By playing, you agree to uListen's Terms and Privacy Policy and YouTube's Terms of Service.
a16zHow AI is Reshaping Labor Markets: A $Trillion-Dollar Opportunity Explained
Summaries, chapters, transcripts, and FAQs on this page are made by uListen, not YouTube.
CHAPTERS
- 0:00 – 4:28
From Brawn to Brains: Software's First Eras
Steph Smith asks whether turning capital into labor is new. Alex Rampell argues past technology augmented physical labor, then walks through the first era of software as digitized filing cabinets, from Sabre to Quicken and PeopleSoft.
- •Earlier technology like the steamship and loom replaced brawn, not white-collar brains
- •Sabre, built with IBM around 1959-60, digitized the airline filing cabinet
- •Round one of software was a database with a front end
- •The 50 HR people stayed 50 HR people even after the filing cabinet and its gopher went away
- 4:28 – 7:58
The Cloud Era and Financial Services Bundling
Rampell describes round two, when Salesforce, NetSuite and Zendesk moved filing cabinets into the cloud. Bundling payments and other financial services then made software viable for restaurants and HVAC contractors, as with Toast and ServiceTitan.
- •Cloud software ran roughly from 1998 to 2010
- •Salesforce, NetSuite and Zendesk put the Rolodex, accounting and email support in the cloud
- •Payments bundling made the market big enough for restaurant software such as Toast
- •The same human teams persisted from 1960 to 2024
- •AI now allows actions on the filing cabinet, such as Workday handling a dependent change
- 7:58 – 12:02
Wages vs. Software: The Scale of the Labor Opportunity
Steph asks how scale differs in this new era. Rampell compares wage markets to software markets, using nurses and NetSuite collections as examples, and raises the question of whether customers will accept paying far more for software.
- •About 4.7 million U.S. nurses earn over $600 billion in wages a year, roughly the worldwide software market
- •A software product could do part of a nurse's work, such as multilingual calls before a colonoscopy
- •A collections agent costing $80,000 a year could be replaced by software at $2,000 a year
- •It is an open question whether customers will see this as software budget bloat or labor savings
- 12:02 – 15:16
Why Systems of Record Set Up the AI Era
Rampell explains that the prior eras of digitization and the cloud made AI agents possible, because data and compute now sit together. Angela Strange adds that Toast now earns 80% of revenue from financial services and that AI could expand software revenue 2X to 10X.
- •AI in 1960 would not have worked without digitized data
- •Systems of record hold the detail of running a business, from laundromats to restaurants
- •Data and compute both being in the cloud lets them be combined
- •Toast and Mindbody show vertical software making most revenue from financial services
- •AI could let software budgets absorb labor budgets
- 15:16 – 21:15
Pricing Disruption: Zendesk, Copilots and Autopilots
David Haber raises how per-seat pricing may need to change, and Rampell works through Zendesk's numbers. He explains how copilot and autopilot tools could shrink seat revenue, while offense could multiply revenue for incumbents like Salesforce.
- •Zendesk charged $115 per seat per month, about $1.4 million a year for 1,000 seats, against roughly $50 million in labor
- •Copilot tools could cut needed seats by 90%, and autopilot could cut them to zero
- •Incumbents face defense and offense, with possible 10X upside
- •Systems of record give incumbents some moat
- •Rampell notes that farmers displaced by tractors moved into other productive work
- 21:15 – 23:36
The Messy Inbox Wedge and Tennr
The discussion turns to how startups can enter against incumbents. Haber describes his "Messy Inbox Problem" wedge strategy, in which software takes over judgment-intensive intake work and expands downstream, with Tennr in healthcare as the example.
- •Humans extract information from emails, faxes and calls into systems like EMRs, ERPs and CRMs
- •The wedge starts upstream and eats away downstream workflows
- •Tennr trained a model on about 4 million healthcare documents for patient referrals
- •Tennr cuts about 90% of intake admin costs and is expanding into scheduling and eligibility
- 23:36 – 25:45
Differentiation vs. Defensibility
Steph asks whether proprietary models or data provide defensibility. Haber distinguishes differentiation from defensibility and argues that traditional moats still matter, a point Rampell supports with the history of software starting where budgets were large.
