The Soul of Enterprise

The episode delivers a coherent, substantive critique of AI token/credit pricing grounded in real pricing theory - actuarial portfolio logic, learning-curve economics, and value-vs-cost-determines-price - but the core thesis is restated many times without deepening, and long personal anecdotes (Tesla charging, Opus clips usage) dilute the idea-per-minute ratio significantly.

“every doubling of the production reduces the cost to serve by 30 or 40%”

“The job of financial management is not to insist that prices recover costs. It is to insist that costs are incurred only to make offerings that can be priced profitably, given their value to customers.”

The 'ABC in a hoodie' framing is a genuinely clever application of a long-standing pricing critique to a new domain, and the AOL-by-the-hour parallel and actuarial 'no bad risks, only bad premiums' angle are fresh touches; however the underlying framework is explicitly the hosts' 30-year thesis on value-based pricing, not new thinking, just competently re-aimed at AI.

“This is why, you know, we call this, um, ABC and a hoodie. This isn't so much the finance or the cost accountants, these are the engineers.”

“I think we're going to look back on this and maybe two years, five years, not sure but and say this is just like when AOL charged by the hour.”

There are no external guests; the episode is two co-hosts, one of whom (Ron Baker) is a legitimate pricing thought leader with decades of practitioner credentials in professional services, but neither host has operated inside an AI company or priced AI products at scale - they are commentators on others' data, referencing Mansard, Stiving, and Hochberg without having them present.

“I'm at class with my friend and co host and co founder of Threshold, Ron Baker”

“I developed an agent, my first agent. I'm very proud of it.”

The episode has genuine concrete data points - Hochberg's GPU cost breakdown, Ed's real API cost-optimization numbers, Ramp's 13-fold spending figure, JP Morgan's $10K/month per-employee example - that lift it above pure opinion, but many assertions are hedged or vague ('I think,' 'last time I looked,' 'I don't know') and the sourcing for several claims is indirect or approximate.

“a gigawatt worth of Nvidia GPUs cost roughly $50 billion to deploy...the GPUs and the network that connects them are about 35 billion. That's about 70%”

“ChatGPT 5.5, the highest version, was costing like $3.50 per transcript...I got it down to where I was getting. Be able to get throughput of four to five transcripts per hour at a price of 60 cents a transcript”

With no guest to interview, the craft ceiling is structurally lower; the hosts do engage seriously with the strongest counterarguments (Mansard's 75-page white paper, Stiving's reluctant defense) and occasionally push back on each other's analogies, but they fundamentally agree throughout, producing reinforcement rather than productive tension, and several tangents (Tesla electricity rates, West Side Story) slow the intellectual momentum.

“Yeah, But Ed, open AI has what, 800 million users last time I looked. Or 900 million. And, and I don't know, anthropic and perplexity. They've got to have some data on this”

“This is just lazy. This is just friggin lazy. And they're putting the burden on the customer and they need to take the burden because they have the portfolio.”