Tyler Mitchell

Web Engineer

[Tyler Mitchell @tylerdmitchell]

@thsottiaux - Some kind of “Directives” feature that allow agents to gate decisioning so user doesn’t need to repeat the same things multiple times.
- Change worktrees for any thread without hassle
- Move any thread to any project

[Tyler Mitchell @tylerdmitchell]

@alightinastorm @rauchg @shadcn All shadcn did was popularize a headless component library pattern that existed years before shadcn was ever a thing. React-aria, radix-ui, and now base-ui; these are the actual drivers.

[Tyler Mitchell @tylerdmitchell]

@alightinastorm @rauchg @shadcn It would be like saying 5 year from now that much of Three.js’s success came from Vibe3D… when Three.js was already objectively successful and popularized on its own.

[Tyler Mitchell @tylerdmitchell]

@alightinastorm @rauchg @shadcn this is a rewrite of history…

[Tyler Mitchell @tylerdmitchell]

@rauchg @shadcn Nah, the headless component pattern was already established long before shadcn… shadcn just popularized it.

[Tyler Mitchell @tylerdmitchell]

@thsottiaux @thsottiaux Tibo acting like he didn't reset-bait for the last 3 days

[Tyler Mitchell @tylerdmitchell]

@evanqjones Nicee. What approach did it use here ? Are these tensor fields?

[Tyler Mitchell @tylerdmitchell]

@RyanSael Nice. How does the city's road network and vehicle traffic work? Wondering because it doesn't look 1:1

[Tyler Mitchell @tylerdmitchell]

@thsottiaux There's too much friction around Git Worktrees in the Codex Desktop App. I shouldn't have to create an entirely new thread just to use worktrees--especially if I have a large ongoing thread that I want to start a new worktree for.

[Tyler Mitchell @tylerdmitchell]

@thsottiaux Codex constantly contaminates codebase by codifying literalist interpretations of user messages instead of extracting the underlying intent and grounding to the conventional ontology relevant to the domain of what's being worked on

[CG @cgtwts]

Anthropic CEO:

“open source is kind of a distraction you still can’t really see what’s happening inside the model, so it’s not truly “free.”

People should be paying a lot more attention to what Anthropic is doing.

[Cody Bennett @CodyBennett]

@dangreenheck Sad to see and recognize work of myself and others contributed in exchange for nothing but attribution and education to then be convolved and stripped of attribution or knowledge of prior art. What remains is a shrinking circle of knowledge and disenfranchised contributors.

[Tyler Mitchell @tylerdmitchell]

@timelesstaco Yes

[Tyler Mitchell @tylerdmitchell]

@thsottiaux In every release after the May 8th build, the entire app will randomly reload--usually after clicking on a response from a running thread--and when the thread loads up again, large portions of the thread history are erased entirely. This happens on a completely fresh install.

[Tyler Mitchell @tylerdmitchell]

@Westoncb What changed specifically? I did notice that last update they hid the context window size indicator which is absurd, but are you talking about the update from today?

[Tyler Mitchell @tylerdmitchell]

@gdb @TheRealAdamG In every release Codex desktop app (MacOS) after May 8th—on fresh install—uncollapsing response item in threads will often result in the entire window going blank for a moment, and when it loads again the messages are missing in the thread that was running

[Tyler Mitchell @tylerdmitchell]

@thsottiaux I've been using it since it released on the pro plan. When GPT 5.5 first released it was amazing. Over the past 2 weeks my GPT 5.5 xhigh reasoning agents have become slow, lazy and don't follow simple instructions.

[Tyler Mitchell @tylerdmitchell]

@rezoundous Yes, and suddenly my agents have stopped being able to follow simple instructions.

[Tyler Mitchell @tylerdmitchell]

@thsottiaux I was at ~95% usage and got rug-pulled down to 0% mid-run and can't use Codex right now. Your communication is appreciated.

[cocktail peanut @cocktailpeanut]

The thing that interests me the most about this library is not what it launched, but when it launched.

For the first time, you don't need to understand a tool to experiment with it. You just ask AI. no need to waste time.

Here's my experiment.