Kelsey Hightower

github.com/QwenLM/Qwen3...

Pull requests · QwenLM/Qwen3 github.com

I'll be speaking at the @wearedevelopers.bsky.social event in San Jose this week. I'm doubling down on my belief that people shouldn't be burning AI tokens when a more efficient option is available. www.wearedevelopers.com/world-congre...

ZTA: Zero Token Architecture · WeAreDevelopers World Congress North America wearedevelopers.com

This stage of the AI push reminds me of peak cryptocurrency, which was supposed to change everything. I was never afraid of blockchain the technology, I was afraid of the people behind it, and their vision for the future.

"It's unacceptable that Chinese open source AI models may have been built on IP theft". They are so close. Now do OpenAI and Anthropic next.

I hope we keep writing docs for people who want to understand and build things without using AI.

I'm not excited by what AI can do. I'm excited by what I can do.

People don't hire AI agents, nor do people hire mouse and keyboards. AI agents are software tools; not people.

I talk to a lot of startups building security tools for AI agents, and my advice to all of them is, treat AI agents like malware. Even if agents provide a lot of value to those who use them, agents are a security nightmare to those who have to secure them.

AI is a trojan horse. Accept the gift with caution.

Based on industry predictions, I'll be in the small percentage of people who won't have a personal AI agent working on their behalf.

If I wasn't retired, I'd be going deep on LLMs to avoid getting fired, and remain marketable. However, I wouldn't allow an LLM to do work for me without eventually learning how to do it too. I'd use an LLM as a tool, not a replacement. I'd also start contributing to an open weight model as a hedge.

I get the same reactions telling people I don't use AI that I get at celebrations when I tell people I don't drink.