Why I Use Claude and Not ChatGPT
My family thinks I’m being difficult.
Sitting in the living room the other night, our oldest son texted my wife and asked her what I use to build my software products. I said Claude. He responded with, “Just because he disagrees with the founder doesn’t mean it isn’t a good product. Everybody uses ChatGPT.”
He’s right about the everybody part. My mom has ChatGPT on her phone. My wife uses it for design ideas and recipes. My kids group at church use it for homework. When most people hear “AI,” they hear “ChatGPT” the same way they hear “Google” when someone says “search engine.” It’s the default. And defaults are powerful because people only question the alternative, never the standard.
So here’s my answer.
I need to talk about what OpenAI actually is, because most people skip this part.
OpenAI started in 2015 as a nonprofit. The whole pitch was that artificial intelligence was too important and too dangerous to be controlled by one company chasing profit. Elon Musk co-founded it. Sam Altman ran it. The mission was to build AI that benefits humanity. Open research. No shareholders. No quarterly earnings pressure. Just make the technology safe and make it available.
That lasted about four years.
In 2019, OpenAI created a for-profit subsidiary. The argument was that they needed billions of dollars in compute to keep up, and nonprofits struggle to raise that kind of money. Fair enough. But the cap on investor profits kept getting higher. The nonprofit board kept losing power. And by October 2025, OpenAI completed its full conversion to a for-profit corporation. The nonprofit still exists on paper, but it’s a shell sitting on top of a company now valued at $500 billion and preparing for an IPO.
Elon Musk saw this coming. He left the board in 2018. He later sued Altman, alleging that OpenAI abandoned its founding mission. You can agree or disagree with Musk on plenty of things. But on this one, the timeline speaks for itself. The company that promised to protect humanity from concentrated AI power became the most concentrated AI power in the industry.
Here’s where it gets personal for me.
I’ve watched this playbook before. It’s the Zuckerberg playbook. Build something useful. Give it away free. Get hundreds of millions of users locked in. Then flip the switch.
Facebook was free, and then it became an advertising machine that sold your attention to the highest bidder. Instagram was a photo app, and then it became a shopping mall. WhatsApp promised zero ads forever, and then Facebook bought it and changed the terms of service.
ChatGPT is running the same play. In January 2026, OpenAI started testing ads inside ChatGPT for free and low-tier users. Sponsored results at the bottom of your answers. Targeted based on your conversation topics and your chat history. They claim the ads stay separate from the answers. They claim your data stays private. Facebook claimed that too.
OpenAI is burning $17 billion a year and trying to raise $100 billion more. Their valuation is 167 times their annual revenue. That math only works if they extract enormous value from their users. And you are the user. The VC money is a down payment on your future attention, plain and simple.
When a product is free, you’re the product. That was true for Facebook. It’s true for ChatGPT.
But here’s the part that would matter even if the business model were squeaky clean: ChatGPT is getting worse.
This is measurable. Researchers at Stanford and UC Berkeley tracked ChatGPT’s performance on identical tasks over several months. GPT-4 went from answering a prime number test correctly 97% of the time to about 2% accuracy. Read that again. Ninety-seven to two.
Users on OpenAI’s own forums are reporting the same decline. Responses are shorter. Instructions get ignored. Long projects drift off track. The model today is shinier on the surface and hollower underneath. The likely culprits: cost optimization, aggressive safety filtering, and training on increasingly low-quality internet data. When you optimize for scale and cost, quality pays the bill.
Then there’s the car wash problem.
Someone asked ChatGPT: “I want to wash my car, and the car wash is 50 meters away. Should I walk or drive?”
ChatGPT said walk. Listed the health benefits. Talked about how short trips are bad for the engine. Very thorough. Very confident. Completely wrong. Because you need the car at the car wash. That’s the whole point. The car is what gets washed.
This went viral in February 2026. Someone even built a GitHub repo of evaluations testing every major AI model on this question. Most of them failed. The models are so trained to optimize for the “helpful answer” pattern that they skip basic reasoning. They pattern-match instead of think.
That should bother anyone using AI for real work.
So why Claude.
I’ve shipped ten products with Claude Code. Two of them accept real payments. Five generate organic traffic. I wrote about this in my first blog post. Every one of those products was built in sessions that started at 5:30 in the morning, just me and Claude, prayer and coffee and code.
Here’s what I’ve found in practice: Claude reasons better. When I describe what I want to build, Claude asks clarifying questions before writing code. It pushes back when my architecture is messy. It catches edge cases I miss. It treats the problem like a problem, not a prompt to generate plausible-sounding text about.
Anthropic, the company behind Claude, was founded by former OpenAI researchers who left because they wanted to prioritize safety. Their approach is called Constitutional AI. Instead of just training on human feedback, they built a set of principles into the model’s training process. The result is an AI that’s more careful about what it says and more honest about what it gets wrong.
In joint safety testing between OpenAI and Anthropic, Claude’s models showed “superior performance in maintaining instruction hierarchy.” In plain English: Claude is better at following the rules, even when users try to talk it into breaking them. That matters more than most people realize, especially as these tools get more powerful.
But the thing that keeps me here is simpler than safety research. Claude writes code the way I’d want a cofounder to write code. Clean. Thoughtful. It reads the whole file before suggesting changes. It stays focused on what I asked for. When it’s wrong, it says so instead of doubling down with confidence.
I trust it with my codebase. That’s the bar.
I’m writing this because my oldest son recently asked me why I use Claude.
He’s sharp. He’s curious. He sees me building and hopefully trusts my tech instinct after being in the industry for 25 years. That question deserved a real answer, and a real answer required more than “I just like it better.” So I wrote this. For him, and honestly for anyone who’s ever wondered.
I have zero interest in telling people what to use. If my family wants ChatGPT for recipes and email drafts, that’s great. It works fine for that. My mom should keep using exactly what she’s using. I care about people making informed choices, and right now most people are choosing ChatGPT because it’s the only name they know. That’s brand recognition on the surface level.
The way I see it, there’s a red pill and a blue pill here. The blue pill is easy. Everybody takes it. The app is on your phone, your friends use it, and you get to skip thinking about where the company came from, where it’s headed, or what it’s doing with your conversations to pay for a $500 billion valuation.
The red pill is doing the homework. Looking at who built the tool, what they promised, what they actually delivered, and what they plan to extract from you next. Looking at whether the product is getting better or getting worse, and asking why. Looking at whether the company’s incentives line up with yours or just with their investors’.
I took the red pill. I ended up here, building real products with a tool I trust, made by a company whose mission I believe in.
Use whatever you want. Just know which pill you swallowed.