Talk Your Book: The Two Biggest Stories of the Year: AI & Tariffs

Episode Summary

Executive Summary: The episode centers on Ramp economist Ara Karazian’s data-driven view of AI adoption, arguing that business spend—not hype—best shows whether AI is a bubble. He says AI usage and retention are rising, enterprise adoption is still early, and government surveys likely undercount usage. The discussion also covers how Ramp’s receipts and invoice data reveal tariff pass-through, software spending patterns, and even workplace behavior like weekend work in San Francisco.

Main Topics: AI adoption should be measured by business spend, not sentiment (Priority: 5/5): Karazian argues that private-company AI adoption is best tracked through actual corporate spending, renewals, and contract size rather than public-market narratives or surveys. Ramp’s transaction data shows what firms are truly buying and retaining. Ramp’s data advantage and AI index methodology (Priority: 5/5): The conversation explains how Ramp’s expense and invoice visibility lets Karazian see line-item details, API spend, renewals, and vendor mix across firms. This powers the Ramp AI Index and makes it useful for benchmarking adoption by sector and company size. Government AI adoption estimates likely undercount usage (Priority: 4/5): Karazian critiques the government survey question as outdated and too narrow, suggesting it misses modern uses like customer service automation and software engineering assistance. He says even Ramp’s figures may undercount because of free AI tools. Tariffs are showing up, but slowly and unevenly (Priority: 4/5): Ramp data suggests tariff charges are increasing in manufacturing and retail, but pass-through into invoices and prices is gradual due to legal, logistical, and operational frictions. The transcript emphasizes that announcements do not instantly translate into costs. Software spending is still broad, with AI changing the mix (Priority: 3/5): Rather than a collapse in legacy software, Ramp sees continued growth in software spend, including tech companies spending heavily on other tech companies and AI tools. The winners and losers of the software shift are still unclear. Labor-market effects are narrow so far (Priority: 3/5): The group discusses how AI is affecting white-collar work, especially software and support tasks, but broader employment remains stable. Karazian sees current displacement as limited to specific sectors, with productivity gains already visible. New use cases in Ramp data reveal behavioral shifts (Priority: 2/5): The episode briefly touches on Ramp’s ability to detect trends like Saturday work in San Francisco via food and expense data, showing how transaction data can surface broader workplace and economic shifts.

Key Arguments: Actual business spend is a better indicator of AI adoption than surveys or public-market commentary because it captures real purchasing behavior, renewals, and contract sizes. AI adoption is still early, but the trend is real: retention is high, contracts are getting larger, and more companies are signing meaningful contracts. Government data likely underreports AI use because the survey question is outdated and framed around narrow notions of production rather than today’s enterprise workflows. Ramp likely undercounts AI adoption too because it only sees paid usage, not free tools like Gemini in Google Workspace or employees using ChatGPT on their own. Tariffs are not instantly flowing through the economy because implementation is messy, legal constraints exist, and firms can delay, re-route, or reclassify imports. The economy is not in an internet-bubble analogue yet because AI vendors are generating revenue and profits, and customers are buying the products. AI’s labor effects are real but currently concentrated in a small set of white-collar sectors, especially software-related work, not the bulk of U.S. jobs. Smaller companies may benefit through efficiency gains, but large firms with dedicated technical teams may be better positioned to extract value from AI. Ramp’s data can help firms make better software-buying decisions by showing what similar businesses are actually purchasing and retaining. The absence of immediate price and adoption shifts after policy or technology announcements does not mean nothing is happening; it may reflect lags in the real economy.

Data Points: NVIDIA market cap: $5 trillion - Mentioned as an illustration of the scale of the AI boom. Ramp AI product retention: 80% in 2024 vs. about 50% in 2022 - Used to argue that AI products are sticking with customers more over time. Average AI contract size: About $1 million next year (estimate) - Karazian says AI contracts are getting larger as adoption matures. Companies on Ramp with a significant AI contract: About 40% in 2025 - Indicates broad but still incomplete enterprise AI adoption. Government-estimated AI adoption: About 9% in 2025 - Karazian says the government survey likely undercounts AI usage. Government-estimated AI adoption in 2023: 3% - Shows the low baseline in official data. Ramp-estimated AI adoption: 44% - Based on actual business spend across Ramp customers. Tariff incidence on invoices: About 3% - Ramp sees tariff line items on a small but rising share of transactions. Average tariff rate / incidence context: About 1.4% previously - Referenced as a prior level before the recent increase in tariff-related transactions. Weekend work in Bay Area: About 40,000 workers - Estimated increase in workers in San Francisco/Bay Area working Saturdays and Sundays.

Pivotal Quotes: "when we're talking about whether or not something is a bubble, I feel like people are grasping for all the different data sets that they can possibly use to support their answer." — Ara Karazian: On why AI bubble debates are so noisy and why spend data matters. "if you look at the government estimate, you know, they were at 3% in 2023. It's now only about 9% in 2025. That seems really low." — Ara Karazian: Critiquing official AI adoption statistics as too low and likely outdated. "If the size of contracts start to go down, and if retention rates start to go down, that means that companies are trying these AI products and services and they're just not working for them." — Ara Karazian: Explaining what would signal that AI demand is weakening.

Implications: Listeners should watch real spend, renewals, and contract sizes—not just hype—when judging AI, tariffs, or software trends. The transcript suggests AI adoption is genuine but early, tariffs are slower to bite than expected, and transaction data is becoming a powerful economic signal.