Over 15 Years of APIs with Kevin Swiber
TL;DR
Kevin Swiber has worked across the API ecosystem, from designing gateways at Apigee and Postman to advising investors, startups, and enterprise architects. On The Good Thing he argues that infrastructure is getting boring, and that boring is what teams want, because nobody wants to wake up at 2am to an outage. Autonomous team models in large organizations produce sprawl instead of innovation, one company paying three vendors for seven address-verification APIs, which is why governance and a real launch process matter. AI and MCP are moving fast, but most teams are still early, and LLM agents do not remove the need for APIs: MCP servers still call out to external services, so API creation moves to a different layer and usually produces more APIs. Swiber judges API companies on whether they are credible, relevant, and differentiated, and expects hypermedia patterns to return through MCP interfaces even if nobody calls them hypermedia.
Kevin Swiber has worked across nearly every corner of the API ecosystem: from designing gateways at Apigee and Postman to advising investors, startups, and enterprise architects. Now operating independently, they bring a rare 360° view of the API space.
He joined Stefan and Jens on the latest episode of The Good Thing to talk about Model Context Protocol (MCP) , AI’s growing role in developer tooling, and why infrastructure is getting boring again—in a good way.
Swiber opened the conversation with a reflection on their shift from vendor-side to consulting.
When you are in that vendor space, it’s kind of easy to forget that you’re not the center of the customer’s world.
As they explained, startups often treat every deal as urgent, while enterprise buyers are juggling “15 different initiatives”, and your product might be a small part of just one.
You might get to a no decision on the deal because that person has changing priorities and you’re just a small part of their world.
Despite the buzz around AI and MCP, Swiber reminded listeners that many teams are just getting started.
Normal people who aren’t in tech… they don't have 15 claude code instances running and they aren't trying to automate their whole lives. They’re just trying to figure out, ‘Okay, I’m using ChatGPT. Now my boss wants to use ChatGPT. What does that look like?’
Even highly technical teams are still exploring tools like GitHub Copilot , Cursor , and Windsurf .
As a product company, there's always in the back of your mind: Am I going to be irrelevant in the next five months?
Swiber advises investors too, and says their AI-related questions aren’t rooted in fear, but in clarity:
They’re asking, should we be investing more money? Should we be advising our portfolio to start doing more with AI or MCP?
His framework for evaluating API companies is simple: “credible, relevant, differentiated.” That lens matters even more now that infrastructure is seen as utility grade. It's stable, interchangeable, and no longer a source of competitive edge.
Swiber described the side effects of distributed ownership models in large orgs.
I work with tons of companies who say, ‘We went to autonomous teams… and now we have no visibility. We have seven address verification APIs, and we’re paying three different vendors on the backend.’
The result isn’t innovation but sprawl. And sprawl needs governance.
Let’s go back to the process. How do we actually launch software? How do we get the most out of the investments we’ve already made?
Some believe LLMs will replace the need for structured APIs, but Swiber doesn’t buy it:
Every time we say we’re trying to get out of doing APIs, we end up just creating more APIs.
As LLM agent architectures evolve, teams often try to bypass traditional API design. But in practice, MCP servers still need to call out to external services. That means the need for APIs doesn’t go away, but moves to a different layer. They often centralize functions and serve multiple consumers, creating new APIs in the process.
Swiber is known for creating Siren , a hypermedia type for APIs. When the conversation turned to LLM-powered agents, Jens asked, “Isn’t this finally hypermedia’s moment?”
Swiber sees those patterns returning, even if no one calls them hypermedia anymore.
We probably won’t call it hypermedia anymore. But the patterns—we’re coming back around to them.
They pointed to MCP UI flows where tools return responses containing forms. This echoes classic hypermedia patterns, just not by name.
There’s this whole MCP UI track… where a tool response comes back that essentially has forms in it. So you know what to do next.
What happens when LLMs become the primary interface to the internet? Swiber likened the experience to the early web: stripped-down, text-based, and impersonal.
We’re losing the entire experience of the web that someone has crafted for us.
That simplicity has benefits, but also trade-offs.
What happens to ad revenue in a ChatGPT world? And for the folks who went too far, showing me five ads before I can read an article—you better believe I’d rather go to ChatGPT.
The shift puts pressure on publishers and platforms alike.
There's now a third party in the mix. And everyone's kind of beholden to that third party. What information are they going to share with me?
Asked about trends in platform engineering, Swiber didn’t hesitate:
People want boring infrastructure. Because they don’t want to wake up at 2am and hear everything went down.
Jens echoed the sentiment from his own experience:
We tried fancy vendors. Then we moved to GCP Kubernetes. It’s more expensive—but we’ve had zero issues in two years.
Sometimes the best tools are the ones you don’t think about at all.
Near the end of the episode, the conversation turned reflective. Stefan and Jens asked Swiber how their thinking has changed over time.
Like many seasoned engineers, they described their journey through the hype cycles.
First, you think you have it all figured out. Then someone says, ‘Why do you have a factory-factory-factory pattern?’ It’s too complex. Eventually, everything becomes simple again.
And as for GraphQL? It’s not going anywhere—but it’s also not everything.
GraphQL went from ‘this is all I’ll ever need’ to ‘here’s where it fits.’ Same with LLMs and MCP. It’s additive.
🎧 Want to hear the full story? This episode covers even more—including GraphQL fatigue, the value of analysts like Gartner, and what enterprise architects are actually buying today.
Listen now on YouTube .
Frequently Asked Questions (FAQ)
No. As Kevin Swiber puts it, every time we say we are trying to get out of doing APIs, we end up creating more APIs. Teams building LLM agent architectures often try to bypass traditional API design, but MCP servers still need to call out to external services, so the work moves to a different layer and often centralizes functions for multiple consumers, creating new APIs along the way.
Swiber describes companies that moved to autonomous teams and lost visibility as a result, one of them running seven address-verification APIs while paying three different vendors on the backend. The result is sprawl rather than innovation, which is why he points teams back to the process: how software actually gets launched, and how to get more out of the investments already made.
Boring infrastructure is stable, interchangeable, utility-grade technology, and Swiber says people want it because they do not want to wake up at 2am and hear everything went down. Jens gave WunderGraph's own example: after trying fancy vendors, the team moved to GCP Kubernetes, which costs more but has had zero issues in two years.
Kevin Swiber uses three questions: is the company credible, relevant, and differentiated. That lens matters more now that infrastructure is treated as utility grade, stable and interchangeable, and is no longer a source of competitive edge on its own.
Swiber, who created the Siren hypermedia type, sees those patterns coming back even though we probably will not call it hypermedia anymore. He points to the MCP UI track, where a tool response comes back with forms in it so the client knows what to do next.
Swiber likened it to the early web: stripped-down, text-based, and impersonal. As he put it, we are losing the entire experience of the web that someone has crafted for us. He also raised the open question of what happens to ad revenue in a ChatGPT world, and noted that a third party now sits in the middle deciding what information gets shared.
Editor’s Note: All quotes in this article are attributed to Kevin Swiber from Episode 25 of The Good Thing podcast. Some quotes have been lightly edited or paraphrased for clarity and readability.
Content Manager
Brendan Bondurant is the Content Manager at WunderGraph, owning technical content across GraphQL Federation, API tooling, and developer experience. He partners with leadership on product and company messaging and works cross functionally to align positioning, terminology, and content strategy across channels.
