Theo drives 8 T3 Code threads across machines from one MCP server
The feature is behind a labs toggle on the nightly build, and Theo says it is still very early.
Theo says multi-machine orchestration is working in T3 Code. In a post he labelled Day 28, he credited Julius for the work and described the result as one thread spanning models, harnesses, apps, operating systems, machines and networks. It is behind the labs section in settings on the latest nightly, and he adds that it is still very early.
One thread, multi-model, multi-harness, multi-app, multi-os, multi-machine, multi-network.
A day later, in his Day 29 post, he went further. He says you can now control your whole fleet through a single MCP server in T3 Code, and posted a clip of himself spinning up 8 threads on different machines via Grok Bot. MCP, the Model Context Protocol, is the interface agents use to reach tools, so the fleet shows up to the agent as one more tool endpoint rather than as a pile of separate sessions.
He then clarified what the demo is not. The control path runs over the internet, not via computer use, the approach where a model drives a remote desktop by looking at pixels and moving a cursor. His example is that he could turn off his laptop and still control his other machines from his phone through Grok Bot.
If I turned off my laptop, I can still control my other machines via my Grok Bot on my phone.
What this is and is not
This is the author demoing his own product two days running, not an independent test of it. There are no numbers on reliability, no failure cases, and nobody outside the project has posted a run. Theo's own framing, still very early and hidden behind a labs flag, is the honest read.
For working engineers, the interesting part is the shape rather than the demo. Most agent setups today assume one machine, one working copy and one session you babysit. Treating a set of machines as addressable targets behind a single protocol endpoint moves the unit of work from a terminal tab to a fleet, which is where several harnesses have been drifting as people run more parallel sessions than one laptop can hold. Whether that holds up once threads on different operating systems start stepping on each other is not something the sources answer.
Have the good opinion, not the loud one
One email a day. Under three minutes. What working engineers found out about coding with AI.