E383:Why the Next Fortune 500 Companies Will Be Built on AI

Episode Summary

Executive Summary: The conversation argues that agentic AI is early but already reshaping work, especially B2B processes where waste is highest. The speakers see AI as a productivity unlock that will raise wages, create new jobs, and expand markets through better tools, data flywheels, and new infrastructure like agent identity, audit automation, and credit data for private businesses.

Main Topics: Why agentic AI is arriving now (Priority: 5/5): AI agents are seen as a breakthrough because they automate repetitive, low-value work like copying, logging in, compliance steps, and transfers, freeing workers for higher-value tasks. The speakers believe the current moment is the “first inning” of a decade-long transformation. Productivity, wages, and historical analogy (Priority: 5/5): The discussion frames AI like past labor-saving revolutions in agriculture, washing machines, and sewing machines: fewer people do routine work, but productivity rises, costs fall, and wages can increase when workers become more capable. Why B2B agentic AI is the main opportunity (Priority: 5/5): The speakers focus on businesses because that is where huge inefficiencies exist and where AI can quickly improve throughput, quality control, and decision-making. They argue B2B workflows are structured, measurable, and ripe for automation. New infrastructure: identity, audit, and credit (Priority: 5/5): Agentic commerce will require new layers such as verifying agents’ identities (KYA), automating audits, and building data systems that let private companies prove trustworthiness and access credit. Data flywheels and defensible moats (Priority: 4/5): Companies that combine proprietary data, workflow expertise, and reinforcement learning can build compounding advantages. The speakers argue this matters more than generic LLMs, which they compare to utility providers like AWS or Azure. Market structure and venture implications (Priority: 3/5): AI is accelerating winner-take-more dynamics in venture and enterprise, but the speakers also believe small, specialized firms can thrive near universities and in vertical niches where relationships and domain expertise matter. Human nature, abundance, and social change (Priority: 3/5): The conversation contrasts fast technology change with slow-changing human behavior. The speakers expect both abundance and disruption, but argue that people remain status-seeking, relationship-driven, and motivated by meaningful work.

Key Arguments: Agentic AI is now practical enough to automate mundane work at scale, and it is still at an early stage. The biggest near-term value is in B2B workflows, where waste, friction, and manual inspection are concentrated. AI should increase productivity enough to support wage growth, since wages rise when workers produce more value. Historical precedents show that labor displacement can coexist with massive gains in output, safety, and affordability. New categories of work will emerge around AI deployment, data management, compliance, and agent orchestration. Vertical AI companies with proprietary data and specialized process knowledge will be more durable than generic model providers in many markets. Agent identity verification (know your agent) and daily automated audit systems will be foundational to agentic commerce. Private company financial data, if structured and shared selectively, could unlock new credit markets and reduce counterparty risk. AI systems improve through data flywheels: user interaction produces more data, which improves the model, which attracts more users. Large foundation models may become utilities, while value accrues to domain-specific applications and data-rich workflows.

Data Points: US labor in agriculture today: about 1% to 2% - Used as a historical comparison to show how automation reduced agricultural employment while increasing output and safety. US labor in agriculture 250 years ago: 95% of Americans - Illustrates how entire sectors can shrink in employment without reducing total production. People globally with zero electricity: 700 million - Example of unmet basic needs showing room for technological progress and development. People globally with less than 4 hours/day of electricity: 4 billion - Used to emphasize global infrastructure gaps and the potential for productivity gains. Average white-collar worker cost: $170,000/year - Estimated employer cost for knowledge workers across banks, law, consulting, marketing, and executives. Wages and benefits portion of worker cost: $160,000 - Part of the total cost of an average white-collar worker. Software spend per white-collar worker: $10,000/year - Estimated spend on software tools such as Microsoft, Google, Oracle, and AWS. Software industry revenue: $1.5T to $1.7T annually - Reference point for the current enterprise software market size. Software industry market cap: about $17T - Illustrates the scale of the existing software sector. Work hours McKinsey expects agentic AI to disrupt: 57% - Used to argue that a large share of work can be transformed by AI agents within a decade. Spend potentially transformed from wages to agentic software: about $100,000 per worker - Derived from 57% of the $160,000 wages and benefits portion. Medical device output before AI inspection: 10,000 units/day - Baseline production level before smart camera and AI quality-control improvements. Medical device output after AI inspection: 166,000 units/day - Example showing a 16.5x increase in production from computer-vision automation. Increase in medical device production: 16.5x - Highlights the scale of productivity improvement from AI-powered inspection. Defect detection rate before: 94% - Prior quality-control performance in the manufacturing example. Defect detection rate after: 100% - Claimed improvement with AI inspection and smart cameras. ChatGPT adoption: 100 million users - Cited as evidence of unprecedented speed of AI adoption. Savings at Citrin Cooperman from one audit agent: 10% of spending - Initial production savings from automating one audit quality-control process. Projected savings after second audit step: 25% of spending - Expected next-stage savings for the audit workflow example. Audit spending example: $1.2M saved out of $12M - Breakdown of first-phase savings in audit quality control. LP meeting timing: yesterday in New York - Context for the Overview manufacturing example being discussed as a live portfolio case. Private company data example: opt-in benchmarking and blind sharing - Mechanism proposed for benchmarking spending versus peers while protecting confidentiality. VC concentration: 75% of venture capital to top five firms - Used to describe power-law dynamics in fundraising and capital allocation. Addepar-tracked AUM in report: $1.5T - Referenced to discuss how institutional capital is allocated across asset classes. Addepar-tracked VC allocation: 2.8% - Share of tracked AUM in venture capital, surprising the speaker. Private equity allocation vs VC: roughly double VC - Used to show PE was a larger allocation than venture in the cited report.

Pivotal Quotes: "“Most likely, to say a cliche, you're more likely to lose a job not to an AI, but to someone else who is using an AI when you're not.”" — David: Argues that AI adoption, not AI itself, will be the main competitive pressure on workers. "“The future of Agentic AI is about understanding process.”" — David: Explains why vertical, workflow-specific AI systems may outperform generic foundation models in many enterprise settings. "“Know your agent.”" — David: Introduces the idea that agent identity verification will become a necessary layer in agentic commerce, analogous to KYC.

Implications: Expect rapid enterprise adoption of AI agents in structured workflows, with major demand for identity, audit, and data infrastructure. The biggest winners may be vertical companies with proprietary data, not just model providers.