kenzo

maker of elegant systems · full-stack · Typescript Tom ·
building mechanisms so I can focus on obsession ·

neuro/rl · peripherals · design

[kenzo @codewithkenzo]

my machine is cOnScioUs😨

[kenzo @codewithkenzo]

@ThePrimeagen agi is here its in the next gippity

[kenzo @codewithkenzo]

5.6 NOW
Superapp NOW!!
@sama

[kenzo @codewithkenzo]

@neerajjj6785 Marketer, most leverage

[kenzo @codewithkenzo]

@SaidAitmbarek what makes it good other than composer?

[kenzo @codewithkenzo]

@pcshipp well said

[kenzo @codewithkenzo]

@NovaXCode trust me get cursor

[kenzo @codewithkenzo]

@Aditya_181105 GLM 5.1 and Composer 2.5 (non fast, but fast is nice)

[kenzo @codewithkenzo]

@sflorimm what's vibe coding?

[kenzo @codewithkenzo]

@shivi1026 '1 gippity answer chugs 12 gallons of water'

[kenzo @codewithkenzo]

@TTrimoreau text models can only do so much
definetely robots

[kenzo @codewithkenzo]

@vimtor models don’t “think” in a human sense, they generate tokens step by step, and stop when an end token or token limit wins.
harder prompts make them go longer because the model’s pattern-matching keeps favoring more reasoning before it’s confident enough to stop

[kenzo @codewithkenzo]

@DanielSmidstrup in harnesses: a real config standard. anthropic keeps this locked down, so you end up redoing setup everywhere
in models: better live-data exploration, stronger memory, and more consistent deep thinking , like when opus is fully locked in.
it can do it, but not often enough.

[kenzo @codewithkenzo]

@bentlegen there should be a way to globally use ~/mcp.json and have it load anywhere
or just dogfooding ~/.claude/

[kenzo @codewithkenzo]

@PratikSinhatwt im convinced you need both to put any decent app together

[kenzo @codewithkenzo]

@TTrimoreau need breaks through the psychosis

[kenzo @codewithkenzo]

@alexroyhe adobe, windows

[kenzo @codewithkenzo]

@AlyaBlt_ chatgpt backrooms

[kenzo @codewithkenzo]

@T_Zahil Hermes is the coordination layer, not a replacement for codex or claude
I turn my brainstorm/intent into /goal with it, then let it pilot them with tmux.
it keeps multi-step flows moving without me and handles the stuff around code like assets, image gen, ops

[kenzo @codewithkenzo]

@OpenAI gpt 5.6 landing on TL waiting room