Summary

AI turns software from a passive filing cabinet into an active labor substitute, opening a market potentially far larger than software itself. Alex Rampell traces a 65-year progression from on-premise databases to cloud systems of record and financial-services-enabled vertical SaaS; agents can now perform the work those systems merely recorded. The new formula is “Input, Coffee, Output, Code.”
The addressable budget shifts from software spend toward wages: US registered nurses alone represent more than $600 billion annually, versus under $600 billion for the entire worldwide software market. AI cannot perform CPR, but it can call patients before a colonoscopy, converse in 45 languages, and absorb work hospitals cannot staff. The operative question is how far customers let “their software budget bleed into their labor budget.”
Per-seat incumbents face a brutal cannibalization choice: lose most revenue as AI reduces seats, or reprice around outcomes and potentially grow 10×. Rampell’s Zendesk example pairs roughly $1.4 million of annual software spend with $50 million of support labor; copilots could cut 1,000 seats to 100, while autopilot could eliminate them. For Salesforce and Zendesk, AI is “both defense and offense.”
The strongest startup wedge is the “messy inbox problem”: automating judgment-intensive work between unstructured inputs and legacy systems of record. David Haber’s healthcare example, Tennr, trained against, he thinks, roughly 4 million documents and cut patient-intake administration costs about 90%, then began eating into scheduling, eligibility, and benefits. The AI capability may commoditize, so durability still comes from owning workflows, integrations, network effects, and the eventual system of record — “moats still matter.”
Previously uninvestable niches become venture-scale when software captures labor budgets or bundles labor with a 10×-better replacement system. Compliance is the model: it is reportedly America’s fourth-fastest-growing job, often runs on Excel, and remains chronically understaffed. David Haber describes AI agents clearing tens of thousands of alerts while helping introduce a better transaction-monitoring system, in the context of TD Bank’s $4 billion fine related to transaction monitoring.
AI may automate routine white-collar tasks while making scarce human interaction more valuable. The panel expects at least every white-collar job to gain a copilot, with some roles fully agentic; Rampell’s extreme formulation is that people may either “tell a computer what to do” or be “told by a computer what to do.” Yet once automated outreach becomes ubiquitous, relationships built face-to-face — even “over golf” — may command a premium.
Business fundamentals do not change, but falling costs expand both market size and competitive risk. Investors still need retention, gross profit, overhead discipline, and the “present value of future profits”; meanwhile, AI makes software easier to build and pushes prices inexorably downward. The most attractive hunting grounds are obscure industries where domain experts understand a workflow, the technology is already good enough, and 30-year-old systems can become “10× better.”

Deep dive

1. Software has graduated from storing work to performing it

Rampell’s historical frame begins with capital replacing brawn — steamships displacing synchronized rowers and looms mechanizing sewing — while white-collar work remained human. AI breaks that boundary: software engineers can now build agents that perform work done by end users “in 1960, 1970, 1990, 2000, 2010, 2023, 2024.”
Rampell said he thought SABRE, developed by American Airlines with IBM around 1959 or 1960, exemplified software’s first era: replace filing cabinets, erasers, and “gophers” with a database and front end. Quicken digitized financial files, PeopleSoft did HR, and email digitized mail — but the same 50-person HR department largely remained.
Cloud software was the second era: Salesforce moved the Rolodex online, NetSuite moved accounting online, and Zendesk moved support email online. Financial services then made small verticals economic; David Haber notes that roughly 80% of Toast’s revenue now comes from payments, insurance, and related services rather than the restaurant software itself.
Those eras created the necessary substrate. “All the data is here in the cloud, all the compute is in the cloud, and now you just kind of mix them together”; agents can act because decades of digitization already captured messages, records, customers, and workflows in accessible systems.

2. Labor budgets make the new opportunity radically larger

Rampell’s scale comparison: approximately 4.7 million US registered nurses earning a little over $120,000 produce a wage market above $600 billion annually. That single US profession exceeds his estimate of the worldwide software market, yet dedicated nurse software has historically had “probably zero” budget.
AI need not replace the whole nurse to reach that budget. It cannot be a phlebotomist or perform CPR, but it can call before a colonoscopy, explain fasting instructions, and converse in 45 languages — useful capacity when hospitals cannot hire enough nurses or need language-specific coverage.
In finance, Rampell imagines NetSuite acting on accounts receivable instead of merely displaying it. Five collections hires might cost $400,000 annually; NetSuite could hypothetically charge $2,000 annually for the capability, while his customer comparison uses $10,000 in software versus $400,000 in labor.

