Dear Friends,
After a hiatus of a few months, my co-author Arnab Sinha and I are resuming the publication of this newsletter. So far in 2025 we have found that the pricing hexagon and the pricing games have continued to resonate as business leaders address acute challenges such as tariffs and longer-term challenges such as the integration of AI.
In this edition of the newsletter, we look at some of the current realities around AI and pricing.
Exploring the limits of AI and pricing
Whenever I tell someone that pricing and setting prices are not the same thing, they often roll their eyes and accuse me of making things too complicated.
They’ll say that “it’s just math,” implying that the entire pricing profession will soon surrender to suites of AI tools that can do all the work much faster and much better.
That’s one of many reasons why Project Vend – an AI experiment recently conducted by Anthropic – made me laugh out loud.
The experiment seemed simple: Anthropic let a version of its Claude LLM manage an office vending machine, which seems like a souped-up version of a kid’s lemonade stand. Nicknamed “Claudius”, the model would select and procure inventory, respond to employee requests and feedback, set prices, and collect money.
What could go wrong? Spoiler alert: a lot, as this video explains.
Anthropics own overview of the experiment faulted Claudius specifically for its pricing behavior, which mirrored some of the most common pricing missteps that human managers make. Leaving the frequent and often humorous hallucinations aside, Claudius performed poorly at:
- Recognizing profit opportunities: It ignored a selling opportunity with a potential markup of more than 500% and declined to raise prices amid higher demand for a certain product
- Discounting: It offered unnecessary discounts, even when asked to stop, and sold a relatively expensive high-demand product at a loss.
- Reading the competition: It continued to charge a relatively high price for a product that was available free of charge in a nearby refrigerator.
These are not mechanical errors in price setting, but rather errors in judgment and perspective. I have no doubt that Claudius could “do the math” but running even the smallest business successfully requires some intricate decision making.
Broader implications: The singularity, chaos, and the end of jobs
If you’re like me and you enjoy speculating about the future roles of AI and humans and the limits on what AI can do, Project Vend yields an entire buffet of food for thought. It illustrates that the singularity – the moment that AI achieves general intelligence and humans become marginalized – is not so much a fixed-point in time as the word suggests, but rather a range or a period of time.
We all know that there are already many algorithms that can plan and run all the tasks, pricing and beyond, of a vending machine business. Yet putting it all together with judgment and the necessary interactions is more complicated. It was audacious of Anthropic to let a general purpose LLM like Sonnet 3.7 run the whole thing. LLMs will probably get to the necessary level soon, but assembling the building blocks of entire businesses anchored in day-to-day physical reality in a reliable human way will take time.
Now on the pure pricing side, AI might be exceptionally good at recognizing patterns and optimizing things in isolation. But the kind of intelligence that still makes humans special – the kind that successful pricing requires – goes beyond those facilities. We need to deal with abstract ideas such as value, we need to interpret the patterns we see, and we need to make tradeoffs that may require sacrifices in some areas that improve the overall performance of a system.
The pricing hexagon highlighted the multiple interactions between three inputs and three economic frameworks. We know from chaos theory that three partial differential equations are enough to create unpredictable situations. That’s why I suspect markets will continue to surprise us humans and machines.
Finally, on the question of whether AI will replace jobs, let’s focus on what we can control. There is no doubt AI agents are the latest iteration of digital capabilities that make us more efficient by doing some of what we do. Let’s embrace that efficiency and focus on our differentiation by elevating our thinking around the added value we bring.
In 1800 the price of a loaf of bread was equivalent to two hours of unqualified manual labor, as it had been for centuries in Europe. But with technological progress, by 1900 it took only one hour and by 1950 only half an hour to pay for a loaf of bread. In the following 50 years, the price of bread in weight continued to go down and represented only 10 minutes of work in 1975.
At that point it became impossible for bakers to survive with bread sold by weight, so they focused on differentiation and added value. The bakery moved to “baguettes” in France and specialty “Broetchen” in Germany, higher-end breads with more margin that showcased the skill of bakers. Then came croissants and vienoiseries.
We are all bakers in the age of AI.
Where do you stand on the role of AI and humans, whether in pricing or in your own field?
Original article can be found here.