rich

Building AI/dev experiments in public.
Currently: Locwars + an AI decision-mapping tool. Writing the weekly “Could I..?” series.

[rich @RichDevLab]

I’m starting a weekly “Could I..?” series.

One question. One experiment. Public notes, code, mistakes and results.

Mostly AI, dev tools, games, automation and data.

[rich @RichDevLab]

how have I only just seen this cool little Codex addition?

lovely bit of UI interaction

[rich @RichDevLab]

@Tech_girlll because now you're debugging the prompt and the code path at the same time

this is why vibe coding doesn't replace traditional knowledge of patterns, proper implementation etc.

[rich @RichDevLab]

great to see this finally confirmed

set phasers to fast mode, reset inc

[Tibo @thsottiaux]

Ola. The Codex team is investigating issues where some accounts are seeing faster usage draining than intended. We believe this is related to some abuse & fraud prevention mechanisms that we have that might be overflagging.

Stay tuned.

[rich @RichDevLab]

@kylegawley i've seen this a few times where the first version was early, not wrong. infra gets cheaper, habits change, then the same idea suddenly works

[rich @RichDevLab]

@tibo_maker this is why plain text launch posts fall flat, they tell me what shipped but not why i should care. auto-picking the most impactful thing is the smart bit

[rich @RichDevLab]

@jturntdev my issue isn't even the delay by itself, it's that 'ai for everyone' sounds a bit hollow if frontier access is basically invite only

[rich @RichDevLab]

@itsolelehmann i agree, but the tricky part is knowing what actually made it work

[rich @RichDevLab]

@sharyph_ this is clean

subagents being context-isolated is the bit most people miss, otherwise the main agent starts dragging old newsletter context into everything

[rich @RichDevLab]

generalists are set up insanely well right now

if you can think across:
- tech
- business
- people

and connect things fast, that's a goldmine

[rich @RichDevLab]

@Road_Kill11 i'd add ci too

it still assumes one clean human change moving through a line. with agents you get a pile of partial edits, and the hard part becomes proving what is actually safe to merge

[rich @RichDevLab]

@zats oh this is really cool

[rich @RichDevLab]

@zdogmode feels useful for structure, risky for voice

once every post starts using the same wording and cadence people can tell even if they can't name why

[rich @RichDevLab]

@kylezantos i'd take this just so people can say 'agents' and not mean 4 completely different things

once you know whether someone builds this stuff or just tried it once, the whole thread gets way less wobbly

[rich @RichDevLab]

@theslowtell i've followed tiny accounts when the profile immediately told me what i'd get more of. not polished, just obvious why i'd come back

[rich @RichDevLab]

@pixelcoder08 yep

launch week gives you borrowed attention, week two tells you if anyone actually comes back without the spike

[rich @RichDevLab]

@monindie yeah, and after enough of it you stop giving posts a chance at all

[rich @RichDevLab]

@RijnHartman i haven't tried it, but yours is the first negative take on GLM 5.2 i've seen

did you try any iteration with it?

[rich @RichDevLab]

@omarvvvr ai removed a lot of excuses, but it also made it way easier to ship something nobody actually needs

[rich @RichDevLab]

@tomhacks seeing a lot of this recently from big accounts

i think they prefer their own hot takes more than they dislike the thing they're talking about

[rich @RichDevLab]

@akoskm this is how tool switches actually happen, not benchmark charts, one random hour where the fallback is just.. better