5 min read

Claude Just Dethroned ChatGPT. I Saw It Coming.

Last Tuesday, 1.5 million people showed up outside OpenAI's headquarters in San Francisco. Actual humans with actual signs. Accounts deleted on the spot. The whole damn thing played out live on every feed I follow.

Wednesday morning — ChatGPT uninstalls up 295%.

Thursday — Claude, #1 in the US App Store. Eleven million daily active users. Paid subscribers doubled since January.

I was sitting in my home office in Aurora, Ontario watching this unfold, and... look, I don't want to sound smug about it. But I'd been building my entire content automation system on Claude Code for months at that point. So no, none of this caught me off guard.

How the Pentagon Broke ChatGPT's Brand

If you missed it, the short version: OpenAI signed a weapons-adjacent deployment contract with the Pentagon. Not a research thing, not a safety consultation — an actual deployment deal. Details leaked. People lost it. A grassroots movement called QuitGPT went from a Reddit thread to 1.5 million people in nine days. Nine days.

Meanwhile, Anthropic — Claude's parent company — got blacklisted by the Pentagon for refusing to loosen their safety guardrails for military use. The DoD tagged them as a "supply chain risk."

Think about that for a second. One company said yes to the military and hemorrhaged users. The other said no and their servers literally couldn't handle the influx.

I spent 20 years in enterprise networking. I know what "supply chain risk" actually means in practice — it's the designation you slap on vendors you don't trust inside your infrastructure. The Pentagon meant it as punishment. The market treated it like a stamp of approval.

That's not just a PR moment. Something fundamentally shifted in how people pick their tools.

I Was Already There

The Pentagon drama isn't why I switched to Claude. I'd switched months earlier because — and this is going to sound boring compared to all the geopolitics — it was just better at helping me build things.

Quick context: I'm a network architect who pivoted into AI. Building three products right now — OpenClaw (business automation), Adara Protocol (blockchain for AI agent economies), and Pincer (cybersecurity). I needed an AI that could actually do engineering work, not just spit out Wikipedia summaries with a confident tone.

What I've shipped with Claude Code in the last 90 days:

All Claude Code. Single tool.

I bring this up because nobody in the press is covering this part. They write about download numbers and Pentagon drama all day long. What they're not asking is: does the tool actually deliver once you're past the hype and trying to build something real?

Yeah. It does.

GPT-5.4 Missed the Point

Five days after the protests, OpenAI dropped GPT-5.4. First model with native computer use — it can click through your desktop apps, fill out forms, navigate interfaces. Genuinely impressive piece of engineering.

But I keep thinking about this: OpenAI shipped a capability upgrade. Anthropic earned something you can't ship — trust. Those are fundamentally different games.

When you're building a business on an AI tool — when your revenue flows through it, when your content pipeline depends on it, when customer data passes through it — you care a lot less about benchmark scores and a lot more about whether you trust the company behind it.

I've architected networks for some of Canada's biggest enterprises. Twenty years of that. The first question we ever asked about a new vendor wasn't "what are the specs?" It was "do we trust this inside our perimeter?" AI is going the exact same direction.

If You're Picking Your Stack Right Now

If I were sitting across from you at a coffee shop, here's what I'd say:

Forget the benchmarks. GPT-5.4 wins some tests. Claude Opus 4.6 wins others. Doesn't matter. If you're using AI to automate workflows, create content, or build a product — what matters is whether the thing understands what you're trying to do and follows through. Does it feel like working with someone, or like wrestling a search engine?

Pay attention to the business model, not the model. OpenAI raised $110 billion, took the Pentagon deal, partnered with Disney. Anthropic got blacklisted because they wouldn't bend. One strategy optimizes for quarterly earnings. The other is a slower bet on trust. Pick which one you want underneath your business.

Just pick one and build. I see so many people stuck in this loop of "testing" tools for weeks without committing to anything. Pick Claude. Pick ChatGPT. Honestly, I don't care. Just pick one and build something real this weekend. You'll learn more in two days of shipping than six months of reading comparison threads on Reddit.

My whole business runs on Claude now — content, product dev, research, all of it. I made that call early, and watching 11 million people land on the same conclusion doesn't make me feel smart. Mostly it makes me feel relieved that the tool I depend on isn't going anywhere.

The Part That Actually Keeps Me Up

Okay, here's where it gets dark.

The U.S. military used Claude in the Iran strikes. After they blacklisted Anthropic. Pentagon designated them a supply chain risk, then turned around and deployed their AI for target selection — over 1,000 targets in 24 hours — through partner channels. Without Anthropic's direct involvement. Without their consent.

Sit with that for a minute.

It means something uncomfortable for anyone building on these platforms: your tool's values don't mean much if the infrastructure around it has different ones. Anthropic said no. The government used it anyway.

For builders like me — and maybe you — that's the conversation worth having. Not which chatbot gives nicer answers. Who actually controls the infrastructure? Does saying no mean anything when governments decide they want in?

I don't have a clean answer for that. But I'd rather build on a platform that tried to say no and got overridden than one that never even asked the question.

One Thing You Can Do Right Now

If you've been thinking about switching to Claude, or you're already using it but just for chat — try Claude Code. Not the chatbot. The CLI tool. It lets you build software, automate workflows, and wire up real systems using natural language. Completely changed how I work.

I'm going to document exactly how I built my 17-platform content system in upcoming posts. Stay tuned.

The model wars are over. This is about who you trust.

← All posts