In partnership with
Dear Readers,
Today, we have something very special for you in the newsletter: an exclusive interview with Microsoft economist and researcher David Rothschild about his groundbreaking article "The Agentic Economy".
David argues that generative AI’s most disruptive impact won’t be just automating tasks - it will be drastically reducing communication friction in markets. You can find out what he means by that and to what extent it will be disruptive in a conversation with him.
Many thanks to Microsoft for making the conversation possible.
David Rothschild is an economist at Microsoft Research. He has a Ph.D. in applied economics from the Wharton School of Business at the University of Pennsylvania. He has written extensively, in both the academic and popular press. His work pushes the boundaries of varying data and methods: polling, prediction markets, social media and online data, and large behavioral and administrative data. His work focuses on solving practical and interesting questions, including: mapping and updating public opinion, the market for news, the effect of advertising, finance, and an economist's take on public policy.
You can find David’s full biography at ResearchDMR.com (opens in new tab).
All the best,
Walled Gardens vs The Web of Agents
The Takeaway
👉 AI’s impact depends on how agents communicate: The real shift isn’t just smarter models, but whether agent-to-agent communication is open or controlled - because whoever controls those channels controls economic power.
👉 Open networks spread value; walled gardens concentrate it: Proprietary ecosystems are likely in the short term, but open, interoperable systems would distribute benefits more widely to consumers and small businesses.
👉 Human-style web interaction is a bottleneck: Agents pretending to click forms and browse websites are limited; direct, structured agent-to-agent communication unlocks richer automation, negotiation, and entirely new market structures.
👉 The economy shifts from attention to preference: As agents search and decide for us, ads lose dominance and reputation, trusted data, and product quality become the key drivers of discovery and revenue.
Before we get into the interview, a quick summary of what his article “The Agentic Economy” is about.
Rothschild and co-authors argue that generative AI’s most disruptive impact won’t be just automating tasks - it will be drastically reducing communication friction in markets. Today, consumers and firms waste time “re-explaining” context (preferences, constraints, history) whenever they switch providers or negotiate complex services. Forms and menus only partially help, often shifting effort to the user and making interactions rigid.
Their vision is an agentic economy: every consumer has an assistant agent, every business has service agents, and these agents communicate directly, flexibly, and unscripted. That could slash search and switching costs, expand available options, and reorganize market power. But we’re not there yet. Most current systems are either siloed service agents (chat interfaces inside one company, like Amazon/Expedia) or end-to-end agents that “computer-use” websites by simulating clicks. End-to-end agents can look impressive, but without true business-side agents, they remain constrained, brittle, and potentially adversarial - especially for ad-driven sites.
The key fork: will inter-agent communication happen inside closed “agentic walled gardens” (platform-controlled ecosystems, like app stores for agents) or an open “web of agents” (interoperable like the early web)? Walled gardens can offer discovery, trust, safety, and even insurance—but risk power concentration and fragmentation. A web of agents would be more decentralized and innovation-friendly, but requires shared standards plus strong discovery, trust, and security infrastructure.
Second-order effects: advertising may change because attention becomes less scarce when assistants can evaluate massive option spaces; the scarce resource becomes high-quality preference/feedback data (“preference economy”). Payments could shift toward micro-transactions as assistants seamlessly switch providers. And agents enable extreme unbundling and rebundling of digital goods—hyper-personalized products assembled dynamically, potentially supported by micropayments that compensate creators.
At the core of your article is a strong thesis: that the way AI agents communicate will determine whether AI broadens or concentrates economic power. How would you summarize this thesis in plain language—and what do you think most people currently underestimate about it?
AI will reshape the economy, but not just because agents provide super-intelligence, but because they can talk to other agents at scale, speed, and extremely low cost. Once your assistant agent can communicate seamlessly with the service agents representing any business (and them with each other), the friction that has defined the digital economy for 25 years (e.g., forms to start a new business relationship, set menus that define prices and product offerings, search boxes to explore for more information) starts to disappear.
What most people underestimate is that communication architecture is power architecture. If agents can only talk within a few propriety ecosystems, then those ecosystems will capture most of the value. But if agents can communicate openly, more like email or the early web, then the benefits spread widely: consumers get better options, small business get direct access to consumers, and innovation accelerates. People tend to focus on model intelligence or benchmarks, which are important. But I focus on who controls the channels through which agents interact, because that determines whether AI becomes a democratizing force or a centralizing one.
Today, many AI agents still operate by imitating human behavior—clicking through websites and filling out forms. Why do you see this as a dead end, and what fundamentally changes once AI agents can communicate directly with each other?
Having AI “pretend to be a human user” is a clever short-term solution, but it’s fundamentally limited. It forces agents to operate with interfaces designed for people: tiny text boxes, rigid forms, brittle workflows. It also creates adversarial dynamics: businesses built on advertising or controlled funnels don’t want bots scraping their sites, and they can block or throttle them at any time. Once agents can communicate directly with agents on both sides of the interaction, everything changes, and there can be richer interactions: instead of filling out a form, your agent can send a detailed, structured request: “plan a three-day trip with these constraints, negotiate price, and check cancellation policies.” It introduces new products and services: businesses can expose capabilities they could never fit into a website UI. It lowers friction: switching providers becomes trivial because your agent can instantly convey your preferences and history. There can be true automation: agents can negotiate, compare, and transact at machine speed, not human speed.
Direct agent-to-agent communication is the difference between AI being a “faster web surfer” (augmenting our current markets) and AI enabling entirely new market structures (creating new AI-first markets).
