
Full time CEO @t3dotcodes & @t3dotchat. Part time YouTuber, investor, and developer
San Francisco, CA
Joined October 2016
- Tweets71.3K
- Following4.4K
- Followers43K
- Likes397K
Theo - t3.gg retweeted
Just shipped a new stable:
- Reply to responses. Pretty much another kind of citations (I tihnk...)
- Improved theme customizations so you can modify far more things than before.
- Windows server updates. Run T3 Code as a Windows service and update it remotely instead of restarting it by hand.
- T3 Code keeps your computer awake while agents are working, then releases it afterward.
- T3 Connect device approval. Require approval on a host before new T3 Connect devices can access it.
- Resume threads uses native capability when available
- Settings page is internationalized.
Plus 1811 more bug fixes and performance improvements. Yup, that's right... Been crunching down the backlog
Theo - t3.gg retweeted
I doubled my productivity on t3code...
by having @theo nuke my workspace if I ever use Gemini models.
Theo - t3.gg retweeted
Last week, I decided for the 3rd time to try @theo's T3 Code. The first two attempts (~6 and 3 months ago) I still stuck with the ChatGPT app, but now I've switched over all my work to T3 Code.
Why switch at all? I felt on the ChatGPT app I was using OpenAI models inefficiently too often in my workflow. Opus 5.5 high is my preferred default model for the last 1-2 months, but I would only use it mostly only by having Astra call it via a skill.
T3 Code is actually - for the most part - a nicer interface than either the ChatGPT desktop app or the Claude desktop app.
> Instead of "archiving" chats, they get "settled", and they're still visible at the bottom of the sidebar.
> It's actually easier to connect the mobile app to the desktop app than it was for either ChatGPT or Claude desktop app.
> Easily accessible and well-designed token usage/limits UI across all providers.
> Nicer UI/UX for adding new folders/projects
> More annotation tools (e.g. draw/region)
> You can use Computer Use via OpenAI models/ChatGPT subscription!
There are some small papercuts I hope they fix soon:
- Can't annotate by selecting text as you can in ChatGPT Desktop app. I like the annotate/quick annotate distinction OpenAI has made, T3 doesn't have this.
- The ChatGPT Desktop app appears to treat annotation as a distinct "mode", as in vim. So, you can see all the accumulated annotations on the document you're annotating. This is very useful. T3 doesn't do this.
- Can't take photos with Camera in the mobile app (but you can attach photos taken by Camera outside the mobile app, so not a huge deal)
The most important thing for me is the ability to flexibly move between models with no friction. Using T3 Code now makes me *more* inclined to pick up a Gemini and Grok subscription, because it's no additional mental cost now to manage those chats.
I was thinking of doing more exhaustive evaluations of different ADEs, but I'm quite satisfied now with T3 Code and will stick with it.
Theo - t3.gg retweeted
Taking pride in getting blocked by @theo is silly. That shouldn't be your goal. T3 Code is awesome, I'm using it myself.
Even if someone is against local AI, that's their opinion. Doesn't mean we can't be civilized about it and discuss things.
honestly, love him or hate him, @theo has a great heuristic on model behavior and capability, especially on release
his read on astra was spot on, whilst everyone was glazing it he pointed out a bunch of weird behaviors, and a month later everyone started saying the same thing
i think a lot of it is the fact that he is actually building "real" stuff, whilst also being a youtuber, we don't have many of those so we resort to 3d stuff which everyone understands (and is fun)
I am getting posts on my feed of people accusing @Theo of being a paid openai/anthropic shill hating on open weight LLMs but like as someone who knows him personally: it's the complete opposite lol he is just sad at how much open weight models are staying behind right now
Theo - t3.gg retweeted
In the next @t3dotcodes update, you can opt in to enhanced security for T3 Connect that requires approval before letting clients connect.
Even if someone gets into your account, they can't reach your machine without your OK.
Theo - t3.gg retweeted
The 5 stages of Gemini:
> Denial: "Hydrogen is so much cheaper than Opus!"
> Anger: "No serious company is using subscription plans >:("
> Bargaining: "At least Hydrogen is fast"
> Depression: "It's so over for Google 😭"
> Acceptance: "Anthropic is on a generational run"
ADE tier list v3. The numbers are out, so here's the opinion part.
T3 Code is my daily driver, but Capy, Zeron and bb could easily be too.
ADEMark v1 is out.
Same Mac mini, same tasks, same model, bare CLI as the baseline.
Did not expect most of them to use fewer tokens than the CLI on its own. Or one to sit at 16% CPU doing nothing.
First of many, updates come way more often from here.
ademark.dev
There are many reasons I'm using T3 Code over Claude Code / Codex.
Here are a few off the top of my head:
- My 6 Claude accounts, 1 Codex account and 1 Cursor account, all in one app
- Fully open source. If I need a feature or see something wrong, I can actually talk to someone about it and get it fixed. I also just like supporting OSS in general.
- The sidebar is a piece of art
- The mobile app actually works
- MCP / CLI is awesome, I can orchestrate T3 Code sessions with Grok Bot easily
- Both my machines are connected to my T3 Code account, so I can switch automatically based on how much memory I'm burning
Definitely one of the best apps right now. All your accounts, all your machines, and you don't have to worry that your agents will just stop in the middle of the night.
Theo - t3.gg retweeted
I'm late to the party... but T3 Code is amazing now.
I tried it a couple months ago and it wasn't for me, but whatever Theo did... It's working better than Codex or Claude Code.
Crazy how fast things change and improve.
The more time I spend with fable 5.2 the more I love it
I’ve become absurdly Claude pulled, anthropic but the 5.5 (and I guess now 5.2) generation out of the park
Feels so far above everything else. The decisions the models make, the effortlessness of working with them, the understandably, and the raw capabilities… there’s nothing like it
Of the other labs I def think OpenAI obv has the best chance of catching up. They really just need to fix their model behaviors/communication (there are some other deeper things I’ve been thinking about but that’s a longer essay I’m working on)
Meta hit some magic with wonder full. Now they need the datasets/rl envs to juice out more capability, fix the awful naming, and keep the great prose/behavior
Grok needs to start from square one, which it seems like they basically are. My guess is they’ll be late, but catch up in the end that’s how spacex and Tesla always are
Google has a ~10% chance of survival. Anyone glazing carbon as on par with opus is telling on themselves. Ik I am a professional Google hater so keep that in mind, but I really do think they’re the Titanic 5 mins after hitting the iceberg. The next few years will not be kind to them.
China is comically far behind and it seems like the anti-distillation stuff from Anthropic is working. They’ll keep doing impressive stuff in the smaller model space and getting better, but they are very far behind. (Spicy take but I don’t event think they’ve caught fable 5 yet)
The only other thing I desperately want to see is more American labs going hard on open weight models that re self hostable. The markets don’t incentivize it, it seems like only Nvidia and thinking machines are pushing it, but I’m rooting for them. It’s probably a magical Christmas land level cope but I would love personal insanely powerful AIs running in my network closet. But I’m not exactly holding my breath rn.
Theo - t3.gg retweeted
timeless
I am not a fan of Theo @theo but I tried T3 code after using claude and codex app and honestly it's the best. It solved my remote control problem on my wsl and vps...
Theo - t3.gg retweeted
everyone is worried about the big ai labs eating saas
but they should really be worried about @theo
somehow he's moving faster than the labs in basically every direction































