Imagine you’re shopping for insurance online. A helpful AI agent appears. It asks about your coverage needs and budget. It gathers your auto insurance history, compares policies, negotiates with providers on your behalf, and makes the purchase. The tool saves you hours of work and gets you a price you’re happy with.
While the experience was positive, it has blind spots. You have no view into how the agent was designed, if it truly provided the best offer, if there is any agent incentive to promote “sponsored” products, or if it will keep your information private. While you saved time, there is little transparency—and no incentive for the company to provide it.
AI agents offer incredible promise, helping companies optimize customer experiences and increase efficiency. Agents are also a potential boon for consumers, taking the grunt work out of switching mobile phone carriers, negotiating discounts from their internet providers, or making sure they have adequate insurance coverage.
In the recent paper, “The Coasean Singularity? Demand, Supply, and Market Design with AI Agents,” MIT Initiative on the Digital Economy research scientists examine where agentic AI is potentially heading and how this could impact the way consumers and companies interact.
The Dawn of AI Agents for Consumers Use
AI agents are autonomous software systems built on large language models (LLMs) that can perceive information, reason through problems, and act across multiple steps without human direction at each stage. Depending on the design, they can execute complex plans, connect to external tools and data sources, and operate within larger digital workflows on behalf of a user.
This type of agentic AI technology is also in the early stages of shopping, negotiating, and searching for products for customers. E-commerce and financial institutions are moving quickly to advance agentic commerce and overcome challenges, unlocking a new market estimated to grow to $175 billion by 2030.
Visa, Mastercard, Google, Amazon, Perplexity and others have announced partnerships that will ideally connect consumers, retailers and financial institutions for seamless shopping experiences.
As these agents evolve, they are likely to disrupt markets and shift the role of humans that currently serve as intermediaries. In their paper, IDE research scientists propose a framework for how that will unfold.
AI Agents and the Effect on Markets
When consumer-facing AI agents gain traction, the researchers argue, they are likely to do so in markets where human intermediaries already exist, such as real estate, job search, insurance, and financial services.
Why these markets?
AI agents, the authors assert, are poised to dramatically reduce transaction costs, and that when those costs collapse, market structures change, too. Anywhere there are high-stakes decisions, large numbers of options to evaluate, and meaningful information gaps between buyers and sellers, markets are likely to be affected.
The researchers call the potential shift in transaction costs a “Coasean Singularity,” in a nod to the economist Ronald Coase, who in 1937 posited that transaction costs shape how firms and markets organize.
For example, someone could use an AI agent to search for and secure an apartment instead of a human apartment broker, suggested Ben Manning, a paper co-author and MIT Sloan PhD candidate who works with John Horton’s AI, Marketplaces and Labor Markets research group at the IDE. One implication of launching this type of agent is that it could lower demand for human brokers and increase demand for AI-powered apartment finder services.
What unifies these markets isn’t the industry. It’s a property of the transaction itself.
“Human intermediaries usually exist for one of three reasons: they’re cheaper than doing it yourself, they have expertise the consumer lacks, or they’ve done that kind of transaction many more times than the consumer ever will,” said Peyman Shahidi, one of the paper’s authors and a PhD candidate at MIT Sloan, also working in Horton’s research group. “AI agents fit the same niche for the same reasons. The specific industries are almost incidental; what matters is the shape of the transaction.”
Lower Transaction Costs Increase Opportunity
Today, when consumers want to make a purchase, they are limited by time, attention, and expertise. These aspects limit how much they search for options and leave them susceptible to behavioral nudges, confusing pricing, and conflicting information.
AI agents, on the other hand, can compare thousands of options simultaneously, negotiate with multiple sellers in parallel, monitor markets over time, and make decisions based on a user’s preferences rather than what the algorithm surfaces.
By lowering transaction costs, AI agents could also enable people to act in markets they previously avoided. Tasks that weren’t worth the effort, such as sourcing contractors, comparing insurance policies, or negotiating a lease, become feasible when the legwork costs almost nothing.
In this sense, AI agents for e-commerce challenge markets built around capturing attention and helping navigate confusing or overwhelming options, and they will likely face disruption as a result.
Two Kinds of Agents, Two Very Different Loyalties
Beyond market disruption, agentic AI systems will affect how consumers interact with the companies they buy from. How this plays out depends on the type of AI agent model used.
The authors draw a distinction between what they call a “bowling-shoe” agent and a “bring-your-own agent.”
The “bowling-shoe” model is provided by the platform hosting a transaction. Like rented shoes at a bowling alley, it’s convenient, ready to go, and optimized for the environment it lives in. While they integrate seamlessly and require little to any user set up, the lack of transparency creates strong incentives to prioritize platform interests, which could result in harm to the consumer, such as higher prices, fewer options, or deceptive practices that inaccurately portray one option as better than another.
In contrast, a “bring-your-own” agent is user-controlled, portable across platforms, and designed to be aligned with your preferences rather than a vendor’s goals. You own it, configure it, and it works for you across different services and marketplaces.
Platform-provided agents may never surface a competitor’s product, even if it’s the better fit for you. They optimize what the platform values, not necessarily your outcome. As the researchers note, these agents benefit from “access to proprietary features and first-party telemetry,” but carry a “risk of self-preferencing.” Most users will find the agent helpful, without realizing its choices were constrained from the start.
Which AI agent will win out?
Because agentic AI for consumer use is still in the preliminary stages, organizations that wish to adopt this technology must now decide what type of AI agent model they plan to implement, or allow to be used, on their website, platform, or e-commerce store.
“This is a choice a lot of companies need to make soon. Will agents be native? Will they be cross-platform? These may require vastly different infrastructures,” said Manning. “Those who get it right first may have a big advantage in rapidly changing markets.”
Shahidi stresses that neither outcome is preordained.
“Both worlds are technically feasible. We could end up with consumers all using platform-provided agents, or all bringing their own; the technology can evolve to support either,” he said. “Which one we get will be decided by firms choosing what to build and what access to grant, and by regulators deciding which kinds of agent access are protected. Those decisions are being made right now, and they’ll be hard to reverse once defaults are set.”
The paper treats this not as a forecast but as a fork: the same capabilities can produce very different market structures depending on the platform and policy decisions made now.
Based on information released by credit card companies and online retailers to date, the architecture and methodology behind AI agents for commerce is already being made for the consumer.
- In March, a California federal judge temporarily blocked Perplexity AI from using its agentic shopping tool on Amazon after allegations that Perplexity unlawfully accessed Amazon the site without its permission.
- The authentication-focused industry association known as the FIDO Alliance announced in April that it will launch two working groups, in collaboration with Google and Mastercard, to develop industry standards for validating and protecting AI agent-driven payments and transactions.
- Last year, Visa released a whitepaper that outlined its path for agentic commerce and touts a future where autonomous shoppers navigate purchasing “securely, instantly, and effortlessly,” by building fraud detection, security and other guardrails into the system.
“Agentic Commerce is no longer a distant vision,” the Visa whitepaper says. “It is unfolding now. The future of payments and AI is inseparable, and those who act decisively today will lead tomorrow’s economy.”