Hey Reader,

Most people think AI tools are neutral.

Put a prompt in, get an answer out. The tool is just a tool.

That's not how it works.

Every major AI company was built by someone with a specific worldview — about what intelligence should do, what it should refuse, what it should prioritize when it has to choose. Those beliefs don't live in a mission statement. They get encoded into the training data, the guardrails, the reward functions, the things the model will and won't say.

The output you get isn't just a function of your prompt. It's a function of the founder's philosophy running underneath it.

I've been spending time with all 8 of the major players. Here's how I read each one:

Dario Amodei / Anthropic

Dario left OpenAI over safety disagreements. That wasn't a career move — it was a conviction. Anthropic's entire research agenda is built around the idea that AI could go seriously wrong if we rush it. That belief shows up in Claude constantly: it slows down, hedges where it's uncertain, refuses things that feel close to a line. You can find that frustrating. Or you can read it as intellectual honesty baked into the architecture.

Sam Altman / OpenAI

Altman's worldview is that the benefits of powerful AI outweigh the risks of moving fast. That shapes everything — the release cadence, the partnerships, the willingness to ship before things are fully understood. ChatGPT reflects that: confident, broad, capable of doing almost anything, and also capable of saying wrong things with complete certainty. It's the product of someone who believes momentum is the point.

Elon Musk / xAI

Musk believes most AI companies have overcorrected on safety to the point of ideological capture. Grok is the result — a model built to say what other models won't. That's genuinely useful in some contexts. It's also a model with the politics and personality of its founder running through it. When you use Grok, you're using Musk's version of free speech as a design principle.

Jensen Huang / NVIDIA

Huang doesn't have a consumer AI product. What he has is the infrastructure every other model runs on. His worldview is that compute is the actual scarce resource in the AI race — that whoever controls the chips controls the outcome. That's not a philosophy about intelligence. It's a philosophy about power and supply chains. It's worth understanding, even if NVIDIA never produces a chatbot.

Yang Zhilin / Moonshot AI (Kimi)

Zhilin built Kimi around the idea that most AI models are fundamentally limited by how much context they can hold. His bet is that the real unlock in AI isn't faster reasoning — it's longer memory. Kimi's architecture reflects that: it can hold and reason across far more information than most Western models. That's a specific intellectual thesis about what intelligence actually requires.

Liang Wenfeng / DeepSeek

Wenfeng's philosophy is efficiency as a form of defiance. DeepSeek was built to show that frontier AI capability doesn't require frontier AI compute budgets — partly to prove a point to the US companies that assumed cost would keep Chinese AI behind. The model punches well above its compute class. The tradeoff is that it operates inside a regulatory environment that shapes what it will and won't engage with. That constraint isn't incidental. It's structural.

Mark Zuckerberg / Meta

Zuckerberg made a calculated decision to open-source Llama, and his reasoning was public: if Meta gives the model away, it becomes the standard, and competitors who try to sell closed models are fighting against free. That's not a philosophy about AI. That's a competitive strategy that happens to look like one. Llama is genuinely open and genuinely good. Just know that the openness is a business move, not an ideology.

Sundar Pichai / Google

Pichai is running the hardest balancing act in tech. Google's ad business is existentially threatened by AI that gives answers instead of links. His job is to develop AI fast enough to stay relevant while not deploying it in a way that kills the core revenue model. Gemini reflects that tension at every seam. It's a good model with one hand tied behind its back.

None of these tools are neutral.

They each carry the fingerprints of the person who decided what the model should and shouldn't be. Understanding that doesn't make any single tool better or worse. It just means you're working with something real instead of assuming the output is objective.

The model isn't the mind. But the person who built it left their thinking inside.

If you want to talk through what any of this means for how you're running your AI setup, you can grab a free call here: https://calendar.app.google/kcChWiDuZXxiCquH9​

— Louis

P.S. This week's email is brought to you by our sponsor below. Every click helps me keep these emails free. Thanks!