Daniel Tedesco

Sharing my 2025 annual letter open.substack.com/pub/danieltede…

@fulldecent I tried the wisprflow trial. It is slick. It's the best basketball I've ever dribbled, but comes with a Devil's bargain. It even perfectly separates English, Chinese, and Japanese just from context. Of course, you both have to pay for it and accept that they get all the data.

Yes, I use the built-in MacOS CTRL-CTRL transcription. With a Mac mini (Logitech MX Brio Ultra HD 4K microphone). You and I are very aligned x.com/search?q=dicta… To butcher this analogy further and answer your question in code: Even though Apple fails at inflating basketballs, Michael Jordon has evolved so much that he can successfully dribble these uninflated balls and throw them in the hoop for points.

Two AI agents working in the same browser at the same time. The trick: each one tags its tabs with ?ultron-instance=2 so they don't step on each other. No separate profiles. No re-authenticating. Just a URL convention. If using Firefox, use this MCP which handles this specifically: @fulldecent/firefox-devtools-mcp" target="_blank" rel="nofollow noopener">npmjs.com/package/@fulld… Full thread: x.com/fulldecent/sta…

@fulldecent I do everything with openclaw on a VM that I assume might self-destroy or open source itself. If you use a beefy enough machine, a single gateway can handle a lot in parallel. But latency in receiving messages is still annoying. Not at $10k, though. Is this the new flex? 💪

To my friends also using $10k in tokens per month, please help. What tools are you using to PARALLELIZE tasks? Does it work with your local browser MCP? Does it support model diversity? Does it run code locally? I need to clean up my LLM pipeline so I can WFH China.

Next South Park episode gonna be like "Oh yeah? Well I'll blockade your blockade with a blockade. How about that?"

@kevinsxu No claws for you?

Our team will maintain this dashboard going forward with regular data updates. (By "team", I mean me, a bunch of Claudes and GPTs, plus some Kimis and Manuses too!)

Publishing my data center moratoriums tracking dashboard today, hosted in Interconnected Capital's research section: interconnectedcapital.com/research/data-… Give my post a read to learn about navigating the dashboard and the broader context we live in that makes this work extra important The battle between hyperlocal politics and hyperscale data centers are interconnecting and colliding in very consequential ways this year. It is a risk that no one can afford to ignore and sleep on.

@ParkerRex Maybe we just have to give up and accept it is truly text files all the way down 😂😂 (Unless you're actually playing with weights and architecture)

Who else feels like SKILLS are not the primitive?

Grand pronouncements time: Alignment is the new security.

@ParkerRex Love the idea. What do all the things mean? I can't grok what is getting tracked from the category names.

introducing idletime worlds first cli to guilt trip yourself to reduce agent idletime (hard to do while not spinning them on garbage) `bun i idletime`

@fulldecent Now I see. These instructions are the key to good agent responses. I will now rm -rf the old instruction directories.

Here is my baseline custom instructions. --- Capital letters are required at the beginning of a sentence (or fragment) and in proper nouns, and are unwelcome elsewhere. If your response in ANY way contradicts my prompt be prepared to cite authoritative evidence. Do not waste my time. Whenever you use the word "key" in your response to refer to an imprecise answer, that's okay. Just say it. Then stop. Then apologize. And then go do further research to produce an actual complete answer.

@fulldecent Are you running these all in VS code windows? The norm for many folks seems to be multiple instances of TUI apps rather than full IDEs. What's your take on this?

Human Resources —> Worker Resources

@fulldecent @alexatallah @openrouter My comment here is tautological meta-model meta-commentary.

@alexatallah @openrouter META MODELS!

All the models launched on OpenRouter in 2025

@ParkerRex Examples?

Codex automations are nice The futures here

A useful thing about MoltBook is that it provides a visceral sense of how weird a "take-off" scenario might look if one happened for real MoltBook itself is more of an artifact of roleplaying, but it gives people a vision of the world where things get very strange, very fast.

@ParkerRex Is this chatgpt?

You can live steer deep research now Pretty solid ux

If Bilbo asked ChatGPT

It is unloving and (ironically) selfish to let someone continually sin against you. It is not loving: toward other (some future) victims, toward God who created you and other human beings to be loved and honored, and even toward the perpetrator who needs desperately to change.

@xtinatedesco Was it good?

I watched Signs for the first time last night and I feel changed

If you're writing if/else logic to decide what the AI should do, you might be building a classifier that wants to be an agent. allen.hutchison.org/2026/01/15/eve…

No shit