Every major technology wave arrives with the same promise: this time, the middleman dies. AI was supposed to be the final blow, intelligent systems that connect you directly to what you need, no gatekeepers, no tolls, no unnecessary layers. But if you look at what's actually happening, the pattern is running in the opposite direction. The middleware layer is growing, not shrinking. And the companies building it are capturing more value than the ones on either side.

The disintermediation fantasy

The pitch is always the same. New technology reduces friction, which means buyers and sellers can find each other directly, which means intermediaries become obsolete. The internet was supposed to eliminate travel agents, stockbrokers, and retail middlemen. Blockchain was going to remove banks, notaries, and clearinghouses. AI is supposed to cut out everyone from recruiters to customer service reps to software consultants. And yet, every time the dust settles, the intermediaries are still there. They just look different. The travel agent became Expedia. The stockbroker became Robinhood. The retail middleman became Amazon. The function didn't disappear. It migrated to a new form, often one that's more powerful and harder to displace than the original.

The middleware companies that won

Consider three companies that emerged from the internet era: Stripe, Twilio, and Shopify. None of them make the thing their customers are selling. None of them are the end consumer. They sit in the middle, and that's exactly why they won. Stripe is a payments middleman. It sits between businesses and the banking infrastructure that moves money. Before Stripe, accepting payments online meant navigating merchant accounts, payment gateways, PCI compliance, and a tangle of banking relationships. Stripe collapsed all of that into a few lines of code. It doesn't move the money itself, the card networks and banks do that. But it owns the abstraction layer, and that layer now processes over $1 trillion in payments annually. Stripe's valuation hit $91.5 billion because being the middleware between commerce and finance turns out to be extraordinarily valuable. Twilio is a communications middleman. It sits between businesses and the global telecom infrastructure. Before Twilio, sending an SMS or making a voice call programmatically meant negotiating carrier contracts, managing phone number provisioning, and handling the complexity of telecom regulations across 180+ countries. Twilio turned all of that into an API. It doesn't own the phone networks. It owns the developer-friendly layer on top of them, and that layer now handles over 2.5 trillion communications annually. Shopify is a commerce middleman. It sits between merchants and the internet. It doesn't manufacture products or buy inventory. It provides the storefront, the checkout, the payments infrastructure, and the logistics coordination that lets anyone sell online. Shopify's merchants generated over $270 billion in GMV in 2024. The company captures a percentage of that by being the layer that makes selling possible. The pattern is identical in all three cases. A complex, fragmented infrastructure exists on one side. Customers who need to use that infrastructure exist on the other. The middleware company builds the abstraction layer that makes the connection simple, reliable, and scalable. Then it charges a fee on every transaction that flows through.

Railroads, ad networks, and the historical pattern

This isn't new. The pattern predates the internet by more than a century. When railroads connected American cities in the 19th century, they didn't eliminate intermediaries. They created a new one: the freight broker. Before railroads, goods moved slowly and locally. After railroads, goods could move anywhere, but someone needed to match shippers with carriers, negotiate rates, and coordinate logistics across a suddenly vast network. Freight brokerage became an entire industry. The Motor Carrier Act of 1935 formalized the role, and deregulation in the 1980s turned freight brokerage into the dynamic, multi-billion-dollar market it is today. The internet did the same thing with information. It didn't eliminate the middleman between content creators and audiences. It created the ad network, arguably the most profitable intermediary in economic history. Google doesn't create content. It organizes access to it and sells attention along the way. Facebook doesn't produce the posts and photos that keep people scrolling. It provides the platform and monetizes the engagement. These companies are middlemen at a scale that no previous intermediary could have imagined. The mechanism is always the same. A new technology expands the number of possible connections between two sides of a market. That expansion creates complexity. And complexity creates demand for someone to sit in the middle and manage it.

In AI, the orchestration layer is the new middleware

AI is following the exact same pattern. The technology is powerful, but it's also complex. Foundation models from OpenAI, Anthropic, Google, and Meta are the raw infrastructure, analogous to the railroad tracks or the telecom networks. End users need AI to do useful things in their specific contexts. And between those two sides, a massive middleware layer is forming. LangChain raised $25 million and then $25 million more because it provides the orchestration framework that helps developers chain together AI model calls, tool use, and memory management. It's the plumbing layer between the model and the application. Vercel's AI SDK does something similar for frontend developers, providing the connective tissue between AI models and the web interfaces that users actually interact with. The Model Context Protocol (MCP), introduced by Anthropic and now governed by the Linux Foundation, has become the standard for connecting AI agents to external tools, with 97 million SDK downloads and over 10,000 active servers. Even products that don't look like middleware are playing the middleman game. Notion sits between your team's knowledge and the AI models that can reason about it. Cursor sits between developers and the code-generation models that assist them. Perplexity sits between users and the search results that AI can synthesize. In each case, the value isn't in the model or the user. It's in the layer that connects them. The winners in AI aren't the model makers or the end users. They're the companies that own the orchestration layer, the abstraction that makes AI usable, reliable, and integrated into real workflows.

