Pinned Tweet
Here’s @Max_Alexxander from SkyroDigital after we rebuilt their agency operations.
“It used to be a mess.”
“Tasks slipped through the cracks.”
“Now it runs smoother with 20 clients than it did with 8.”
“I don’t even worry about signing new clients anymore.”
[Full VSL]
paste this into claude code:
Install t3-sharks for me: clone github.com/hunterbohm/t3-sha… to a temp folder, read install.sh so you know what it does, run it, then tell me to quit and reopen T3 Code.
mac only. t3 itself isn't touched, so updates keep working. sharks by @stackedlol
I think about this a lot
It’s only a matter of time before things change, as they always have
- Principles for Dealing with The Changing World Order @RayDalio
My @bot is amazing and always
*just works*
But I feel a pull deep in my soul to use Hermes over grokbot for almost most things.
I try to use grokbot as a tool it is and not build upon it bc I can’t shake that feeling if you don’t own the code then it not worth building your system on
And Hermes with opus 5.5 is literally magic
Learning about gradient descents and trying to build the cursor “Projects” abstraction into @t3dotcodes on my flight
I fully expect to never have this merged and may not even PR bc I know @theo isn’t taking feature requests and that’s totally cool
But we’re at point with agents that if you’re not building custom tools for yourself to make you more productive… ngmi at least imo
We’re all unique no tools fits everyone perfect
Hey @grok what does this say about simulation theory
~ $840 in ai costs a month
- 3x claude max
- 1x 20x codex
- opencode go
- grok super
Will change next month lol
Got it
Sick dashboard design @theo
i spy Hermes with my little eye
My Hermes setup has been absolutely banging recently.
I’ve been able to get a lot of real work done more easily from my phone than ever before.
That mainly comes from understanding better the nature of personal assistant agent harnesses and Claude code codes etc
Granted I know Hermes is much more powerful than just a personal assistant. But using it almost as an executive assistant instead I’ve found to be a great combo
What does that mean?
I do alot of my work in t3 code using Claude code or codex as the harness. But with orchestrator v2 feature which allows agents to spawn and manage threads inside t3 code.
So anywhere I am I can get updates on my work or know of something needs me and it’s truly been s tier
Some of my setup:
I’m using my default Hermes profile on opus 5.5 with many different specialist profiles setup like content, code, research.
I have the default use native Hermes -p “ “ profile. To have the default delegate work or use the t3 code mcp to spin up threads in whatever machine and repo and harness I want while I’m out.
Which I have visibility into from my phone. S tier expierence
And also just using default Hermes memory. I find that works the best and I’ve tried mem0 honco, Viking, hindsight, gbrain
But I find that they normally clutter the agent with context that is rarely relevant and if the agents need something they can search for it.
Some of that does come down to agentic workspace repos for knowledge work where I already have things documented. But native Hermes session search is magical.
And doing a TON of regiment with how I interact with Hermes and optimizing the soul Md edit after edit to reply shorter and more coherent
I wanted to do this because of how much I like grokbot. But not wanting to build myself into someone else’s walled garden
Next will be owning the tokens that get generated. Probably either a m5 ultra or some dgx shakes but not sure for now
@theo @maria_rcks pls copy projects
Actually huge
Things included in this PR:
- Pi support.
- Auto-resume when limits reset
- T3 Code MCP (create, launch, message, wait on, read, search and interrupt threads)
- `delegate_task` lets an agent start child agents on any provider or model
- ACP Registry: add any registry agent (Devin, Cline, Kimi, Droid, and more)
- Cursor runs on the official Cursor SDK, not the (broken) CLI
- OpenCode 2 support
- Agents can rename threads, regenerate titles, link PRs and settle threads
- New MCP tools for worktree handoff, and for listing and closing previews
- Proper server-side queueing and steering
- Thread forking
- Switch provider or model in the middle of a thread
- Native subagents show as child threads in a lineage view, with their model, status and history
- Use `@` with a thread name, or drag a thread from the sidebar, to attach it as context
- Scheduled tasks
- Tons more I'm forgetting
Pumped this landed. We'll have a lot of cleanup to do to get it perfect, so expect lots of updates throughout the next few days :)






















