SaaS is Losing Its Head

TL;DR: AI agents are replacing web UIs as the primary way we interact with SaaS. The services survive, but the “head” (the dashboard, the web app) becomes vestigial. Companies that make themselves accessible to agents will thrive. The ones clinging to their web interfaces will get outmaneuvered.

One of the best ways to build the future is to live in it. This is possible because, as Gibson famously said, “the future is already here, it’s just not very evenly distributed.” I’ve been living in a particular corner of the future for a few months now, and I think I have a useful intuition for what’s coming for SaaS. I think there will be a massive transformation in how we use SaaS. Rather than web-based “headful” SaaS, we will evolve to SaaS that seamlessly integrates into your AI workflow: “headless” SaaS.

What I Mean by “Headless”

The future of SaaS is headless. What I mean is: you won’t go to PostHog’s website to look at analytics. You won’t go to Cloudflare’s dashboard to configure DNS. You won’t go to HubSpot to manage your CRM. Instead, your AI agent will do the heavy lifting, interacting with these services through skills, CLIs, APIs, or MCP servers. The web UI, the “head” of SaaS, becomes vestigial.

This isn’t a prediction. It’s how I work right now.

Why Now?

APIs have existed for decades. So why is this time different?

Because AI agents are now good enough to be the intermediary. A year ago, telling an AI to “configure my DNS” would have been an exercise in frustration. Today, an agent can read documentation, call the right API endpoints, handle errors, and verify the result.

The key enabler is skills: packaged instructions that explain to the AI how to use a service’s tools in a way that is both executable and context-preserving. A skill tells the agent what endpoints exist, what they do, and how to combine them to accomplish real tasks. It’s the missing layer that turns a raw API into something an agent can actually use on your behalf. Skills are becoming an open standard that works across AI platforms, not just one vendor’s ecosystem.

The people doing this today aren’t just early adopters building custom tooling. Hundreds of thousands of developers are already using Claude Code, Cursor, and similar agent environments where skills and MCP servers are a natural part of the workflow.

A Letter from the Present Future

Here’s how I build apps today. I use CodevOS’ Tower UI, which you can think of as Claude Code with a framework around it for organizing work. It adds features like multi-model review, specs as first-class citizens, structured development workflows, and an Architect agent that coordinates the work of other Builder agents.

When I need to configure Cloudflare DNS, set up a Railway deployment, or create a new GitHub repository, I avoid the web UIs if I possibly can. I tell my agent to handle it, preferably through a skill, or failing that through the CLI or API. If I have to touch a web interface, I curse under my breath.

Why? Because the agent already has all the context. It knows the project, the names, the configuration details. If I open a web UI, I’m in for a session of cut-and-pasting context from my development environment to the browser. And I’m guaranteed to get at least one detail wrong.

What’s more, I tell the AI to capture what we’re doing to a bringup.md file that documents the sequence of steps if I ever need to bring up the service again. The process documents itself.

Some providers make this incredibly smooth. A standout example is Resend, which makes its email capabilities available as a Claude Code skill. Here’s what that looks like in practice. I recently sent a newsletter to 2,500 people without visiting their website once. The AI helped me draft the message, run test sends, select the audience, and fire the broadcast. Why would I need a web page for that?

I know what you’re thinking: this guy built his own AI development framework. Of course he works this way. But the same pattern is playing out across the developer ecosystem. Anyone using Claude Code, Cursor, or Windsurf with skills or MCP servers is already doing a version of this. These tools already have hundreds of thousands of users, and every one of them is a potential headless SaaS customer.

The Question That Changed My Product

I have an application in development that helps you build a business using AI. It does pre-market research, creates and instruments landing pages, helps you build a mailing list, sends newsletters, helps you create content, and tracks what content actually resonates. It’s not ready for release yet, but if you’ve seen recent things I’ve built like the codevos.ai landing page, this tool has been behind the scenes.

Early on, I was thinking: if I want to offer this as a SaaS, I need a web interface, right? It needs to look vaguely like a HubSpot-for-bootstrappers, right?

Then I caught myself. If I actually had to go to a web UI to use my own product, I’d hate it. What if the entire day-to-day experience of using this product didn’t involve a website at all?

