AI and Marketplaces: The Future is Agent-Led Marketplaces

The rise of agentic AI will transform marketplaces from passive aggregators to active facilitators of transactions. The future is Agent-Led Marketplaces.

Hi, it’s Colin! Welcome to Take Rate. I am excited to have you join the 2,400+ marketplace founders, operators, and investors who subscribe. Join the fun!

---

The tool of the week is…

folk - Finally, a CRM that doesn't overcomplicate it.

Backed by investors like Accel, Folk is a CRM that builds genuine client connections that are simple to use and integrate. It covers all your CRM needs: Pipeline Management, Chrome Extensions, Dashboards, one-click Enrichment, Integrations, and Messages and Sequences.

I use it for all of my fundraising for Yonder. If you would like a 20% off discount code, please restack 🔄 this post, and I will message you directly with the code.

---

Setting the Stage

We all remember the first time we used ChatGPT. The experience was magical, and the raw potential was eye-opening. In the last two years, ChatGPT and AI, generally, have changed the world as we know it. If software was eating the world, it was at a measured pace before, but now it is devouring it at a breakneck speed. Creative destruction is rampant: businesses are forced to evolve or perish, labor markets are in turmoil, and the future looks unclear.

What a way to end 2024 and begin 2025!

Far from being immune to AI disruption, marketplaces are uniquely positioned to benefit from the tailwinds AI agents unlock. The revolution will fundamentally change marketplaces as we know them.

Marketplaces in the Age of AI

Marketplaces are at the heart of economic activity. From the old bazaars to today’s global digital platforms, they have evolved to connect supply and demand more efficiently over time. Yet, despite all the advancements in modern marketplaces, a fundamental aspect remains unchanged: they rely on humans to drive decision-making. Buyers search, compare, and choose; sellers optimize, adjust, and compete. But what happens when AI steps in—not as a tool for efficiency but as the engine driving marketplace dynamics?

The answer is a massive unlocking of economic potential that could revolutionize marketplaces as we know them. Marketplaces can move from being passive products to being proactive ones.

Marketplaces are already harnessing AI, but the applications are becoming more robust with LLMs. AI-first search engines have been created to index and proactively use preferences to make suggestions. Today, they are prompt-based, but with the advent of agentic AI, marketplaces can harness AI agents to match supply and demand proactively.

I call these new types of marketplaces Agent-Led Marketplaces, or ALMs for short. They have the potential attributes:

1. Autonomous: Act without prompting to achieve the goals of the market.

2. Goal-Oriented: Pre-emptively and proactively aggregate and match supply and demand.

3. Adaptive-Learning: Iterative usage and learning help build better customer experiences by increasing trust and personalization.

The Evolution of AI in Marketplaces

The current wave of AI has primarily improved marketplaces' bottom-line efficiency. They use AI to cut costs, streamline operations, and enhance productivity.

Marketing, for example, has greatly benefited from AI. AI has improved efficiency through content creation, personalized campaigns, targeting, and optimized ad spending. These innovations are powerful but fundamentally incremental. They refine what we already do, making it faster and cheaper.

However, these systems rely on human intervention to set goals, analyze outputs, and make decisions. Even when AI contributes to decision-making, its role is bounded by predefined inputs and objectives, making AI a latent tool.

The real potential of AI in marketplaces lies not in optimizing processes but in reimagining them entirely. AI is the key to a new model in which the heavy lifting of aggregation, matching, and decision-making is no longer human-driven but agent-led.

One of AI’s greatest strengths is its ability to synthesize massive amounts of data to make complex recommendations. It doesn’t just filter options—it can deeply understand needs, predict preferences, and surface the best matches with unparalleled precision (when appropriately trained and not hallucinating). In traditional marketplaces, buyers must search, evaluate, and select from various options when they have time.

Imagine booking a vacation. Today, you might spend hours comparing flights, hotels, and activities on travel sites. Take Airbnb, for example. You must find and select the filters you care about and then submit them to find the rental that meets your needs. This process is very clunky and time-consuming.

Imagine not having to select all the filters.

What if you could have a natural language search that considered your preferences and filtered accordingly? What if the agent used your past bookings, reviews, and other data to improve those recommendations? What if the agent knew I took a trip every year at Spring Break with my kids and proactively made suggestions?

