EP30: AI Agents vs SaaS: The Future of Software with Kenneth Auchenberg

So since you've moved to innovation endeavors, you wrote a piece that's really, really interesting called A Decade of Software Infrastructure and the things that you've learned and everything else.

And you developed this concept of agent-first infrastructure, which I think is super important for where we're going.

Tell us a little bit about what that means, what's the insight that you have, and how are people like us that are starting companies, how should we think about this?

Yeah, I think maybe we can take a step back on what's kind of happening and what has happened to software engineering.

I think I just talked about VS Code, a product we built more than a decade ago.

I'm not using VS Code anymore because my agent is writing the code.

So we are fundamentally a different kind of reality when it comes to writing software, is that when you're building an infrastructure company today, you're not really targeting the individual developer anymore, but you're targeting the agent.

And I think we spent the past 15 years talking about developer experience and optimizing everything for developer experience, but we're now going through this reality where we're transitioning from developer experience to agent experience.

And that's because we have a new consumer of everything that we do.

So I think that's kind of what we've outlined in our thesis about how we look for the next decade.

I think you can kind of summarize into two things, is that that's going to be agent-first infrastructure as we're pivoting all aspects of our developer tools and our infrastructure chain to be optimized for an agent-first.

And then there's the go-to-market side of that.

um and i think that's kind of implied assumptions of what it's going to mean to have an mcp server have some skills um but i think that the other side that we'll be recovering is like what does an agent first world mean for distribution what does it mean for procurement and because we're already starting seeing the early innings of that of like agents selecting tools making decisions for you etc but but if i were to just expand on the product pieces i think um Today, when we're sitting and writing code, we're using Codex or Cloud or OpenCode, whatever flavor of the Wego you're using, but that's really producing the code.

So that means all the abstractions that we provide have to work really well in that kind of agenda context.

But it also means that how we think about end-to-end full-stack software is changing.

And I think the chronical example is probably, let's say, Salesforce.

Salesforce went out and announced at the last conference that they're going headless.

What does that mean? It basically means that the future of Salesforce looks more like an MCP server and API.

And the kind of UI that made their business, they're kind of like fading that away because they know that people don't want to sit and click in a crappy UI, but they want the agent to do things.

And I can even see that in how I'm using software.

If I kind of could close to use it, I'll just turn and use something else.

Or I'll ask Claude to build me a version of it.

So I think even the consumer behavior is changing really, really fast.

So that's kind of one thing.

But I think if the UI kind of goes away, what does that mean for typical software? I think another piece we outline is our thesis around engines and kernels in the sense that If we reduce the UI, you actually reduce a lot of the surface area you provide to be more of a kernel or rendering engine.