The project today
I am the fork’s sole user. I focus on product behavior and high-level or technical specifications, while coding agents implement much of the experimental code.
T3 Code
Testing agent workflows in my own T3 Code fork
A personal fork lets me specify, prototype, and dogfood workflow ideas quickly enough to learn which tools earn a place in my engineering work.
T3 CodeScreenshot coming soon
What I’m solving
Agent work becomes difficult to coordinate, monitor, and continue across providers, machines, and devices. Starting an idea on a MacBook should not require keeping the laptop open or reconstructing context before work can continue on a persistent Linux machine.
My responsibility
I identify workflow friction and product possibilities, write specifications, direct agent-assisted prototypes, and evaluate them through daily use before keeping, revising, or removing them.
01
Key decisions
Cross-provider orchestration
Extend orchestration across provider boundaries so one agent can initiate work in another provider’s thread, such as Codex creating a Cursor Composer thread.
Visible parallel work
Keep parallel work understandable through a monitor, pinning, tagging, and usage visibility.
One-action handoff
Develop a single handoff action that preserves task and workspace context while moving active work to a remote Linux environment for background execution and later phone control.
Where it stands
Orchestration, monitoring, pinning, tagging, and usage visibility have become retained parts of my workflow.
Cross-machine handoff remains active product work rather than a finished outcome.
What I learned
The evidence is personal utility, not external adoption. The project is most credible when I show what survived dogfooding and remain explicit about implementation authorship and unfinished work.