Large Language Marketplaces - a new paradigm in commerce

Will the LLMs transition from models to marketplaces?

Not only do I think they are capable of being the intermediary of most of our commerce, I believe this is their end game.

The question may make you think of eBay in the early aughts, but that comparison doesn't do it justice.

At its peak, eBay had ~180 million active buyers in 2018.

ChatGPT reported having nearly 900 million weekly active users (over 2 billion monthly visits) at the end of 2025 and Gemini surpassed 750 million monthly active users in February of this year.

My belief is not just based on LLM usage numbers, it's also founded in the growing coverage in use cases.

Let's start with the shift in search itself - a component of the original LLM use case of information sourcing.

According to AI-powered digital experience platform Optimizely, last October 75% of queries were answered on Google without a click-through thanks to AI answers, with 100M users turning on AI mode (no more blue links).

I can only imagine these numbers have grown since then.

AI has converted search engines into answer engines - bypassing the need to click through to a site in many cases. This has injected the LLMs as not only an intermediary between information seeker and source, but has positioned them to be the curator and provider of answers to the users' questions themselves.

The nature of how we learn, connect, and buy is changing.

Search was always about connecting the user to the source and getting paid for placement or clicks, but this is much bigger than a transfer of link sourcing (and ad spend) from the traditional search engine to the LLMs (Google is going to be just fine here).

There is also a shift on where the transaction occurs and Retail and Commerce industries are leading the way. They represent 27.5% of global LLM adoption and are the proving ground for “Agentic Commerce” where we see early integrations from Walmart, Instacart, and others that enable frictionless purchases originating from the chat interface of the LLMs.

They're laying the groundwork for commerce to occur directly from the LLMs.

This shift is not limited to ecommerce.

By adding Agentic capabilities, the LLMs have positioned themselves to be where much of the "work" will take place as well. And the definition of work expands to almost any task a human or business may need to complete.

Over 3 million custom GPTs – mini agents – have been created through OpenAI’s models. These agents handle just as many use cases spanning from personal to organizational, private to external-facing, administrative function to product powering and value delivering.

Anything that requires coordination with other parties can be automated, improved, and optimized online and through these agents powered by the LLMs. This coordination will largely be done through vendor, customer, and platform agents working on behalf of this new system of work.

And these systems learn . . .

They learn from doing the tasks, they learn from their users, they learn from the compounding effect of data. The LLMs capture much of this data, setting them up for further expansion in capabilities and scope.

Plan and book a trip? Build a product (software or tangible)? Design a website, or even a building? Manage inventory or pull business analytics and reporting? Run a marketing campaign? Orchestrate and facilitate the family's entire set of predictable needs, not just their calendars?

All eventually can and will be done through these agentic ecosystems. The term "ultimate broker" comes to mind.

This will require an obscene number of integrations and a common operating language (MCP entered the chat). But if billions of people are already spending most of their time within the LLMs building and initiating their "workflows" through an army of agents, then this will force the hand of literally any business to set up virtual shop as close to their customers as possible.

Competition and FOMO are real dynamics at play that will accelerate capabilities and adoption - a flywheel that will perpetuate itself and make the LLM evermore sticky and central to our existence.

And the world will gravitate towards a model of frictionless access and consumption and will pay for that convenience and the scaling value it brings. They'll pay for the access, the coordination, and the outcomes - much of which will go through the LLMs.

But what about the things robots can't do yet?

Anything requiring physical presence to complete will still be done in person, but capabilities will be built around the LLMs to augment such work.

How about fixing a leaky pipe? A plumber can certainly do that without any help.

Sure, the experienced plumber can do it in their sleep, but what about the apprentice?

OR - hear me out here - what if the plumber leverages agents within the LLM system to track cold fronts, map out the ages and architecture of the homes in their service area and use that data to predict not only the anticipated number of calls (i.e. busted pipes) they'll receive, but what type and specification of materials they'll need and have them ordered and shipped just in time for that cold front? And then, of course, automatically initiate and time a targeted online campaign around the "busted pipe" problem.

In this future world, that plumber was likely found, booked, tasked, routed and paid through the common interface as well.

The LLM-powered agents enabled a small business to access a deep stack of advanced technology and marketplace capabilities that it would have otherwise never stitched together due to traditional legacy constraints (time, know-how, multi-vendor coordination and costs).

See what I mean?

Numerous underlying services and capabilities underpin the performance of that single plumber. There will still be several independent providers involved in the scenario above (and likely one who pulled it altogether on the LLM platform), but neither the plumber nor his customer will have to pick up the phone, log into numerous apps, and work each piece of the flow individually.

The system handles the coordination for them - it removes any constraints - IF they've built or evolved their businesses to operate in this system.

This new technology marketplace will touch everything.

Controlled access, enablement, and brokering of all goods and services is the north star and the reward for these LLMs and the reason trillions of dollars are being funneled to them by the venture world.

What about everyone else? Will the other platforms and pure-play businesses have lost all leverage with their customers and their wallets?

No, not necessarily. But it's going to take some soul searching and focused intent to adapt and thrive in this new system.

There is still value to be delivered and reasons a customer should choose your product or solution over others - even when the entry point has been consolidated and even if the LLMs themselves are capturing margin from the coordination.

And customers still need to be aware of your product and have frictionless access to it.

First step?

Identify your competitive advantage, the problems you can solve for customers in an advantageous way that can't easily be replicated - certainly not through a prompt . . . unless that prompt leads to your service.

And then with the new market lens, determine what your product or service looks like, how your customer finds you, signs up, and realizes value - and what underpinning capabilities, systems, and integrations are required to deliver this new model.

For the companies whose primary strategy with AI is productivity gains, I have bad news for you. You're already losing the game, because you're missing the larger impact of AI altogether.

This is where real strategy work comes into play and it's going to be fascinating to see how creative companies get with their value propositions, offerings, and go-to-market strategies. A whole new generation of HBS case studies are coming our way and they won't be forged by following the old proven business models.

When I said "creative" companies, I meant it.

This is an opportunity for all of us to rethink the problems we're solving for and how they can be addressed in completely different, better ways within this new system of work.

Having a competitive moat will be key to thriving in the LLM marketplace, because when a prospective customer can research, compare, price, access and test a product within minutes, having a clear, compelling, and defensible reason why you are the best choice will be the difference between success and irrelevancy, regardless of whether or not they're using the LLM to interface with your service.

Creativity will be needed to survive. This was always the case, but there was a larger margin of error in the past – there was more time to find your way out of a competitive squeeze.

I think with the speed of disruption those who can bring creativity to their work and their solutions will be in high demand in this new market. Thinking outside the box around market problems and the customers who have them will be needed to dream up a better system that removes their constraints at a deeper level than anyone else.

Afterall, the new marketplace will reward the businesses that bring clarity to what problem they're solving, for whom, and how it's superior to anything else out there.

The future favors the bold (and the creative).