3. Seat-based incumbents must cannibalize themselves before agents do

Rampell’s Zendesk arithmetic makes the conflict explicit: 1,000 support seats at $115 per month generate about $1.4 million annually, while the employees may cost roughly $50 million. Zendesk wants seats to grow, but the larger prize lies in charging against the work those employees perform.
Copilot can be cannibalistic: if each representative moves from answering 10 questions daily to 100, the customer needs 100 people rather than 1,000 and Zendesk loses 90% of seat revenue. Autopilot goes further — route questions directly to the agent and “I need nobody,” leaving no conventional seats to sell.
The offense is correspondingly large. A provider capturing part of the $50 million labor pool could grow revenue 10×; mishandling the transition could erase most of it. Rampell applies the same fork to Salesforce: its $200 billion-plus scale and customer data are advantages, but it still faces declining human-seat revenue if it fails to adapt.
The pricing pushback — worth keeping: customers accustomed to 99-cent apps or fixed software budgets may resist a sudden increase even when it lowers total costs. Incumbents must move from per-seat pricing toward work or outcomes, yet many may not evolve quickly enough, creating a startup wedge.

4. The messy inbox opens the door, but conventional moats keep it open

Haber’s “messy inbox problem” describes administrators extracting information from emails, faxes, and calls, then entering it into an EMR, ERP, or CRM. That judgment-intensive work historically sat upstream of software; AI can now automate it, wedge into the workflow, and gradually become the AI-native system of record.
His concrete example is Tennr, which addresses specialist referrals still sent by fax. After training against what Haber thought was roughly 4 million healthcare documents, it can extract patient information programmatically and reduce pre-clinician intake administration costs by about 90%; from that entry point, it is beginning over time to eat into scheduling, eligibility, and benefits.
Haber distinguishes differentiation from defensibility: replacing a human workflow can be “a thousand times better” and feel magical, but the model capability itself may become commoditized. Protection comes from owning downstream workflows, integrating with every relevant system, becoming hard to remove, and adding familiar advantages such as network effects, platforms, and virality. “Moats still matter.”
Rampell’s alternative is to find labor-heavy categories with almost no incumbent software. Compliance officers — described as America’s fourth-fastest-growing job, behind manicurists at No. 1 — often rely on Excel, Word, and browsers. Agents could address staffing backlogs first, then potentially turn that wedge into a purpose-built system of record.

5. AI reopens failed theses and rewards aligned business models

David Haber says each technology shift forces investors to revisit ideas previously judged unworkable. Legacy financial systems often survived because replacements were only “2× better”; combining better software with scarce labor can make the offer 10× better and finally overcome resistance to ripping out a 30-year-old system.
Haber’s transaction-monitoring example ties the two together: in the context of TD Bank’s $4 billion fine related to transaction monitoring, banks may face old systems producing too many false alerts while still confronting tens of thousands of alerts they cannot staff. An agent bundle can clear the backlog while introducing a better monitoring platform, improving both the sales wedge and defensibility.
The investment metrics remain orthodox. Rampell still wants customers, retention, gross profit per customer, and manageable overhead because valuation is still “the present value of future profits.” Even the social-era “smile curve” — usage falls after installation, then recovers and plateaus around 50%, 70%, or 90% — remains useful; AI does not suspend economics.
Market size does change. Haber points to the US North American Industry Classification System, or NAICS, which has roughly 600 industry categories: a niche with 1,000 buyers paying $1,000 monthly was previously framed as only a $12 million market and unattractive for venture backing. Once agents tap the industry’s labor budget, the same narrow vertical can become materially larger.

6. Deflation expands demand, while obscure expertise identifies the winners

Haber sees a business-model fork in professional services. Hourly law firms may face revenue pressure if three hours of work becomes three seconds, prompting pitches for full-stack AI-native competitors; contingency-based plaintiff firms are aligned with productivity because better intake lets them accept more valuable cases rather than merely bill fewer hours.
In his plaintiff-law example, firms accept roughly one case per 100 leads. AI can evaluate medical and employment records, draft chronologies and demand letters, file a complaint, and walk through pre-litigation and litigation, enabling lawyers to handle 3× or 4× as many cases; the software cost may be passed through as a familiar technology expense.
Rampell’s qualified call is that well-executed technology is deflationary: he “can’t see a scenario” where agent prices exceed human labor or stop falling significantly. Lower costs also create demand — a $2,000-an-hour trademark service becoming $5 could make filing ubiquitous, while nearly free translation makes even an ancient-Greek version conceivable.
The panel’s builder brief is deliberately unglamorous: seek founders with a decade of knowledge in farming, mining, insurance, financial services, or another obscure workflow. Autopilot is not ready everywhere, so timing matters; the opportunity is where present technology is already sufficient, legacy systems can become “10× better,” and incumbents cannot smoothly change either product or pricing.

Victor Riparbelli, CEO @Synthesia: OpenAI vs Anthropic vs X.ai - Who Wins and Why | E1246 Tal Zaks - Bridging Science, Medicine, and Returns - [Invest Like the Best, EP.406]