The "just use the API" fallacy

There's a recurring argument in technical communities: why pay for a wrapper when you could just use the API directly? It's the same argument people made about Stripe ("just integrate with the bank yourself"), Twilio ("just negotiate a carrier contract"), and Shopify ("just build your own e-commerce site"). The argument is technically correct and commercially irrelevant. Yes, you can call the OpenAI API yourself. You can manage your own prompt templates, build your own tool-calling infrastructure, handle your own rate limiting, implement your own error recovery, and maintain your own evaluation pipeline. You can also change your own oil, but most people don't. What the "just use the API" crowd consistently underestimates is how much value sits in three things: abstraction, reliability, and developer experience. Abstraction means you don't have to understand the underlying complexity. Reliability means someone else is handling the edge cases, the failures, the version migrations. Developer experience means you can ship in days instead of months. These aren't nice-to-haves. They're what Stripe, Twilio, and Shopify were built on. And they're what the AI middleware layer is being built on now. The companies that package complexity into something usable aren't adding unnecessary cost. They're capturing the value of saved time, reduced risk, and faster iteration.

Aggregation theory, updated

Ben Thompson's Aggregation Theory, first published in 2015, explains how internet-era companies win by aggregating demand. Google aggregates search intent. Netflix aggregates viewing attention. Uber aggregates ride demand. The aggregator controls the user interface, commoditizes suppliers, and captures the margin. AI is shifting this framework in an important way. The new aggregators aren't just aggregating demand. They're aggregating capabilities. Consider what an AI orchestration platform does. It doesn't just connect users to one model. It connects users to multiple models, tools, data sources, and workflows, selecting the right combination for each task. The orchestration layer aggregates the capabilities of the entire AI ecosystem and presents them through a single interface. This is a more powerful position than traditional demand aggregation. A search engine aggregates web pages, but the web pages exist independently. An AI orchestrator aggregates capabilities that often can't function independently, models need tools, tools need context, context needs memory, and memory needs persistence. The orchestrator is the only layer that holds the full picture together. As one venture analysis put it, we're moving from systems of record to systems of action. The middleware layer isn't just storing information or routing requests. It's making decisions about which capabilities to deploy, when, and how. That's a fundamentally different kind of intermediation, and it's harder to displace because the value compounds with every integration added.

Distribution is the real product

There's a reason why being the middleman keeps winning, and it has nothing to do with rent-seeking or market manipulation. It has to do with a basic economic reality: distribution beats product. It doesn't matter if your product is better if nobody can find it or use it easily. The middleman's core function is distribution, connecting producers to consumers in a way that's convenient, reliable, and scalable. That function is genuinely valuable, and in most markets, it's more valuable than the thing being distributed. This is why Spotify is worth more than most record labels. Why the App Store captures 30% of every sale despite writing none of the apps. Why Google captures more revenue from content than the people who create it. Distribution is leverage, and the middleman owns the distribution. In AI, the same dynamic is playing out. The companies with the strongest distribution, the ones embedded in existing workflows, integrated with existing tools, and sitting at the point where users already are, will capture more value than the model makers. OpenAI knows this, which is why they're building ChatGPT into a platform rather than just selling API access. Google knows this, which is why Gemini is being woven into every Google product. The model is the commodity. The distribution layer is the business.

The middleman isn't a bug

It's tempting to be cynical about this. Another generation of companies that own nothing but charge for everything. But intermediation isn't a market failure. It's a market function. Middlemen exist because they solve a real problem: they reduce friction, aggregate options, manage complexity, and save people time. The freight broker existed because coordinating shipments across a railroad network was genuinely hard. Stripe exists because payments infrastructure is genuinely complicated. The AI orchestration layer exists because connecting models to tools to data to users in a reliable way is genuinely difficult. The criticism that middlemen are rent-seekers has some truth in specific cases. Platforms that lock in users and then raise fees are extracting value rather than creating it. But the net function of intermediation is positive. Without Stripe, fewer businesses would accept online payments. Without Twilio, fewer applications would have communication features. Without the AI middleware layer, fewer organizations would be able to deploy AI in production. Every technology disruption is supposed to cut out the middleman. Instead, it creates new chokepoints, new complexity, and new demand for someone to make sense of it all. The middleman changes form with every wave, from freight brokers to ad networks to AI orchestrators, but the role persists because the underlying need never goes away. The only question is which middleware layer you're building, or which one you're paying.

References