Three months ago I published my annual predictions, and while I rarely revisit them in the middle of the year, I do want to note an interesting development related to prediction #3, which states: “2025 will not be the year AI agents take off.”
Here’s what I said back in January:
“As the bloom came off the Generative AI rose in 2024, everyone started talking about AI agents as the Next Big Thing. Google, Apple, OpenAI, Microsoft, Meta, Amazon — all of them (and about a million startups) are trying to build user agents for both enterprise and consumer use cases. I’m a huge fan of the concept, but for now, it remains just that. Reasoning agents that book your travel, negotiate your insurance bills, or manage your calendar simply will not work if they are beholden to the same business models currently driving Big Tech. There’s so, so much to say about why this is true, but I’ll leave that for another series of posts.”
Indeed, there is a lot to say. I’ve struggled with how to cover this without getting stuck in the messy details of an emerging industry, and for the most part, I’ve just stayed silent. But this piece in The Information — Anthropic’s Not-So-Secret Weapon That’s Giving Agents a Boost — is spurring me to clear my throat.
The piece focuses on MCP, an open source software standard Anthropic developed last Fall. MCP stands for model context protocol, which, in the company’s words, seeks to become “a new standard for connecting AI assistants to the systems where data lives, including content repositories, business tools, and development environments.”
The company goes on to explain the problem MCP is designed to address: “…even the most sophisticated models are constrained by their isolation from data — trapped behind information silos and legacy systems. Every new data source requires its own custom implementation, making truly connected systems difficult to scale.”
Put another way, there’s data, data, everywhere, but without connectivity, there’s not a drop to drink. Want your nifty new AI agent to book a flight for you? Well, it’ll have to work with, let’s see … every major airline’s online systems, every major payment system, every major travel platform, every major calendaring system, and … well, that’s enough to confound your average AI developer right there. For every single possibility, a developer would have to code a custom programming interface, not to mention get their business colleagues to negotiate a deal with each company to access the data in the first place. Those kinds of hurdles are near impossible to overcome for most startups.
This is why, at present, you don’t have an AI agent doing much of anything for you, nor will you anytime soon.
Now, Very Large Technology Companies do have relationships with many, if not most of the players I listed in the example above. With enough determination, resources, and product panache, they could brute force AI agents into the world. Google, Amazon, Meta, Apple and Microsoft are the obvious places one might expect AI agents to first emerge. So why haven’t they happened?
I can think of a few reasons. First, big tech companies are terrible at consumer-facing product innovation. Name one new product from any of the big five listed above that has quickened your blood in the past few years? Sorry, Teams doesn’t count!
Second, big tech companies don’t like to share with others. They may have deals with every payment, travel, and airline company on earth, but those deals don’t contemplate a novel consumer service like a flight-booking AI agent. Re-negotiating those deals will take months and months of haggling between business development, legal, compliance, product, and executive teams, each of which will want to extract their particular pound of flesh (not to mention the lifting of legs). I call this death by biz dev, and trust me, it’s a thing.
But most importantly, there’s the overarching business models of the big tech companies, and in today’s technology landscape, business model is destiny. How might our flight-booking AI agent impact advertising revenues at Google Flights, not to mention search? How might it impact engagement and commerce revenues on platforms like Instagram or WhatsApp? Might our industrious AI agent lay bare Apple’s pretensions around consumer privacy, pretensions which drive its entire marketing model for iOS and the iPhone? I could go on, but I think you get the picture: Faced with the uncertainties such agents present, these large companies are most likely to respond with inaction. If they do ever create an AI agent capable of executing our simple vision, we can be certain of one thing: That agent will not work for us. It will work for them — and in doing so, it will fail as a consumer product.
But … there is another model of how novel innovations spring forth in our technology ecosystem. That model revolves around industry standards — fundamental ways of doing things that everyone agrees to, such that developers and entrepreneurs — and ultimately consumers — can create and benefit from new products and services. The Internet is based on an open set of protocols that no one company controls. Ditto email, texting, and even the cellphone network. When it comes to doing complex, data-driven actions across a massive network of loosely coupled organizations (like travel, for example), the only approach that can scale and thrive will be one based on an open, neutral standard.
I don’t know if MCP is that standard, though The Information reports it’s starting to take off. Suffice to say the world needs a protocol like MCP, and the sooner the better. But will we get it? That remains to be seen, but given the money and power at stake in the AI world today, it’s likely that Apple, Google, Meta, Microsoft or Amazon aren’t going to support MCP anytime soon. There are plenty of examples of good standards finally winning out over time (USB-C comes to mind), but it often takes years, if not decades, for the big guys to finally come around. In the meantime, I think my prediction is safe. Despite all the excitement about AI agents taking over the world, I don’t think it’s gonna happen anytime soon.
—
You can follow whatever I’m doing next by signing up for my site newsletter here. Thanks for reading.