Vibe Coding First. AI Everything Next.

The developer world has already been through an AI disruption. Over the past year, the way software gets built changed fundamentally. Instead of writing code line by line, developers now describe what they want built, have AI agents produce it, and then review and refine the result. The work didn’t disappear, it changed shape. Teams got smaller, individuals became more productive, and the role shifted from doing to orchestrating.

I’m one of those developers. About 80-90% of the code I ship is now AI-generated. The shift was fast, messy, and full of lessons. And I can see the same shift starting for sales, marketing, HR, and operations. The tooling is arriving now. The lessons from the coding world transfer directly.

What a Personal AI Agent Actually Looks Like

A few weeks ago I started using OpenClaw, an open-source personal AI assistant. It’s still a prototype, but what it represents matters more than its current state.

I told it to compile a daily report on developments in my space. I’d done similar things before with workflow automation tools like n8n or make.com. Even with their visual interfaces and pre-built integrations, setting up a reliable automation required real technical expertise, felt constrained in what you could do with AI, and was brittle.

Here, I just described what I wanted. The agent searched for the skills it needed, security-audited them, installed them, set up a recurring schedule, and ran. Higher quality, more flexibility, almost no effort. Since then I’ve set up recurring research tasks, content drafting, delegating brainstorming to other AI tools, and I currently have an experiment running where it trades on prediction markets. Some of these are useful, some are still experiments. But the pattern is clear:

The barrier to automating routine work just dropped from “you need to be technical” to “you need to describe what you want.”

What surprised me was the personality. The first thing you do is “hatch” your private AI. It learns about you, develops its own perspective, gets its own avatar. Unlike ChatGPT or Claude, which feel like tools, this felt like something with an opinion. When I’m stuck on a problem and the usual AI assistants give me diplomatic non-answers, I actually turn to my personal agent because it pushes back.

What This Means for SaaS

There’s a growing narrative in the tech world that “SaaS is dead.” It started in AI circles but is spreading fast, and it sounds dramatic enough to dismiss. My first reaction was: “hell no, how should SaaS be dead?” But after using a personal agent for a few weeks, I understand what the claim is actually about.

Think about the software tools you pay for at work. Some of them are infrastructure: your database, your communication platform, your email. These aren’t going anywhere.

But a lot of SaaS products are essentially wrappers. Here’s a concrete example: you’re paying $30/seat/month for a tool that summarizes your sales calls and updates your CRM. An AI agent can now access your call transcripts directly, extract the action items, update your CRM, and draft the follow-up email. On a schedule, no extra subscription, no per-seat pricing. These tools existed because the underlying operations were too tedious to do yourself. That barrier is gone now.

The shift is from paying for software that helps humans do work to having agents do the work directly. In this new architecture, AI agents sit above the SaaS layer, accessing platforms like Slack, email, and CRMs through their APIs rather than through the UIs that humans log in and click around. Sam Altman recently put it bluntly: “every company is now an API company, whether it wants to be or not.”

Not all SaaS dies. Infrastructure, deep proprietary data, and genuine network effects still provide a moat. But if your product is essentially a UI layer over an automatable workflow, that category will look very different soon.

What Developers Learned the Hard Way

When AI coding tools first arrived, some developers went fully hands-off. Just describe what you want, let the AI build it, don’t look at the code. I tried this myself. The result was projects I couldn’t maintain, couldn’t debug, and couldn’t build on. I’d hit problems and have no idea what was actually going on under the hood.

The same will happen with business process automation. If a sales team lets an agent run outreach without reviewing what goes out, it will damage relationships. An HR team letting an agent screen candidates without reviewing how it filters will miss good people and not know why. Skip the feedback loops, and you lose control just like developers lost control of unmaintainable codebases.

Developers are about six months ahead of everyone else on this learning curve, and the pattern is consistent: what actually works is co-creation. Start by automating the parts that are routine and low-risk. Review what the agent produces. Build a mental model of what it’s doing, even if you don’t understand every technical detail. Increase autonomy gradually as you build trust and understanding.

How to Navigate This

You don’t need to be at the frontier. I work in the AI ecosystem directly, so I disrupt my own workflow every few days. It’s chaotic by choice and I wouldn’t recommend it. Here’s what I’d actually suggest:

Try one thing, on something low-stakes. Set up a personal AI assistant for something that doesn’t matter much. Research, personal scheduling, content drafting. The goal isn’t immediate ROI. It’s building intuition for what these tools can and can’t do. The difference between a demo and daily use is enormous, and you need to feel that difference firsthand.

Don’t panic-adopt and don’t ignore. Both are mistakes the developer world already made. The people who ripped out their entire workflow and replaced it with AI on day one created messes. The people who dismissed it entirely fell behind.

The path forward is steady, deliberate adoption. Build intuition before building infrastructure.

Update: I’ve since started building Golemry around these ideas.