Dots are maybe the most opaque in terms of how they work. Why is everything hidden?

I am NOT giving Zuck my data, but I will take his tokens. Same goes for instinct. Like count down to their first huge data breach. So I’ve rigged up a system where they can email my trusted personal agent and ask for info or request actions like putting stuff on my calendar

So instinct has its own email built in and Muse very much does not

Errbody be building ai meal planners. Announcing my ai meal planner! chef.dmfenton.net

Universities are going to be reinvented around token allocation

Coded and Claude Code are NOT systems thinkers. Spending a lot of time cleaning up things that flow from that. They naturally want to pile fixes on fixes when the proper thing to do is change the design

Yall the games created by kids using AI are going to be 🔥

It seems everyone is building their own family meal planning and task agents… Announcing my new meal planning and task agents!!

Dots are cool but I worry they will abstract away too much engineering. E.g. I often notice that my coding agent is working inefficiently and therefore needs some new tool or skill. I have system instructions and retrospective skills to look out for this but it doesn’t work reliably. Solution may be another loop…

OpenAI dots are virtual employees, full stop

Riding to an artificial intelligence conference in a robot car summoned by my phone. The year is 2026

I find myself increasing asking models “what’s in the design space for this problem?”

Token allocation is capital allocation

This is good news for me because I burned thru my pro weekly credits in a day. Kinda miss the 5 hour limits tbh

And on the seventh day…he ran out of tokens

Instinct is 0/3 for me in doing anything useful. iMessage interface is cool but who cares if it doesn’t work?

To be honest, I’m kind of annoyed that I’m still useful to the models as an architect, engineer, and product manager. I want it to do it all for me!

If Astra is so smart, why did I have to be the one to tell it 12 minutes in CI is unacceptable?

I can’t remember a single unit test written by ai failing. So what’s the point again?