How Companies Are Actually Spending Money on AI Now

Episode Summary

Executive Summary: The episode examines whether AI spending is becoming a real macro driver or a market bubble, using Ramp CEO Eric Glyman’s transaction data to show rapid growth in AI budgets and a shift from experimentation to operational use. It also explores where AI is being adopted most, how it’s changing sales and engineering workflows, and what this means for productivity, competition, and job design.

Main Topics: AI as a macro and market driver (Priority: 5/5): The hosts frame AI as a major pillar of the market and debate whether investment is being matched by revenue and productivity gains, with possible outcomes ranging from a bust to a breakthrough or AGI-level transformation. Ramp’s visibility into company AI spending (Priority: 5/5): Eric Glyman explains how Ramp sees over $50 billion in annual spend across customers and can observe AI purchases at a granular level, including licenses, API usage, and vendor-specific behavior. Shift from experimentation to operational adoption (Priority: 5/5): The discussion highlights evidence that AI products are no longer just trial subscriptions; retention and recurring use are rising sharply, indicating real workflow integration. Where AI is being used inside companies (Priority: 4/5): Ramp says engineering, sales/growth, and marketing are the biggest internal use cases, with AI automating repetitive tasks like lead research, call analysis, follow-ups, and content generation. Competition, model choice, and lack of lock-in (Priority: 4/5): The guests discuss how AI markets differ from prior software categories because developers can multiplex across models and switch quickly based on performance, reducing lock-in and intensifying competition. Productivity, labor, and job redesign (Priority: 4/5): The conversation considers whether AI will cut jobs or make work more interesting by automating low-value tasks, while leaving judgment, creativity, and human connection to people. Government spending and waste reduction (Priority: 3/5): The latter part of the interview uses government procurement as an example of inefficient spending and outdated tooling, arguing that modern software could reduce waste and improve auditability.

Key Arguments: AI spending is rising fast enough to matter at the company level and potentially the macro level, but it still needs to translate into measurable revenue or efficiency to justify valuations. Ramp’s data suggests AI usage is moving from trial behavior to embedded operational use, shown by much higher retention of AI products after purchase. AI adoption is strongest in digital, data-rich functions like engineering and sales because those workflows are easiest to automate or augment. The biggest near-term value of AI may be in automating low-value tasks rather than replacing full jobs end to end. AI markets are unusually competitive because developers can test multiple models and route work to whichever performs best, limiting lock-in. Modern expense and financial workflows can be substantially automated, saving companies time and money and even enabling better pricing intelligence. In the public sector, outdated procurement and legacy systems create waste that modern digital tools could help reduce, though structural budget issues remain much larger. Job disruption may be significant over decades, but historical shifts such as farming’s decline suggest the labor market can adapt over time.

Data Points: Ramp customer spend visibility: over $50 billion a year - Glyman describes the annual spend data Ramp can observe across its customer base. Customer savings since founding: over $2 billion - Ramp says it has reported this cumulative savings figure to customers over four years. Time saved since founding: 20 million hours - Ramp says this is the amount of time saved for customers over four years. Share of savings in past year: about half - Glyman says roughly half of Ramp’s total reported savings came in the past year. Customer count: 30,000+ companies - Ramp says it serves more than 30,000 companies, from small businesses to public firms. AI spend growth per customer: about 4x - Average Ramp customer spending on AI-based products rose about fourfold from the start of 2023 to the end of the period discussed. AI product retention in 2022: 50% chance of not remaining a customer within a month - Used as a measure of experimentation versus operational adoption. AI product retention in 2023: 70% chance of remaining a customer within a month - Signals more durable, operational use of AI products. Anthropic developer market share: 20% in 2024 versus 3% in 2023 - Glyman cites this as evidence of rapid model-share shifts. Ramp salesperson productivity: about 4x as productive as closest competitor - Glyman attributes part of Ramp’s sales productivity to AI automation. Average American company profit margin: about 8.5% - Glyman uses this to argue that cost savings can be very valuable. Potential savings from expense management: about 5% per year - Glyman says Ramp’s customers on average cut expenses by about this amount. Government email shutdown: 4 hours per night - He cites this as an example of outdated systems and operational inefficiency in some agencies.

Pivotal Quotes: "“There’s this growing idea of service as software.”" — Eric Glyman: He describes the next stage of AI where workflows are automated end to end, not just accessed through chat interfaces. "“AI can do your expense reports.”" — Eric Glyman: He uses this as a concrete example of AI automating low-value administrative work. "“The next programming language, in fact, is English.”" — Eric Glyman: He argues that natural language interfaces may become the dominant way people instruct software and AI systems.

Implications: AI spending appears real, fast-growing, and increasingly embedded in workflows, especially in engineering and sales. But the market still needs durable productivity gains and revenue to justify valuations. Expect more specialized tools, more competition among models, and gradual reshaping of jobs rather than instant replacement.