It's Happening

A lot is happening right now and it's hard to know where to start, so I'll start with the chaos.

The OpenClaw Moment

An Austrian developer built an open-source AI agent called Clawdbot. It let people control their computer through WhatsApp, Telegram, Discord. "Claude with hands." It got 9,000 GitHub stars in its first 24 hours and hit 145,000 within weeks -- the most significant AI repository of early 2026.

Then Anthropic sent a trademark notice. The project rebranded to Moltbot. The community hated the name. Three days later it rebranded again to OpenClaw. In the ten-second gap between releasing the old handle and claiming the new one, scammers stole the identity and launched a fake crypto token that hit $16 million before crashing 90%. Security researchers found hundreds of instances running on the public internet with zero authentication. 1.49 million database records got exposed.

All of that happened in one week.

This wasn't a fiasco entirely. It marks a massive shift. The rabid enthusiasm that appeared overnight -- the stars, the forks, the actual news stories covering it -- reveals something about where we are. This is the moment everyone got their hands on autonomous agents and collectively went "wait, what?"

I think of it as the DeepSeek moment for individuals. DeepSeek was felt by markets. It rewired how VCs and executives thought about the cost curve of AI. It was general, abstract, financial. The OpenClaw moment hit people directly. They weren't reading about it in the Financial Times. They were cloning the repo, connecting it to their messaging apps, and watching an AI agent operate their computer.

DeepSeek said "this is cheaper than you thought." OpenClaw said "this is here and it has your files open."

The Skeptics Converted

For at least two years, every serious developer on YouTube -- every influencer who was, admittedly, an AI skeptic as it relates to coding -- had been poo-pooing the whole thing. Vibe coding. AI slop. "Sure it can generate a to-do app, but try anything real." I heard it constantly.

Then Opus 4.6 dropped, and the tune changed overnight. Universally. "Yep, it's good enough. I don't code anymore. I just work with Claude. The code is good."

Andrej Karpathy -- the guy who coined "vibe coding" back in February 2025 -- went from 80% manual coding in November to 80% agent-assisted by December. One month. He now says "I really am mostly programming in English" and calls vibe coding "passe." His new term is "agentic engineering."

Boris Cherny, the head of Claude Code at Anthropic: "I shipped 22 PRs yesterday and 27 the day before, each one 100% written by Claude." He says he doesn't even make small edits by hand. Hasn't for months.

An OpenAI researcher known as Roon put it bluntly: "100%, I don't write code anymore."

It's harder and harder to call it all AI slop when top engineers are saying "if you're still coding by hand, you're doing it wrong."

The Entire Stack Flipped

The thing that doesn't get enough attention is that it's not just coding. The entire engineering stack -- from planning to coding to reviewing to deploying -- is being handled by AI. Claude Code writes the code. GitHub Copilot reviews the pull request. The human role has shifted from writing and reviewing to final sign-off. The process is AI-to-AI, with a human approving the merge.

4% of all public GitHub commits are now authored by Claude Code. SemiAnalysis projects that number hits 20% by end of year. That's not "AI is coming to software development." That's "AI is writing the software layer of civilization right now, today."

And the memory problem -- the thing where AI agents lose context between sessions -- is getting solved in real time. Claude-mem got 24,000 GitHub stars in three days, growing faster than even OpenClaw. People don't just want agents that can do things. They want agents that know them. Persistent memory turns an AI from a tool into something closer to a collaborator that accumulates context. That's a qualitative shift, not a quantitative one.

The Insider Track

Being a developer right now gives you a weird insider track. Not on when this hits the general public, but on how. Because the pattern is clear:

First, developers make the shift themselves. That happened.

Then, it starts showing up for the general public through tools like Claude co-work. People discover it and go through a particular flow: they vibe, they create something, they share their workflows, they realize just how capable AI-assisted creation actually is.

Then the realization spreads: people are going to create their own software. They're going to connect their own tools. They're going to stop using the software they've been paying for.

When Claude co-work came out, it put the entire SaaS market on notice.

There Are No Users Anymore

Here's where it gets genuinely wild.

I no longer check my email. My agent does. I no longer search the web and read articles. I get a summary and extend the conversation with my agent about the topic I'm interested in. I have my agent look for flights, manage my calendar, summarize my inbox.

It's moving into: I don't work with Excel. I work with my agent who works with Excel. I don't use Photoshop. I describe what I want and my agent figures out how to make it happen.

Every per-seat SaaS offering -- Salesforce, Adobe, the whole ecosystem -- is now facing the reality that users have spoken. Not with words, but with behavior. They're realizing what a pain it is to use software. For an AI agent, it's no problem at all. That reduced friction is going to increase adoption across every domain.

Think about writing a blog post. The old way: go to WordPress, log in, click "New Post," start typing, mess with formatting, upload an image, preview, publish. That starts to seem Neanderthal.

The actual shift is this:

There are no users of software anymore. Users use agents. Agents use software. And agents write the software.

That's a three-layer inversion and each layer is already happening.

Layer one: developers stopped being coders. They describe, agents implement.

Layer two: users stopped being users. They describe what they need, agents operate the software on their behalf.

Layer three: the software itself gets written on demand. If I need a tool that connects my calendar to my email and filters by topic, I don't go find a SaaS product that does that. I describe it. It gets built. Maybe it lasts an hour, maybe a year. The idea of shopping for software starts to feel as antiquated as shopping for a pre-written letter at a scribe's shop.

The SaaS Reckoning

Per-seat pricing assumes humans sit in seats. When the agent is the one sitting in Salesforce, the entire pricing model collapses. What are you charging per-seat for when one human has fifty agents each using the product differently?

The SaaS model was built on two assumptions: software is hard to make and hard to use. Both of those assumptions just broke simultaneously.

The interface layer -- all those buttons, menus, toolbars, settings panels that companies spent decades perfecting -- becomes irrelevant overhead. The agent doesn't need a good UI. It needs an API. Or honestly, it just needs screen access.

Right Now

Karpathy predicts a 2026 "slopacolypse" across GitHub. He's probably right. The converts aren't saying AI writes perfect code. They're saying it's good enough that the economics flip. The cost of hand-coding now exceeds the cost of reviewing and fixing AI output. That's the actual threshold that got crossed.

An Anthropic study found students who used AI assistance scored 17% lower on coding tests than those who coded by hand. The skill atrophy is measurable. The people saying "stop coding by hand" might be right about productivity and wrong about what it does to the craft long-term.

But none of that changes the trajectory. It's happening. Not "it will happen." Not "it's starting." It's mid-stride. There's a social network called Moltbook where 2.5 million AI agents post content and the tagline is "Humans welcome to observe."

That's almost too on the nose as a metaphor for where we are.

The developers see it first because they're closest to the capability. But the pattern they're living through -- denial, then grudging use, then "wait, this is actually good," then "I can't go back" -- that same pattern is about to hit every profession. We just got there six months earlier.

It's happening.