In an agent-led marketplace, this could be a reality; you could tell your AI agent, “I want a relaxing beach vacation with great food and some adventure options under $5,000.” The agent would handle everything: finding the perfect destination, booking flights and accommodations, and even curating a personalized itinerary within your budget—keeping you updated and ensuring you’re in control.

Agentic AI can play a decisive role in actively redefining how marketplaces match supply and demand by shifting the paradigm from a faceted search to a prompt or chat-based experience that leads to goal-oriented agents adapting over time to drive marketplace outcomes proactively.

One company working on this today is PartyPlease, a marketplace for party planners, which switched from a standard grid search experience to a long-form chat that combines humans and agents. Jeremy Burton of Platform Venture Studio shared that this change improved booking conversion rates by 30%+, which is unsurprising, given that party planning is inherently customized and not a one-size-fits-all process.

I am not the first to highlight AI's potential to improve search. Many companies use some form of AI today to enhance their search algorithms. In his piece The AI-First Marketplace, Pete Flint of NFX aptly points out that AI can “Reimagine the Search Box.” As he points out, AI will shift from a push to a pull experience in search, making matching a more active than a passive experience.

Olivia Moore of a16z phrases it as “new search modalities” that lower the cognitive load of searching and the shoe leather costs for consumers.

“These changes will reduce scrolling fatigue and make it more likely for a search to convert into a transaction. If you’ve ever spent 20 minutes or more browsing DoorDash, you understand that closing the gap from “inspiration” to actually finding what you’re looking for is crucial to getting an order placed.”

Her post is excellent and highlights several worthwhile examples. Her example of Glaze, which uses natural language and AI to generate results based on style, vibe, etc., rather than standard facets like color, size, etc., illustrates the shift from push to pull experience.

Revolutionized search experiences only paint a portion of the bigger picture. The bigger picture is that agentic agents can act autonomously, allowing marketplaces to add value beyond user input into a search box. Marketplaces will benefit greatly when goal-oriented agents can pre-emptively and proactively perform the fundamental search and discovery tasks involved in making a transaction.

In agent-led marketplaces, you can imagine a system where:

• Buyers express their goals or preferences in natural language once or over time.

• An AI agent (or series of agents) scours the available supply, evaluates the trade-offs, and recommends the ideal match.

• This process happens invisibly, not through endless scrolling or comparison shopping. Decisions are made behind the scenes and presented as a seamless, personalized outcome.

• The preferences, results, and user behavior are saved and iterated over time by the agent to pre-emptively share and recommend products/services or remind users of an upcoming need at a future date.

• This process is repeated iteratively and inherently improves over time with repetition and enhancement.

Agent-led marketplaces have the inherent advantage of improving over time, but they also include softer and situational attributes that are difficult for faceted search to handle. Greater personalization and accuracy lead to trust and build long-term customer relationships, which is paramount for AI agents to be fully adopted by users.

For example:

• In a talent marketplace, an AI agent could match freelancers with projects based on their skills, long-term career goals, team dynamics, and culture fit.

• In e-commerce, an AI-powered marketplace could learn buyers’ tastes over time and proactively source unique items, creating a deeply personalized shopping experience.

Whitney Wolfe Herd of Bumble talks about this at length in her appearance at Bloomberg Tech earlier this year. The clip below is long (I start it at the right spot), but she talks at length about how innovation through AI will evolve the Bumble marketplace into a relationship platform. They will have AI agents that act as dating coaches and potentially even date for you.

A by-product of an inherently proactive system is that it becomes less transactional, which is paramount for dating. Goal-oriented AI opens the door to a more relational marketplace, where adaptive learning and value creation are built into every interaction. Building rapport and trust can allow customers to build a relationship with the marketplace, improving outcomes across conversion, retention, and, ultimately, marketplace revenue.

The Road to AI-Led Marketplaces

The shift to agent-led marketplaces isn’t just about adding AI—it’s about transforming or building the marketplace's core to be AI-first. NFX does a great job articulating the spectrum of AI integration.

