Bowling-shoe Agents

The Coasean Singularity? Demand, Supply, and Market Design with AI Agents

To the broader point of a Coasean Singularity…well….adversarial games be adversarial. To the narrower point of Bowling-shoe Agent: what a brand position!

“AI agents—autonomous systems that perceive, reason, and act on behalf of human principals—are poised to transform digital markets by dramatically reducing transaction costs.”

“We take seriously the idea that, rather than hiring a human agent, one could instead rely on an AI agent.”

“AI agents expand the economic market design frontier. By collapsing the costs of eliciting preferences, enforcing commitments, and verifying identity, agents make mechanisms that were once only theoretically attractive now feasible at scale. This new set of feasible designs has the potential to improve consumer welfare and matching. That said, AI agents are not guaranteed to make markets more efficient or to improve other social objectives. Even if it is individually rational for consumers and firms to adopt agents, the equilibrium outcomes may be suboptimal. This is particularly relevant in environments with externalities across agents, imperfect alignment, and asymmetric information. These factors point to an exciting market design agenda for economists: translating theoretical insights into practical mechanisms that can guide the transition to agent-first environments and capture the full benefits of AI.”

“Second, and perhaps more consequential, agents will enable undertakings that would not have been attempted at all. By lowering the cost of exploration and execution, they expand the feasible set of options and reduce the threshold of “worth doing.””

“The fundamental economic promise of AI agents lies in their ability to dramatically reduce transaction costs—the expenses associated with using markets to coordinate economic activity. This reduction occurs not only through direct task execution but also via advisory services. In the direct approach, agents perform tasks entirely on behalf of users, from price comparison shopping to contract negotiation to job interviews. In the advisory approach, agents assist users in making better market decisions, such as helping job applicants individually optimize their resumes for each job submission. Recent evidence suggests both approaches can be quite effective.”

“By lowering the costs of preference elicitation, contract enforcement, and identity verification, agents expand the feasible set of market designs but also raise novel regulatory challenges. While the net welfare effects remain an empirical question, the rapid onset of AI-mediated transactions presents a unique opportunity for economic research to inform real-world policy and market design.”

“AI agents can significantly alter matching markets. Trained to reason directly over natural language, foundation models already allow a single agent to parse a paragraph describing one’s tastes (Rusak et al. 2025). Soon, they will move beyond just facilitating preference elicitation to discovering them. Like a skilled coach or therapist, an AI agent can help users articulate values by detecting behavioral patterns: a home-buying assistant might notice a buyer consistently favoring houses near parks with large windows—suggesting unspoken preferences for green space and natural light. By surfacing these insights and prompting reflection, agents could help individuals better understand their own priorities.”

Shahidi, P., Rusak, G., Manning, B. S., Fradkin, A., & Horton, J. J. (2025). The Coasean Singularity? Demand, Supply, and Market Design with AI Agents (No. w34468). National Bureau of Economic Research.

https://www.nber.org/system/files/chapters/c15309/c15309.pdf

Mitigating Hallucinations in Large Language Models via Causal Reasoning

Help it think right

“Our experiments across four LLMs on the CLADDER, WIQA, and HaluEval benchmarks demonstrate that CDCR-SFT significantly improves causal reasoning, achieving a state-of-the-art accuracy of 95.33% on CLADDER, surpassing human performance (94.8%) for the first time. Moreover, CDCR-SFT reduces hallucination in HaluEval by up to 11%, confirming that enhanced causal reasoning directly mitigates hallucinations. These results affirmatively answer our research question: improving the causal reasoning capabilities of LLMs can mitigate hallucinations.”

