Labor Pyramid AI Opportunity Map
Map AI opportunities from costly knowledge work down to physical labor
- Difficulty
- Moderate
- Time to result
- ~weeks to results
- Steps
- 6
- Confidence
- 96%
View AI-generated tokens as economically useful units of labor, then map opportunities through a pyramid of occupations. Begin at the top with highly paid, highly skilled knowledge workers such as engineers, doctors, and lawyers because successful assistance in these roles creates immediate, measurable value. Identify products that operate as copilots or autonomous agents within their workflows. As model capability and customer trust improve, move into the next compensation tier, including financial analysts, salespeople, and nurses. The long-term progression reaches physical labor, where robotics can perform dangerous or backbreaking work. This ordering combines market size, willingness to pay, technical feasibility, and expected timing. It helps teams avoid treating AI as one undifferentiated market and instead choose an occupational wedge appropriate to current capabilities.
Origin
Extracted from Masters in Business, where Hamid explains how Kleiner Perkins mapped AI investments after the emergence of ChatGPT.
Core principles
- 01Treat model output as units of labor, not merely software functionality
- 02Start where labor is highly paid and economically valuable
- 03Match AI capabilities to specific occupational workflows
- 04Move down the labor pyramid as technology becomes more autonomous
- 05Expect physical automation to mature later than knowledge-work tools
How to run it
- 1
Quantify the labor opportunity
Estimate the labor component of the economy and identify the portions represented by knowledge work and physical work. Use labor spending as the broad opportunity pool.
Pro tip Express AI value in labor hours or workflow output as well as software-seat revenue.
Watch out Do not assume the entire labor pool is immediately addressable.
- 2
Rank occupational tiers
Organize jobs by compensation, skill requirements, and economic value. Place highly paid engineers, doctors, and lawyers near the top and physical labor near the bottom.
Pro tip Use real compensation and workflow data rather than prestige-based rankings.
Watch out High compensation alone does not prove that a workflow is technically automatable.
- 3
Find high-value workflows
Within the top tier, locate repeatable tasks where AI can increase throughput, accuracy, or access. Determine whether the immediate product should assist a professional or perform a bounded task autonomously.
Pro tip Begin with workflows that already consume substantial professional time.
Watch out Avoid trying to replace an entire profession with the first product version.
- 4
Select an occupational wedge
Build or back a focused product for one professional category and integrate it into the way that category already works.
Pro tip Use domain-specific context and workflow integration as differentiation beyond the underlying model.
Watch out A generic model wrapper may not satisfy professional security, accuracy, or workflow requirements.
- 5
Move down the pyramid
As models improve, repeat the analysis for the next occupational tier, including finance, sales, and nursing. Expand only when capability and economics support the new category.
Pro tip Treat each tier as a fresh market with distinct buyers and constraints.
Watch out Do not assume success in one occupation transfers automatically to another.
- 6
Track the physical frontier
Monitor robotics and embodied AI for eventual access to physical-labor markets. Prioritize work that is dangerous, undesirable, or physically damaging.
Pro tip Use knowledge-work adoption as an earlier signal, not as proof that robotics is ready.
Watch out Physical automation faces hardware, safety, and deployment constraints absent from software.
In the wild
Kleiner Perkins mapped the highest-paid occupations to engineering, medicine, and law, then invested in focused products including Harvey for legal work, Windsurf for software development, and Ambience and OpenEvidence for medicine.
→ The firm built an initial AI portfolio around expensive knowledge work where augmentation could create immediate value.
After addressing doctors, lawyers, and engineers, the firm moved to financial analysts, salespeople, and nurses through investments such as Rogo, Nooks, Revo, and Hippocratic.
→ The occupational map supplied a repeatable sequence for expanding the firm's AI investment search.
Common mistakes
Treating AI as ordinary SaaS
Model output can perform economically valuable work, so evaluating it only as another software tool understates its addressable market and pricing logic.
Starting with whole-job replacement
A focused copilot or bounded agent is often more feasible and adoptable than attempting to automate an entire profession immediately.
Ignoring technical timing
Knowledge-work automation and physical robotics mature on different schedules, so descending the pyramid too quickly can produce premature investments.
Is it for you?
Best for
It is best for investors and founders deciding which labor-intensive market AI can address now and which opportunities require more technical maturity.
Not ideal for
It is not ideal for products whose value cannot be linked to a recognizable job, workflow, or labor cost.
From the transcript
“if you look at tokens and what the frontier model companies provide is it is units of labor”
“we thought about like the labor pyramid, think about knowledge work and we went straight to okay, what is what are the most highly skilled,…”
“we've sort of worked our way down the labor pyramid.”
From the episode
Venture Capital During the AI Revolution with Mamoon Hamid
Mamoon Hamid