Sam Altman's First Guinness and a Bar Conversation About AI Deflation

John Collison pulled Sam Altman into a bar and ordered a Guinness. It was Sam's first Guinness. The two talked about PHP vulnerabilities, OpenAI's math weekends, agent swarms, and why AI will cause deflation.

The top-liked replies under the video mostly discussed the foam on that Guinness. Some asked who poured the beer; others posted photos complaining that the foam was too thick. A user named Barry said the foam on that glass almost stretched the G out of shape. Bar conversations are like that: serious topics and nonsense mixed together.

But there were indeed a few points in this conversation worth noting. A user summarized the key points into an image:

The points in the image include: bringing frontier workflows to mainstream enterprises is an opportunity; coordinating intelligence may be as important as improving the models themselves; the next generation of the internet may be agent-to-agent, and company websites as interfaces will become increasingly less important. Later, someone reminded readers that the third and fourth points are directly related to OpenAI's revenue, so listeners should keep that in mind.

The term agent swarms was mentioned in the conversation. It can refer to many small agents dividing up work and collaborating, or to a planner breaking down tasks and distributing them to executors. Replies also asked which one it meant, but there is no definitive answer yet. The direction, however, is clear: having multiple model instances check and cover for one another is more reliable than a single call.

AI deflation is another topic. Sam Altman believes AI will lead to deflation. This judgment is not new, but the bar setting makes it more concrete: if agents can replace part of software procurement and outsourcing, corporate costs will fall, and prices will come under pressure.

There is a ready-made example in the replies. Keith Richman said his team used Claude to build internal tools and canceled Tableau and several other software subscriptions. The money saved is real, and the revenue lost by the displaced vendors is real too. This kind of deflation will not show up in macroeconomic data first. It shows up first on a company's procurement list.

Sam McRoberts pushes the logic further in a long post. His core argument is: as long as one company chooses to use AI and robots to cut costs, competitors must follow, or they will be eliminated by price competition. This cycle will push prices toward the cost of energy, compute, and raw materials. He calls this defection into abundance. The article link is here for those interested: Defecting Into Abundance.

The relaxed feel of a bar conversation masks a not-so-relaxed fact: AI deflation may happen first in software budgets. By the time we see changes in CPI, many companies' procurement lists will already have been rewritten.