Let's start in November 2024. Many of us in the AI world were losing our minds over Claude Code back then. The potential was massive. But for non-developers, the learning curve was too steep.

Then Anthropic came out with Cowork. The evolution of Claude Code, but for everyone. Not much excitement at first. The calm before the storm.

This week it happened. On January 30, Anthropic published 11 starter plugins on GitHub for Cowork. The catalyst that set everything in motion: the legal plugin.

Thomson Reuters: 6 billion dollars in revenue per year. The business model is built on expensive subscriptions for lawyers. Access to databases, research tools, everything the industry needs.

Then came the GitHub plugin. And the stock? Down 20%. In a single day. Because this one plugin makes a large portion of their business obsolete. The cost? From very expensive to practically free.

It's not just about legal tech. What people are realizing now: vibe coding is no longer a joke. Not AI slop. It creates real, functional software. For free.

I can ask Claude Code to build software for me. In 5 minutes. Tailored to my needs. No price tag. If you follow my channel, you know: we've been talking about this for over a year.

Fun fact: a few days ago was the one-year anniversary of "vibe coding." Andrej Karpathy coined the term exactly one year ago. The fact that the crash is happening right now, right after? Pure irony.

Tech stock markets are reacting violently. Especially companies that rely on software. SaaS businesses, software producers, investors in software companies. Some are already calling it the "SaaS Apocalypse."

The fear: companies that used to be worth billions are vanishing. Poof. Gone. Of course, the markets are exaggerating right now. If you do a lot of vibe coding, you know: there are still tons of issues.

The big problem? Technical debt. The more I vibe-code, the less I understand the structure. How were features implemented? What security vulnerabilities exist? The code often lacks security, scalability, and documentation.

But: for startups, this is a game-changer. More shots on goal. Building an MVP faster. Deploying quickly. Cheaply, without needing much technical talent. The cost of version 1.0? From hundreds of thousands down to practically zero.

If you just want to test whether something works, technical debt doesn't matter. You want momentum, finding product-market fit. If it takes off, you can build everything properly later. But until then: iterate quickly, test, see what works.

And it's not just for startups. Anyone with Claude Code or other tools can build impressive software. You barely need any technical knowledge. A little bit is enough.

For big SaaS companies, this means: the difficult problems remain. But many make a lot of money with very simple software. And that simple software? Any kid can build it themselves now. This won't kill those companies immediately. But long term? The high valuations are based on growth projections. And those are becoming more realistic now.

Today there was another leak: Anthropic is launching new models. Sonnet 5, better, faster, cheaper than Opus 4.5. And probably Opus 4.6 as well. Everyone expected the release today.

We're already seeing references to these models in Anthropic's backend and API calls. But something happened. Either a last-minute decision, not ready, or the leak confused infrastructure updates with the release date. They often build in features before they go live. For testing.

One thing is clear, though: Sonnet 5 and Opus 4.6 are coming soon. And that will be exciting for anyone using AI agents and software building. Better performance, lower costs, higher speed.

How will SaaS companies survive this? I don't expect a total collapse. That’s an overreaction. But long term? Valuations will drop. When more people can create their own software, tailored to their needs, how could that not have an impact?

I have already replaced multiple SaaS subscriptions. With custom software built using Claude Code, Clawbot, OpenCodex. I mainly use Anthropic models for coding. And I'm not alone. Many of you are experimenting with this. We are the early adopters. The geeks.

But when more people realize what these tools can do, more will build their own software. Why pay for software when you can just talk to a chatbot and it builds what you need?

For enterprise use cases with high security requirements, off-the-shelf software will still exist. But for millions of use cases where simple software is enough? Everything is changing.

By the way: I checked my website analytics today. ChatGPT is now one of the top referrers to my site. More than Bing or YouTube or many other search engines. ChatGPT is sending massive traffic. Exciting times.