Li, Y., Shen, Y., Nian, Y., Gao, J., Wang, Z., Yu, C., ... & Zhao, Y. (2025). Mitigating hallucinations in large language models via causal reasoning. arXiv preprint arXiv:2508.12495.

https://arxiv.org/abs/2508.12495

Emergent Communication through Negotiation

Cheap talk

“We showed that by communicating through a verifiable and binding communication channel, self interested agents can learn to negotiate fairly by reinforcement learning, using only task success as the reward signal. Moreover, cheap talk facilitated negotiation in prosocial but not in self-interested agents, corroborating theoretical results from the game theory literature (Crawford & Sobel, 1982). An interesting future direction of research would be to investigate whether cheap talk can be made to emerge out of self-interested agents interacting. Recent encouraging results by Crandall et al. (2018) show that communication can help agents cooperate. However, their signalling mechanism is heavily engineered: the speech acts are predefined and the consequences of the speech acts on the observed behaviour are deterministically hard-coded. It would be interesting to see whether a learning algorithm, such as Foerster et al. (2017), can discover the same result.”

Cao, K., Lazaridou, A., Lanctot, M., Leibo, J. Z., Tuyls, K., & Clark, S. (2018). Emergent communication through negotiation. arXiv preprint arXiv:1804.03980.

https://arxiv.org/abs/1804.03980

Enlightenment ideals and belief in progress in the run-up to the industrial revolution: A textual analysis

Everybody in their corners

“Using textual analysis of 264,443 works printed in England between 1500 and 1900, we test whether British culture manifested a belief in progress associated with science and industry.”

“We document the following three results. First, there is little overlap in works using the language of science and religion beginning in the mid-18th century. This indicates that the “secularization” of science became entrenched during the Enlightenment. Second, while works using the language of science did become more progress-oriented during the Enlightenment, this sentiment was mainly concentrated in language at the nexus of science and political economy. Third, those volumes using language at the nexus of science and political economy that also used the language of industrialization were particularly progress-oriented. We interpret these findings to mean that it was the more pragmatic, industrial works using the language of science—those that spoke to a broader political and economic audience, especially those literate artisans and craftsmen at the heart of Britain’s industrialization—that contained the cultural values cited as important for Britain’s economic rise.”

Almelhem, A., Iyigun, M., Kennedy, A., & Rubin, J. (2023). Enlightenment ideals and belief in progress in the run-up to the industrial revolution: A textual analysis.

https://digitalcommons.chapman.edu/esi_working_papers/393/

AlphaXiv

Progress!

Quickarxiv on any paper URL to get an instant blog with figures, insights, and explanations

https://www.alphaxiv.org/

Reader Feedback

“Hinton’s point about steering AI into motherhood is … dark.”

Footnotes

I’ve been testing the value propositions for the Virtual Twin of Consumer (VTOC) tech. I’ve been enjoying my own champagne. And doing the qualitative research. Listening for doppler in the conversation. Lots of Blueshift.

Some Redshift when I cause fear or reaction.

VTOC’s are positioned in my mind as a serious science that solve a serious business problem. They’re fantastic tools. They’re on about the same level as the invention of the Box Plot. They focus the mind of the user to fantastic effect. The VTOC is the treatment that treats you.

As a part of this work, I tried Pomelli to see what Google would do with it. You can try it out here:

https://labs.google.com/pomelli/about/

It’s a labs project.

My expectations are checked going in.

Let’s have some fun. Some rare meta-meta-meta fun!

This is what it produced:

Neat, right? It’s a senior decision maker that’s literally kept up at night by stalling growth.

The agony! But it got the fingers right! You still got your fingers Mister Mad Man CMO.

How about a smile? How about romancing the gain?

“The Strategy To Find $5M More” is a straight up aligned gain. Good job!

“Emororing discovery!” - what an exciting new task!

It caused all sorts of questions about the gatodo brand key reflected back.

There isn’t too much science on how marketing scientists market to marketers. I’m going for extreme valence as a start. Because marketers can generally tolerate higher degrees of creative divergence. When I look at industry judging preferences, they tend to go for high divergence in ways that the mainstream doesn’t.

The invitation should contain a hint that they’ve been seen. I see you. Do you see me seeing you?

Serious. Dignified. So many arrows pointing to different places on the Twin and the CMO leader looks a bit too serious.

I still have to write a brief. That hasn’t changed. But the feedback, reflection, and feedback cycle is a lot faster. It speeds up the front end of the effort, certainly. But that last, differentiable, mile isn’t there.

Neat?

Never miss a single issue

Be the first to know. Subscribe now to get the gatodo newsletter delivered straight to your inbox