- •AI gives strong differentiation, but a wedge product alone is not defensible
- •Model advantages are ephemeral and capabilities will commoditize
- •Defensibility comes from owning end-to-end workflows and becoming the system of record
- •Network effects, platform status and deep integration remain the moats
- •Software started with airlines because the budgets justified it
- 25:45 – 29:33
Untapped Labor Budgets: Compliance Officers and Bundles
Rampell argues for finding areas with big labor budgets but no incumbent software, using compliance officers as the example. Angela Strange adds that theses once thought unworkable may now work, citing transaction monitoring and a labor-plus-software bundle.
- •Compliance officer is the fourth fastest-growing U.S. job per the Bureau of Labor Statistics, and banks use Excel
- •Opening a business bank account can take a month because compliance is backlogged
- •Strange says past replacements were 2X better, not 10X better, and AI changes that
- •A labor-plus-software bundle helps sales and defensibility, as in the TD transaction monitoring fine example
- 29:33 – 33:12
New Jobs, Changing Roles and the Value of Human Connection
The panel discusses what happens to jobs. Rampell says such shifts are hard to predict but expects human-only skills like relationship building to rise in value as AI-driven outreach saturates. Strange expects every white-collar job to have a copilot.
- •Past shifts such as farmers after the tractor were hard to foresee
- •AI cannot build a relationship over golf, so in-person skills may grow in value
- •Rampell compares AI sales to telemarketing, mail catalogs and junk faxes, which got crowded
- •Strange expects every white-collar job to have a copilot and some to be fully agentic
- •Removing menial tasks could free people for creative and human-facing work
- 33:12 – 36:47
Do Metrics Change in the AI Era?
Steph asks whether new metrics matter. Rampell says the fundamentals are unchanged, explains the retention smile curve, and notes AI makes software easier to build. Strange adds that potential market size has changed through NAICS-coded industries.
- •Rampell says valuation is still the present value of future profits: customers, retention, gross profit and overhead
- •The smile curve describes products like Uber and Facebook whose usage drops, then recovers and plateaus
- •Most AI products monetize via subscription, so monetization is clearer than in the social media era
- •Strange says niche NAICS-coded markets now become venture-scale once labor budgets are addressable
- 36:47 – 39:41
AI-Native Professional Services and Legal
Haber contrasts selling software into incumbents with building full-stack AI-native businesses. He uses law firms as the example, explaining how hourly billing conflicts with AI efficiency while contingency-based plaintiff firms stand to benefit.
- •Hourly billing is threatened if AI turns three hours into three seconds
- •Some founders pitch full-stack AI-native law firms with different cost structures
- •A plaintiff-law company helps qualify cases and draft documents, letting lawyers take 3X to 4X more cases
- •Contingency firms are aligned with efficiency, and software costs are passed to clients
- •Both selling in and building full-stack may succeed
- 39:41 – 42:26
Deflation, Falling Costs and Market Expansion
Steph asks whether AI labor products will be deflationary. Rampell says technology is always deflationary, which makes defensibility a worry, but falling costs also create new demand in areas like trademarks and translation.
- •Rampell says technology done right is deflationary through productivity gains
- •Many competitors can build similar tools easily
- •He cannot see prices ending up above human costs
- •Cheaper trademark filing and translation could expand markets that were previously too costly
- 42:26 – 45:07
Where Builders Should Focus Next
Steph asks where the panel wants to see more builders. Rampell favors obscure domains and founders with deep industry insight, Strange points to old financial services and insurance systems, and Haber to vertical wedges and AI-native horizontal software.
- •Rampell likes founders with years in obscure industries who understand AI's potential
- •The technology is not ready for autopilot in many complicated cases, so early failures are expected
- •Strange sees 30-plus-year-old financial services and insurance systems of record as ripe for 10X improvement
- •Haber expects more messy-inbox wedges in niche verticals
- •Horizontal AI-native software is possible but requires understanding the incumbent's likelihood of changing
