg Index fills that gap: $1.08 trillion in global wage value is AR-coachable across 33.5 million skilled trade workers.
The economics are different from cognitive AI. These workers earn $49,000-$79,000 - expensive enough that productivity gains justify SaaS pricing, not so expensive that automation ROI makes replacement attractive. The labour shortages are acute across every target geography. The hardware just got cheap. The AI just got fast enough.
The physical iceberg has existed for decades. What changed is that addressing it became technically and economically viable in 2025-2026. Coffee AI is building directly into this space.
References
[1] A. Chopra et al., "The Iceberg Index: Measuring Skills-centered Exposure in the AI Economy," arXiv:2510.25137, Oct. 2025. [Online]. Available: https://arxiv.org/abs/2510.25137
[2] U.S. Bureau of Labor Statistics. "Occupational Employment and Wage Statistics, May 2024." [Online]. Available: https://www.bls.gov/oes/
[3] McKinsey & Company. "From Start-up to Scale-up: Accelerating Growth in Construction," 2023. [Online]. Available: https://www.mckinsey.com/
[4] Bessemer Venture Partners. "State of the Cloud 2022: ServiceTitan Case Study." [Online]. Available: https://www.bvp.com/atlas/state-of-the-cloud-2022
[5] CITB. "Construction Workforce Outlook 2025-29." [Online]. Available: https://www.citb.co.uk/cwo/index.html
[6] Centre for Social Justice. "Skills to Build: Fixing Britain's Construction Workforce Crisis," Nov. 2025. [Online]. Available: https://www.centreforsocialjustice.org.uk/wp-content/uploads/2025/11/CSJ-Skills_to_Build.pdf
[7] Statistics Bureau of Japan. "Labour Force Survey 2023." [Online]. Available: https://www.e-stat.go.jp/
[8] Australian Bureau of Statistics. "Employee Earnings, August 2024." [Online]. Available: https://www.abs.gov.au/
[9] ServiceTitan. "SEC S-1 Filing," Dec. 2024. [Online]. Available: https://www.sec.gov/cgi-bin/browse-edgar?action=getcompany&company=servicetitan
[10] ONET OnLine. National Center for ONET Development. [Online]. Available: https://www.onetonline.org/
[11] Office for National Statistics. "Annual Survey of Hours and Earnings 2024/2025." [Online]. Available: https://www.ons.gov.uk/
[12] European Skills, Competences, Qualifications and Occupations (ESCO). [Online]. Available: https://esco.ec.europa.eu/
[13] Skills England. "The UK Standard Skills Classification," Nov. 2025. [Online]. Available: https://www.gov.uk/government/publications/uk-standard-skills-classification-interim-development-report