Today, the most economically significant marketplaces are between 1, 2, or 3, with an average near 1. For existing marketplaces to cross the AI chasm, they must rebuild to be AI-first, which will inherently advantage nimble and AI-savvy newcomers. For incumbents, my experience suggests they will find it challenging to make the expensive and risky bet to refactor and disrupt the way they do business entirely. Their network effects can carry them, but they will erode over time as the AI product experiences rapidly iterate and improve. One of the caveats is whether these incumbents can maintain a unique supply base with little multi-homing.

Olivia Moore provides a nice framework for thinking about the distribution of disruption, which suggests marketplaces with unique and hard-to-aggregate supply will benefit significantly from the shift to AI.

Marketplaces where AI can replace supply are at risk of immediate disruption. We have seen this with businesses like Chegg and Stack Overflow and labor marketplaces with low-complexity digital labor products.

It is essential to realize that while certain businesses can benefit (dark green), they may not be able to harness AI promptly and effectively to prevent disruption. Conversely, companies with AI headwinds (red) may lean into AI and disrupt themselves. Outside of physical supply, there is a world where AI agents replace the supply by becoming the supply.

The response and use of AI will dictate the outcomes, and those willing to adapt will thrive. Furthermore, AI allows startup competition to run faster and more efficiently, dramatically leveling the playing field.

I don’t think all marketplaces will be a perfect fit for the agent-led model, nor is it a foregone conclusion that they will happen. I see the following barriers to the creation and adoption of agent-led marketplaces.

• Trust–Agent-led marketplaces will demand trust from users, who must feel confident that these agents act in their best interests. This challenge requires transparency, accountability, and robust safeguards. If users do not trust the agents, the AI-led marketplaces will not succeed.

• Behavior Change–People are used to the current search paradigm and chat. We have been conditioned for years to type a query into a search box and refine our search. Prompt-based search and action are inherently different and have a learning curve suited to younger generations. Humans also like to feel that they did the work sometimes, and it will be essential to balance that.

• Founders–People matter. Founders and operators have to want to solve marketplace problems using AI. Those same people need the skills to do this, which is a massive gating factor in creating agent-led marketplaces if we can’t get people to work on them.

• Data–Garbage in, garbage out. The data matters a lot in AI, and ensuring that high-quality, relevant, and topical inputs are available is critical. Dan Hockenmaier has a related piece about the “legibility” of supply, which I think is the crux of this issue. He wrote about it in the context of service marketplaces, but I think the point more broadly is that the more information you can surface about the supply, the more you can standardize the process around it. For AI, the data quality around the supply and demand side of the business will be a key variable in its efficacy.

• Speed–One of the key features is adaptive learning, but the question is whether users will provide enough data or use the product enough for it to improve quickly enough. The speed of improvement is directly correlated to user adoption, which is a chicken-and-egg problem in its own right.

• Technology–Generally, there is a risk that the technology can’t do what we hope it will do in the timeframe that we want it to. I think about actual self-driving cars, which must be near perfect before they can go to market, or blockchain, a fantastic technology with unclear product-market-fit. Both technologies keep progressing but have an ever-ephemeral feel. I believe it will happen; the timeline is uncertain.

• Funding–Investor interest in marketplaces has been waning as the last innovation window closes and the previous generation matures. It is unclear if investors are convinced this new model will yield venture-scale outcomes and put material funding behind it. Most AI-first companies will choose to raise funds as AI companies and likely not as marketplaces. I prefer to be optimistic here, but we are entering a period where marketplaces may entirely fall out of favor before resurfacing again with the new AI technology window opening.

Why Now?

Given these barriers for agent-led marketplaces, it is essential to ask: “Why now?”

The technological underpinnings for agent-led marketplaces are already here, and the green shoots are beginning to sprout. Large Language Models (LLMs) have shown they can handle complex reasoning and natural language interaction. Agents and recommendation systems are becoming more adept at understanding context, nuance, and intent. It is also increasingly clear that people like talking to agents and seek more human-like AI products, e.g., AI girlfriends and avatars.

The real challenge is not the technology but the design of systems that integrate these capabilities into a cohesive marketplace experience. Not all LLMs are great at everything, and an agent might need to use multiple LLMs for a given workflow. Platform operators must shift their mindsets. They must see AI not as a productivity tool but as a core enabler of their business model.

