What Ramp’s data tells us about AI, unemployment and more with CEO Eric Glyman | E2192

Key Concepts

• Information Asymmetry: A situation where one party in a transaction has more or better information than the other.
• Notional Value: The total value of an underlying asset in a derivatives trade, not the actual cash amount exchanged.
• Rare Earth Metals: A group of 17 chemical elements crucial for high-tech applications, with China dominating production.
• Global Casino (Crypto): A metaphor describing the unregulated, global nature of cryptocurrency markets, susceptible to manipulation.
• Insider Trading: The illegal practice of trading on the stock exchange to one's own advantage through having access to confidential information.
• Leverage (Financial): Using borrowed capital to increase potential returns, but also increasing risk.
• Market Brittleness: The susceptibility of financial markets to rapid and significant price changes in response to news or events.
• Interchange Revenue: Fees collected by card-issuing banks and payment networks from merchants for processing credit and debit card transactions.
• Price Intelligence: A RAMP feature allowing businesses to compare vendor quotes against aggregated market rates.
• Earned Media: Publicity gained through promotional efforts other than paid advertising, such as news coverage or mentions on podcasts.
• Static Team Size: A trend observed in large tech companies where employee headcount remains relatively constant despite revenue growth.
• Natural Rate of Unemployment: The theoretical lowest unemployment rate that an economy can sustain without causing inflation.
• Specialization vs. Generalization (AI Impact): The shift from a traditional economic model valuing deep human specialization to one where AI can provide broad, deep knowledge, potentially leveling the playing field for non-specialists.
• Large Language Models (LLMs): AI models trained on vast amounts of text data to understand, generate, and interact in human language.
• Tokens (AI): The basic units of text (words, subwords, characters) that LLMs process.
• AI Agents: AI models augmented with tools and the ability to take actions on behalf of a user, rather than just providing suggestions.
• Co-pilot vs. Agent: A co-pilot assists by suggesting actions, while an agent autonomously performs actions.
• Sarbanes-Oxley (SOX): A U.S. federal law that mandates certain practices in financial record keeping and reporting for public companies, influencing expense review policies.
• Handshake Agreement: An informal, non-binding agreement made verbally.
• SAFE (Simple Agreement for Future Equity): An investment contract used by startups to raise capital, providing investors with the right to future equity.
• Punk Rock (Founders): A term used to describe founders who challenge norms, break rules, and operate unconventionally.
• Magnanimous: Generous or forgiving, especially toward a rival or less powerful person.
• Mench: A Yiddish term for a person of integrity and honor.

Market Volatility, Allegations of Insider Trading, and Crypto Regulation

The discussion begins by highlighting significant market volatility following a Trump announcement regarding Chinese tariffs. On a Friday, the news led to a $2 trillion market cap wipe-off from the US stock market, with the crypto market also experiencing a substantial hit. However, a recovery bounce was observed, with the Nasdaq up 2% and the S&P 500 up 1.5%, indicating a material reduction in fear, though concerns about Chinese tariffs and rare earth metals persist.

A key issue discussed is rare earth metals, with China controlling 60-70% of the global supply despite holding only one-third of known deposits. The hosts explain that other countries often don't extract these metals due to higher costs compared to China's supply, a principle of capitalism and globalism. However, China's repeated leveraging of rare earth metals as a geopolitical tool is expected to drive other nations towards greater independence in sourcing, similar to China's efforts to develop its own chipsets due to export restrictions.

A perplexing incident of alleged insider trading in the crypto market is detailed:

• Event: 30 minutes before Trump's tariff announcement, a large short position of approximately $700 million in notional value was taken.
• Outcome: After the crypto market "took an enormous dump," the position was closed, yielding a profit of between $160 million and $200 million.
• Suspect: A hedge fund manager from Hong Kong was highlighted, though he denied having "instant information" or knowing the Trump family.
• Concept: This incident raises "suspicion of information asymmetry," as noted by Joshua Devos of CoinDesk, implying someone might have cheated the market.
• Crypto Regulation: The hosts emphasize the lack of comprehensive new regulations for crypto, describing it as a "global casino" with anonymity and global participation, making it susceptible to market manipulation at an unprecedented scale. Unlike traditional stock markets or even casinos, crypto lacks robust regulatory frameworks against insider trading.
• Advice for Crypto Investors: Given the inherent information asymmetry and lack of regulation, investors are advised to be cautious, invest only a "low single digits" percentage of their portfolio in Bitcoin or stable projects, and avoid leverage due to its amplified risk.
• Market Brittleness: The rapid market decline over a single Trump tweet, even years into his administration, underscores the market's brittleness. Specific companies like AMD (down 8%), Tesla (down 5%), Nvidia (down 5%), Broadcom, Apple, and Oracle experienced significant declines, particularly those with robust valuations or direct ties to China.