It’s not completely website-free. There’s still a web presence for things like account setup, and for any tooling that end users interact with directly (a newsletter signup page, for example). But the point is that the web presence is minimized. The operator, the person running the business, interacts through their agent.

“But what about dashboards?” you might ask. “Don’t you need to visualize what’s happening?” Dashboards exist to help you extract insights from data, and all those charts and graphs are designed to make patterns visible. But you know who is also good at extracting insights and has a lot of context? Your agent. You can just ask it.

And for most practical SaaS applications, you already have some kind of admin interface. Multisage, for example, has a backoffice for cost monitoring, key management, and other operational tasks. When I needed an analytics dashboard, I just told the AI to build one directly into the Multisage admin UI, following the structured development workflow I mentioned above (put together a spec, turn it into a phased plan, run tests and do reviews after every phase). It took a few hours, unsupervised. Why would I go to PostHog’s website instead?

So that’s where my new product is going. Its primary interface will not be a website. It will be accessed through Claude Code, Claude for Cowork, or other agent environments. You’ll have a project, and it will delegate the product-market fit and go-to-market tasks through skills and APIs. And if you need a dashboard, it can inject one directly into your existing application’s backoffice.

What This Means for SaaS

Does it mean SaaS will die? Does it mean the recent contraction in SaaS valuations was the real thing? I really don’t think so. I think in the long term, headless SaaS will lead to an expansion of the market.

Why? Because for companies that manage the transition, it lowers the barrier to entry for their customers. One of the biggest reasons people are reluctant to adopt SaaS tools is the steep learning curve. I gave up on HubSpot via the web. It was so complicated that I abandoned it and decided it was easier to build my own mini-CRM.

HubSpot does have an MCP server (it launched in public beta recently). But when I tried it, it was clearly built for people already deep in the HubSpot ecosystem. It didn’t help me get set up. What I needed was an AI that could bridge the gap between “I don’t know how HubSpot works” and “my CRM is configured and doing useful things.” That bridge doesn’t exist yet, but it could.

If it did, maybe I wouldn’t have given up. The AI would have built the bridge.

To be clear, headless SaaS won’t replace every use case. There will always be contexts where a visual interface is the right tool: collaborative design work, discovery-oriented exploration where you don’t yet know what you’re looking for, or environments where non-technical stakeholders need to interact with the system directly. But for the operational backbone of SaaS (configuration, data entry, reporting, monitoring, notifications), agents are already better.

The Dinosaurs Will Fall

But a headless future does mean a massive reshuffle. Some SaaS companies are at serious risk. The ones most likely to struggle share two characteristics:

1. They fail to make the transition to agent-driven interaction. No skill, no MCP server, no meaningful API. If you can’t be accessed by an AI agent, you’ll be invisible to a growing segment of power users, and eventually to everyone else.

2. Their entire business model is a thin wrapper around what is essentially an API. If all you provide is a GUI on top of a commodity service, and the GUI is no longer needed, what’s left?

Case Study: Mailchimp vs. Resend

A concrete example from my own experience. Mailchimp has no official MCP server and no CLI. The company’s entire product philosophy is oriented around its web interface. And for a growing number of users, that interface is the problem, not the product. Even Intuit’s own CFO has acknowledged that small businesses find the platform “a bit harder to use.” They charge me $85 a month for the privilege.

Frustrated with Mailchimp, I asked my Architect agent one day: could we use Resend for newsletters too? After investigating the Resend skill, it let me know there’s a Broadcast feature that’s perfect for my use case. A few hours later, the first newsletter went out, essentially co-authored by me and the Architect, since it already had most of the context and knew what I’d been working on. No web UI involved. Half the price.

Predictions are dangerous, but I’m fairly confident about this one: if things stay the way they are, Mailchimp will decline as users migrate off it, many of them to Resend. Resend is already a rocket ship, roughly quadrupling its user base in 2025 and hitting 1 million users by December. They’ve embraced the agent-driven world, and it’s paying off.

Losing the Head, Not the Body

SaaS isn’t dying. The body is healthy: the services, the APIs, the data infrastructure, the reliability, the scale. All of that remains valuable. What’s dying is the head: the web UI as the primary interface. The future of SaaS is headless. The companies that figure that out will define the next era. The ones that don’t will wonder what happened.