AI has a substantial market pull, making it the perfect time to build an agent-led marketplace for those willing to take risks. We can see the tiniest glimpses of the future, and there will likely be a significant upside for those willing to be a first-mover—a once-in-a-lifetime opportunity to build a generational marketplace.

The big question is when these marketplaces might start to appear. I am already starting to see them pop up, but they are few and far between. As a former podcast guest of mine, swyx puts it.

Expecting agent-led marketplaces to appear in 2025 magically would be a heroic assumption. This is not because they can’t, but because I think the technology is still being adopted and improved, and the human talent to pull these agents together is not ubiquitous enough. There is a very narrow cross-section of individuals who could pull this off. Given that, I predict (with an uncertain confidence interval) that we will start to see a few emerge this year, but 2026-2027 will be a ramping year for these marketplaces.

Transitioning to an Agent-Led Marketplace

What do you do as a founder and operator to start this transition to an agent-led marketplace? There are many ways to go about it, but Ryan Moser shared a few ideas and real-world examples of how ThredUp implemented AI before LLMs to power its growth.

1. Understanding SKUs and Customers—ThredUp uses AI to cluster millions of SKUs based on customer preferences, allowing it to understand supply/demand imbalances.

2. AI-Driven Pricing—Listing prices are set intelligently using AI, which analyzes recent sales and pricing to optimize for revenue in a dynamic resale market.

3. Dynamic Price Adjustments with Reinforcement Learning—AI systems gradually lower their prices for items that don’t sell immediately. A reinforcement learning model provides continuous feedback, improving the pricing system over time and balancing revenue with inventory velocity.

4. Supply and Demand Optimization—ThredUp’s AI monitors supply and demand signals to optimize operations. For example, the system prioritizes high-performing supply sources while limiting consistently underperforming sources.

5. Personalized Search and Sorting—Customer embeddings are continuously refined using browsing and search history data. This personalization improves search and filter rankings, making finding relevant inventory in an extremely deep catalog easier.

6. Smarter Visual Search with CLIP Models—ThredUp employs CLIP (Contrastive Language–Image Pretraining) models to extract concepts from product photos. This enables richer search terms and improves discoverability, giving customers a more intuitive shopping experience.

Of course, many other areas have clear applications for agentic agents, but these are real-world applications of AI in marketplaces today. This section provides a taste of what is possible and warrants its own post on this topic.

Conclusion

Agent-led marketplaces represent a fundamental shift in how we think about commerce and connection. By leveraging AI’s unique strengths in aggregation and matching, these systems have the potential to move beyond efficiency gains to create entirely new forms of value.

The transition won’t happen overnight, raising important questions about trust, fairness, and control. But the potential is too great to ignore. As AI continues to evolve, tomorrow's marketplaces may no longer be places where buyers and sellers meet—they will be intelligent systems that anticipate our needs, act on our behalf, and make matching supply and demand as seamless as possible. They might even replace the supply they set out to aggregate.

AI is the future of marketplaces. And it’s closer than we think.

🚨 Request for Startups: If you are building or considering creating an agent-led marketplace, please email me at colin@yonder.vc. 🚨

Thanks for reading!

Please don’t forget to share this post and subscribe/follow me on X and LinkedIn for other great marketplace content. If you want to book a call, you can find me on Intro.

And lastly, a HUGE thank you to Casey Winters, Ryan Moser, Dan Hockenmaier, Jeremy Burton, Seann Stubbs, Lucas Dickey, and Carlos Caro for their feedback. 🙏

---

The tool of the week is…

folk - Finally, a CRM that doesn't overcomplicate it.

Backed by investors like Accel, Folk is a CRM that builds genuine client connections that are simple to use and integrate. It covers all your CRM needs: Pipeline Management, Chrome Extensions, Dashboards, one-click Enrichment, Integrations, and Messages and Sequences.

I use it for all of my fundraising for Yonder. If you would like a 20% off discount code, please restack 🔄 this post, and I will message you directly with the code.

---

About Me:

Colin is a marketplace geek and the General Partner of Yonder, a pre-seed marketplace fund that invests in marketplaces that create new economies. He has also been a long-time advisor to marketplaces, helping them with product growth, monetization, liquidity optimization, and strategy. Previously, he was the CPO/CRO at Outdoorsy and worked at Tripping.com, Ancestry.com, Justanswer, and the Federal Reserve.