You describe two possible futures: “agentic walled gardens” controlled by a few large companies, and an open “web of agents.” Which of these futures do you personally find more plausible in the near term—and what economic or political forces will push us in that direction?
In the near term, walled gardens are more likely. Large platforms already have the users, the data, and the distribution channels. It’s easy for them to extend their existing ecosystems—app stores, marketplaces, operating systems—into agent ecosystems. And for consumers, these environments feel safe and convenient, because they are safe and convenient.
But several forces may lead the push toward openness over time: (1) Economic pressure: open systems tend to produce more innovation (higher overall welfare and more distributed welfare among small businesses) and lower prices (high welfare gains for consumers). (2) Regulation: especially in the EU, there is strong momentum toward interoperability and limits on gatekeeping. (3) Technical progress: emerging protocols make open communication increasingly feasible, but there is a lot that needs to be built to ensure quality, trust, and safety to ensure the viability of more open markets.
So, the short-term equilibrium is likely closed, but the long term there will be a strong push to make them more open. The question is how quickly we can accelerate that transition.
If AI agents can search, compare, and negotiate on behalf of users, attention becomes less scarce. What does that mean for advertising models that dominate today’s internet—and what might replace them?
Advertising today is built on a simple fact: humans have limited attention, so businesses pay to get in front of us.
In an agentic economy, your assistant agent doesn’t get tired, bored, or distracted. It can evaluate thousands of options nearly instantly. That means: traditional ads lose power, because they’re no longer the key gateway to discovery. Paid ranking still exists, but it targets agents, not humans, and it competes with far richer signals. The scarce resource becomes high-quality preference data, not eyeballs. Agents will rely heavily on trustworthy feedback, reviews, and user satisfaction signals. We move from an attention economy to a preference economy, where businesses will compete to earn positive feedback from early users, because that feedback trains the agents that future users rely on. Monetization shifts toward reputation systems, microtransactions, and value-aligned ranking rather than interruptive advertising. In short, business resources shift from getting our attention to improving their product fit and proving their quality to get our business.
Looking ahead, what is one concrete, everyday example of how an agentic economy could create new opportunities for individuals or small businesses that are currently locked out by platform dominance?
Imagine a small business owner: let’s say she makes handcrafted spice blends in her kitchen. She finally decides to list her products on a big sales platform. When a customer searches for “spice blends,” the first page is a wall of sponsored listings: big brands and well-funded sellers paying for placement. Her product is technically “there,” but buried under layers of ads. To be seen at all, she has to buy ads too. Suddenly, she’s paying the big sales platform for the privilege of competing on the big sales platform. Then come the fees: referral fees, fulfillment fees, storage fees. By the time a jar sells, the big sales platform is taking a big chunk of the revenue. She raises prices to compensate, but that makes her less competitive, so she buys more ads. It’s a treadmill. And looming over everything is the constant threat of copycat competition. So she spends her evenings tweaking keywords, adjusting bids, responding to algorithmic swings, and fighting for visibility in a marketplace that was supposed to help her grow. In the end, she’s doing everything except what she actually loves: making great spice blends.
Now imagine that same spice blend maker in a world where agents, not platforms, mediate discovery. She doesn’t need to master SEO, buy ads, or reverse engineer the big sales platform’s ranking algorithm. Instead, she sets up a simple service agent, something as easy as filling out a structured profile of her products, ingredients, pricing, shipping constraints, and brand story; the agent becomes her digital storefront and salesperson.
One evening, a home cook tells their personal assistant agent: “Find me a small batch spice blend with no additives, good for roasting vegetables, under $12, and ideally from an independent maker.” Instead of funneling that request through a single marketplace, the assistant agent fans out across the open network of service agents. It evaluates dozens of small producers: most of whom would never appear on page one of a search engine today. It compares ingredients, freshness, shipping times, and even past customer satisfaction signals. And it finds her. Her service agent answers the assistant’s questions instantly: Yes, the blend is additive free; Yes, she can ship within two days; Yes, she offers a sampler pack if the customer wants to try more flavors. The two agents negotiate the details automatically (price, shipping, substitutions, etc.). She gets an order notification only when everything is settled. There are no ad fees, no pay to play ranking, no platform taking a huge cut, and no fear that a giant competitor will clone her product and outrank her overnight. She’s discovered because she’s the best fit for the customer’s needs, not because she paid the most for visibility.
In an agentic economy, the long tail finally becomes accessible. Small businesses aren’t buried: they’re discoverable. And customers get exactly what they want, not what a platform makes the money by showing them.
Why it matters: This matters because agent-to-agent communication could collapse switching and transaction friction, forcing markets to reorganize - shifting power away from today’s dominant intermediaries (app stores, marketplaces, ad platforms) toward whoever controls agent ecosystems, discovery, and trust. It also rewires monetization: attention-based advertising may weaken while preference/feedback data and micropayments enable new, more personalized products and business models.
Sources:
🔗 https://www.microsoft.com/en-us/research/people/davidmr/?msockid=05122a90a981665221f43cc3a8a7671c
🔗 https://dl.acm.org/doi/epdf/10.1145/3747175
Want to get the most out of ChatGPT?
ChatGPT is a superpower if you know how to use it correctly.
Discover how HubSpot's guide to AI can elevate both your productivity and creativity to get more things done.
Learn to automate tasks, enhance decision-making, and foster innovation with the power of AI.
How'd We Do Today?
Stop Stitching Your Business Together
Starter apps create friction. Overbuilt platforms slow you down. Kajabi brings your business together, from offers to marketing to delivery, so growth feels clear and focused, not overwhelming.
