Deel’s growth overshadows SpyGate, Claude learned new skills, and the great “momentum is moat” debate | E2194

Episode Summary

Executive Summary: The episode argues that AI has pushed startups and venture capital into a peak-market phase: opportunity is enormous, but so are the incentives to fight over deal terms, strategy, and power. The hosts debate bottom-up AI TAM, the future of VC as an asset class, momentum versus moat, and how enterprise AI is improving retention. They also highlight Anthropic’s new Skills feature, Deal/Rippling scale, and how founders should adapt to a more competitive, higher-stakes market.

Main Topics: Peak-market chaos and changing startup behavior (Priority: 5/5): Jason opens by saying markets feel hot because people become more combative and irrational when lots of capital is circulating. He frames this as a peak-market phenomenon where founders, VCs, lawyers, and managers get chippy over non-essential issues. AI as a massive labor-and-software TAM (Priority: 5/5): The conversation shifts to a bottom-up model for estimating AI’s market size using working populations, SaaS spend, consumer AI usage, and labor replacement assumptions. The conclusion is that AI could be a trillion-dollar-plus opportunity and may create multiple trillion-dollar companies. Deal, Rippling, and HR-tech scale (Priority: 4/5): They discuss Deal’s $300M raise, its reported $1.2B ARR, and compare it to Rippling. The host argues both are elite SaaS-era companies likely to expand into AI and possibly become global HR and employer-of-record platforms. Venture capital as an asset class (Priority: 5/5): A clip from Sequoia’s Roelof Botha sparks a debate about whether VC really behaves like a true asset class. The hosts discuss power-law math, dilution, entry price, fund size, and why many VC firms may underperform public-market indexes. Momentum vs. moat in AI startups (Priority: 4/5): The show examines whether fast growth itself can serve as a moat. The hosts weigh contrasting views from Brian Kim, Liz Wessel, and Nikunj Kothari, then conclude that momentum can matter now, especially in AI, though retention and long-term mission still matter. Anthropic Skills and AI workflow collaboration (Priority: 4/5): Anthropic’s new Skills feature is presented as a lightweight way to give models company-specific instructions, style guides, and policies. The hosts connect it to broader enterprise workflow and collaboration ideas, including document-level coordination and operational guardrails. Measurement, incentives, and performance management (Priority: 3/5): Jason describes using software to track work patterns and redesign bonuses around effort and impact rather than salary percentage. He argues AI and visibility tools can help managers reward high performers faster and run leaner teams.

Key Arguments: Markets become more chaotic and adversarial when capital is abundant; the result is more fighting over terms and status rather than company building. AI’s economic opportunity is larger than traditional SaaS because it targets labor replacement and workflow automation across the entire economy. A bottom-up TAM for AI suggests a trillion-dollar-plus market even under conservative assumptions, with upside if AI spend per worker rises significantly. Deal and Rippling, despite public controversy, are strong businesses with enough scale and distribution to become long-term category leaders. VC returns are structurally constrained by entry prices, dilution, and power-law dynamics; most funds won’t beat public indexes unless they own exceptional companies at attractive terms. In AI, momentum may now function as a practical moat because it attracts capital, talent, and customers faster than slower-moving competitors can respond. Anthropic’s Skills feature could reduce token usage and make enterprise AI more useful by embedding persistent instructions, policies, and style rules. Enterprise AI products appear to be retaining customers better over time, suggesting the category is maturing from experimentation into durable usage. Managers should reward both effort and effectiveness more frequently, using data and short feedback loops rather than annual, salary-based bonus formulas.

Data Points: Deal annual run rate: $1.2 billion - Described as Deal’s September run rate after its Series E raise. Deal EBITDA margin: 15%–17% - Reported as adjusted profitability for Deal. Deal valuation multiple: ~14x ARR - The host estimates Deal at roughly 14x annual recurring revenue. Rippling ARR: $570 million - Cited as an earlier-2025 reported figure for Rippling. VC capital raised annually in the U.S.: ~$250 billion - Used in a back-of-the-envelope argument about whether venture can produce enough exits to justify inflows. Expected net IRR used in VC math: 12% - A hypothetical return assumption for venture funds in the Roelof Botha discussion. Required venture run-rate multiple: 3.7x - Estimated from $250B annual VC inflows over a seven-year horizon at 12% IRR. Implied annual company exit value needed: ~$1.5 trillion - Derived from the prior assumption that investors own two-thirds of each company. Figma valuation: ~$40 billion - Used as an example of how many very large exits would be needed annually. Anthropic run rate: ~$7 billion - Current revenue run rate mentioned for Anthropic. Anthropic projected year-end run rate: ~$9 billion - Forecast mentioned during the discussion. OpenAI run rate: ~$13 billion - Current revenue run rate mentioned for OpenAI. Developed nation population: 1.4 billion - Used in the bottom-up AI TAM model. Developing world population: 6.8 billion - Used in the bottom-up AI TAM model. AI spend per worker in developed markets: $3,000/year - Jason’s baseline assumption for current developed-market AI-related software spend. AI spend per worker in developing markets: $240/year - Baseline assumption in the TAM model, described as less than $1/day. Consumer/enterprise AI retention in 2022: ~50% - Ramp data on enterprise AI product retention by year-end. Consumer/enterprise AI retention in 2023: ~63% - Ramp data on enterprise AI product retention by year-end. Consumer/enterprise AI retention in 2024: ~80% - Ramp data on enterprise AI product retention by year-end. Launch accelerator investment volume: 100 investments/year - Jason says his program makes about 100 investments annually. Launch accelerator applications: 20,000 - Used to illustrate competitiveness of the accelerator funnel. Launch accelerator acceptance rate: 1 in 200 - Calculated from 20,000 applications and 100 investments.

Pivotal Quotes: "We are in or we're entering into peak market right now." — Jason: Opening framing about hot markets causing irrational and combative behavior. "I don't think venture is an asset class. It doesn't support the numbers." — Roelof Botha: Clip from the Jack Altman interview used to debate venture economics. "Momentum is not a moat." — Liz Wessel: Referenced as a rebuttal to the idea that fast growth alone protects a startup.

Implications: AI is broadening the startup opportunity set while also making capital and talent competition harsher. Founders who move fast and retain customers may win; VCs will need better entry prices or different strategies. Enterprise AI is likely becoming stickier, and more trillion-dollar outcomes may emerge sooner than expected.