RAMP's Data-Driven Approach and AI Innovation

The discussion transitions to an interview with Eric Gliman, co-founder and CEO of RAMP, a "mega unicorn" in corporate cards and AI agents for fintech.

RAMP's Data Insights and Open-Source Philosophy

• Scale: RAMP processes over $100 billion per year in spend across 50,000+ organizations, providing an "incredible index" of economic activity.
• Transparency: RAMP open-sources this aggregated and anonymized data at ramp.com/data, making it accessible to small business owners and finance teams. This contrasts with historical practices where such data was expensive and sold to hedge funds (e.g., counting cars in Walmart parking lots for retail trends).
• Leaderboards:
  • New Customer Count: OpenAI, Intuit, Anthropic, Canva, Adobe.
  • New Spend: HubSpot, Carta, Vanta, Pipe 17, Avalara (sales tax automation, with a temporal spike due to tax deadlines).
• Price Intelligence: A RAMP product feature that allows customers to upload vendor contracts (e.g., Salesforce quotes) and compare their proposed rates against what the rest of the market is paying, similar to Zillow's home value estimates. This helps businesses ensure they are paying market rates and save money.
• Earned Media Strategy: The regular release of this data acts as a form of earned media, generating discussion and brand awareness, akin to Zillow's "Zestimate" which, despite initial controversy, drove engagement.

RAMP's Business Model and Cost-Saving Philosophy

• Counterintuitive Mission: RAMP's core mission is to help customers spend less money, not more, despite earning revenue from interchange fees on card spend and paid seat-based software.
• Long-Term Value: Eric argues that by helping businesses save money (currently 5%+ per year, aiming for 10%), RAMP increases their "health span," leading to longer customer retention and potential expansion into more of the business over time.
• Financial Impact: A dollar saved in cost is mathematically more impactful than a dollar earned, especially for businesses with an average profit margin of 8% (a dollar cut of cost is equivalent to $12 earned in profit).
• Financial Performance: RAMP has surpassed $1 billion in annual revenue, is doubling, and is generating free cash flow.
• Investor Management: Eric acknowledges that some investors, particularly early-stage ones, might view free cash flow as a "bug, not a feature," pushing for more aggressive capital deployment to capture the remaining 98% of corporate and small business card spend not yet on RAMP.
• Competitive Edge: RAMP differentiates itself from traditional corporate cards (e.g., American Express Platinum, United Business) by focusing on cost reduction rather than points and rewards, arguing that the latter's value (e.g., lounge access) has diminished.
• Cost Savings Evolution: RAMP has increased average savings from 2% to over 5% by:
  • Kill Switch: Allowing one-click cancellation of subscriptions or blocking specific merchants on cards, addressing a common frustration for businesses and consumers.
  • Dark Patterns: This feature enables businesses to effectively "turn off" payments, forcing vendors to re-engage for renewals, highlighting vendor dark patterns.

Economic Trends and AI's Impact on the Workforce

• Static Team Sizes: The hosts and Eric discuss a trend of static team sizes at mega-cap companies (Uber, Airbnb, Google, Meta, Microsoft) over the past four years, despite revenue growth.
• Leverage per Employee: RAMP's data indicates a significant increase in revenue and valuation per employee, exemplified by companies like Cursor (50 employees, $20-30 billion valuation).
• Unemployment Paradox: Despite companies becoming leaner, US unemployment is at a 50-year low of 4.1-4.2%, within the Federal Reserve's target range of 4-5%.
• Eric's Optimism: Eric expresses hope that this trend will lead to more companies and more lean, highly leveraged organizations, allowing individuals to achieve more with less, potentially reducing the number of people "stuck in mid-level hell" at large corporations.
• Recent College Graduates: A concerning statistic shows that young male college graduates are experiencing unemployment rates similar to non-graduates (4.9% in 2025 vs. 3.25% in 2019), suggesting a 50% increase in unemployment for this demographic.
• AI's Role: Eric attributes this to AI automating entry-level jobs. He offers a profound insight: Large Language Models (LLMs) know more about specialized skills (e.g., accounting, law) in aggregate than any human, as this information is widely available on the open web. This challenges the traditional path to wealth through specialization via university education. If one can effectively interface with a "digital accountant" or "digital lawyer," the need for deep human specialization diminishes.

