Agentic mashups - just like getting the band back together...

The rapid pace of change and excitement surrounding the Model Context Protocol (MCP) and agentic AI bear a fascinating similarity to the shift that occurred in the early 2000s around APIs and web applications. In both cases, a new, lightweight approach to integration ushered in a paradigm shift that ultimately changed the way enterprises built applications.

Developers were suddenly excited about how data could be woven into web mashups, much like musical mashups, but for creating new user experiences. Eventually, the new lightweight API integration patterns completely overtook the stodgy Service-Oriented Architectures (SOA) popular in enterprises to dramatically speed the composability, cadence, and capabilities of enterprise apps. MCP might do the same, but for AI.

I recently spoke with John Musser, who became an accidental cartographer of the API landscape in 2005 in the uncharted territory. He created ProgrammableWeb as a guide to the few available APIs, which eventually expanded to tens of thousands of APIs. The site became the go-to guide for developers trying to keep pace with the technical, security, and governance challenges of the new API economy.

ProgrammableWeb was eventually sold to Alcatel-Lucent in 2010 and then acquired by MuleSoft (now part of Salesforce) in 2013, before being shut down in 2023. Musser continued consulting on the API economy and is now Senior Director of Engineering for Ford Pro, Ford’s new global commercial services business. Musser says he is having a moment of Déjà vu following the rapid adoption of MCP close to the 20^th anniversary of ProgrammableWeb:

It's almost like the bands are back together in a funny way, where so many of the people who were in that period are struck by and engaged by the MCP AI era. This whole notion of what's going on today in 2025 has so many similarities to what was going on twenty years ago. It's a bit of a cultural moment, if you will. The same but different.

Birth of the programmable web

ProgrammableWeb emerged when Musser was working on startup ideas that could benefit from the new APIs being developed by eBay and Amazon. In these early days, many of the APIs were built using the Simple Object Access Protocol (SOAP) popular in SOA enterprise governance schemes. A lightweight alternative called Representational State Transfer (REST) was starting to emerge.

Musser was surprised there was no website where he could find out about APIs, so he built his own. Over the next couple of months, he documented the thirty different APIs he had found. In these early days, developers were starting to discover a new paradigm for mixing and resampling data into an entirely new construct called web mashups. Musser recalls:

Mashups were kind of a brand new thing in 2005. Housingmaps.com from Paul Rademacher, was kind of like the quintessential, very first mashup. Google Maps didn't even have an official API, but being a good engineer, he figured out how to combine Craigslist listings with Google Maps and put them together, which was essentially the first web mashup. Within a few years, APIs had sticking power. Mashups may have faded as a phrase, but APIs certainly took off, and ProgrammableWeb became a hub for APIs on the web. We tracked the genesis of this development over the next decade. As APIs grew and became more popular, developers relied on what we had built in our community through news and as a directory.

Getting the band back together

Musser now sees a similarity in the recent birth of MCP. It's similar in the sense that it’s the new thing to do and a way to integrate. The big difference is that MCP is for weaving together AIs rather than APIs and can utilize tools and other resources to accomplish new things.

Musser connected with many of the pioneers of the early web mashup economy and now sees them making a comeback with MCP. One was Pamela Fox, who is now Principal Cloud Advocate at Microsoft. Fox had an uncanny ability to win mashup hacking contests in the early days of APIs while a student at the University of Southern California:

Companies would have mashup contests where they have prizes to use their APIs in innovative ways. Pamela Fox could create a mashup that was like the contest-winning one almost every single time in a night in her dorm room back when she was younger. It was kind of amazing. She went on to get a job at Google, which seemed like a good home for her. But lo and behold, there she is, observing last week how MCPs are like mash ups, but the people doing the mashups aren't people. The mashers are AIs, and the developers are now the managers of the mashers. So, it's this sort of funny evolution of what APIs have become, what we're facing here in 2025.

Another API mashup pioneer was Oren Michels, who co-founded Mashery in 2006 to bring governance and security to the wild west early days of mashups and web APIs. Michels recently founded Barndoor.ai as an encore of the same performance but for agentic AI:

Mashery was arguably the first API management company twenty years ago. There was no such thing as API management in 2005, which enterprises take for granted now. During that early API era, API management companies like Mashery and Apigee emerged to help companies manage, govern, rate-limit, and secure APIs. Now, in 2025, we have a new opportunity for startups to help companies manage, govern, and maximize the benefits of agentic AI. There is an absolute need for that. Notably, you have someone who’s had experience in that era twenty years ago, like Oren, who founded Mashery and successfully exited that company to create the new version of what we need today, which is AI MCP management for enterprises, because that is going to be an important space in the tools ecosystem.

The death of mashups

In the early days of what came to be known as Web 2.0, mashups were the hip programming paradigm. Rockstar developers were busy sampling web reality to create a cool new art form. However, people eventually stopped using the term mashup, even though the techniques became more widely used and common. People still talk about mashups in music, video, education, and other art forms. So what happened? Here’s Musser’s take:

Mashups were the hot thing. The phrase ‘mashup’ started with music, and in the mid-2000s, people were mashing up music, which was a subgenre of taking two or more songs and combining them together. And then they got mapped to what people were doing with these early APIs. The heyday of web mashups was probably 2006 or 2008, as they were new, novel, and exciting, and people had never done this before. It was an expression of creativity among developers, allowing them to use their imagination, both in sometimes silly and completely useless ways, and in others that were extremely useful.

So it went from being a novelty to being a ‘Yeah, so what about it? That's how we build stuff, right? We do integration. We use APIs. So what?’ And then mobile came along, and I think the death of the phrase ‘mashups’ got usurped by ‘building an app.’ So, I'm no longer building mashups, I'm building apps. Everything in mobile requires APIs. With the birth of the iPhone and mobile and smartphones, you couldn't really build any smartphone app without using sort of web-based APIs. So, the novelty had worn off, and you call these things apps, you wouldn't call it a mashup.

My take

The paradigm shift involving the birth and death of mashups had a parallel evolution on the enterprise IT side, as both converged into a middle ground that we now take for granted. But in the process both the terms mashups and SOA have vanished.

When lightweight web APIs emerged in the early 2000s, they ushered in a new paradigm, but they did not immediately replace the old-school, governance-heavy APIs used in enterprises. Musser reminds me that most websites or apps maintained both the old-school SOA APIs and the new REST APIs for quite some time. However, as governance, security, and observability infrastructure improved, even enterprises began to move away from the old SOA approaches. When was the last time you heard the term Enterprise Service Bus?

I started writing about technology in the early 1990s, when it felt like the Internet was slowly transitioning from an academic experiment into a new art form of linked web pages. In the early 2000s, I started writing for SearchSOA, when Integration always seemed like a cumbersome exercise with an alphabet soup of integration protocols and patterns that moved at a glacial pace. In the thick of it, my editors would assign me stories about some obscure new protocol or challenge and from within this paradigm, it was difficult to see why things seemed to move so slowly.

There was a palpable sense that enterprises were looking for something easier. My editors must have assigned me to write about some variant of “Is SOA dead… long live SOA,” for the better part of a decade. I saw the writing on the wall in the mid-2010s when microservices took off and SearchSOA was briefly renamed SearchMicroservices before eventually becoming SearchAppArchitecture.

Looking at these agentic AI systems in 2025, it seems apparent that they are ushering in an entirely new kind of mashup for resampling different flavors of intelligence. Whatever comes from it will normalize the fast pace of early pioneers with enterprise governance, and may usher in a new term for this status quo as well.