RAMP's Advanced AI Agents

• RAMP's AI Use: RAMP is a heavy user of LLMs, consuming over a trillion tokens and spending millions on AI. Its primary use is accounting automation, categorizing spend, collecting receipts, and integrating with accounting software, making it faster and more accurate than human accountants.
• AI Model Strategy: RAMP employs a sophisticated strategy for using LLMs:
  • Load Balancing and Comparison Shopping: They actively evaluate and switch between models (OpenAI, Anthropic, Gemini, Grok, open-source options) based on function, cost, and accuracy.
  • GPT-4 Mini Example: For tasks where a cheaper model (like GPT-4 Mini) is 100% accurate 90% of the time and completely inaccurate 10% of the time, RAMP routes the 90% to the low-cost model and the remaining 10% to a more expensive, robust model.
  • Impressive Models: Anthropic is highly regarded for coding/engineering, Gemini for complex tasks with long context windows, and Grok for physics and math.
• AI Agents vs. Intelligence: RAMP's new AI Agents (rolled out for controllers in Q3 and accounts payable in Q4) differ from their earlier "intelligence" features (co-pilots) by having the permission to do things rather than just suggest.
  • Functionality: Agents can automatically approve reports, initiate purchases, and enforce expense policies.
  • Policy Agent: This agent understands a company's expense policy in detail, reviews transactions with 99%+ accuracy, automatically approves 90% of in-policy transactions, and flags the remaining 10% for human review. This saves significant time and catches more out-of-policy spend.
  • Sarbanes-Oxley (SOX) Context: The policy agent addresses the need for independent transaction review, a requirement stemming from SOX, by automating low-value tasks that previously consumed significant human effort.
  • Adoption: Adoption of agents is rapid, even among non-tech customers, because they are embedded, intuitive, and directly solve problems like closing books faster and improving convenience.
  • Training and Trust: The model continuously learns from controller approvals and denials, building trust over time, leading to less human review.

Y Combinator Drama and Founder Ethics

The episode concludes with a discussion about a recent controversy involving Y Combinator (YC) and a founder named Daniel Jung of Omen, an "agentic investing platform."

• The Incident: Daniel Jung's company, Omen, was accepted into YC, used the YC brand to hire, but then declined the traditional $500,000 SAFE investment and left the program before signing the deal.
• YC's Reaction: YC partners, including Pete Kumman and Tyler, publicly criticized Jung on Twitter for breaking a "handshake agreement" and a "commitment and contract," expressing significant anger.
• Founder's Defense: Jung argued that if YC encourages founders to "drop out of college," he should be able to "drop out of YC," maintaining that he was grateful for the opportunity but chose a different path.
• Jason's Perspective:
  • "Punk Rock" Ethos: Jason argues that YC, known for cultivating "punk rock" founders and rule-breakers (like Sam Altman), should not be surprised or upset when a founder acts unconventionally.
  • Unsigned Deal: Since the SAFE was not signed, Jung had the right to back out, especially if a better offer emerged.
  • YC's Overreaction: Jason views YC's reaction as an overreaction, potentially stemming from increased competition from other accelerators (e.g., A16Z Speedrun, Sequoia Arc, Pair, Launch Accelerator, Tech Stars, Antler) which often offer better terms and more bespoke experiences.
  • "Negging" by YC: Jason points out that YC itself has a history of being "rough" with founders, citing an email from a YC partner (initials MS, possibly Michael Seibel) to a struggling team, giving them stark options including shutting down or returning money. If YC can "neg" founders, founders should also be able to act in their own best interest.
  • Advice for Founders: Jason advises founders to seek counsel before making significant deals or equity grants, emphasizing his role as a "mench" (like Dave Goldberg) in the ecosystem, offering guidance without personal gain.
  • Conclusion: Daniel Jung's actions are deemed "punk rock," and YC, as a powerful entity, should be more magnanimous and supportive of founders' choices, even if they deviate from expectations. The power imbalance means YC's public criticism can be perceived as bullying.

Synthesis and Conclusion

The episode provides a multifaceted look at current events impacting startups and the broader economy. It highlights the fragility of financial markets in the face of geopolitical news, the ethical complexities of insider trading in unregulated spaces like crypto, and the transformative power of AI. RAMP's success is presented as a case study in leveraging data and AI to create value, not just by facilitating transactions but by actively helping businesses reduce costs and operate more efficiently. Eric Gliman's insights into AI's impact on the workforce, particularly the challenge to traditional specialization and the rise of AI agents, offer a forward-looking perspective on economic shifts. Finally, the YC drama serves as a commentary on the evolving dynamics between accelerators and founders, emphasizing the importance of integrity, competition, and the need for established institutions to maintain a magnanimous and founder-friendly approach, even when challenged. The overarching takeaway is the rapid pace of change across technology, finance, and the workforce, demanding adaptability and thoughtful engagement from all participants.