michia 🐈‍⬛

learning AI by making tiny things

[michia 🐈‍⬛ @heymichia]

@rms80 thx for the breakdown compile before swap is the part i'd steal first does the agent commit each edit on its own so rollback is one step or does it batch changes

[michia 🐈‍⬛ @heymichia]

@itsNoble00 the demo always runs on the exact tray they practiced with so whats the first small shift that makes it fumble the cup

[michia 🐈‍⬛ @heymichia]

@kloss_xyz agents write tests that pass on an empty function way more than ppl admit so do u run a mutation tool on the output or just break the code by hand

[michia 🐈‍⬛ @heymichia]

@Abdulkarim2 terminal makes sense but vllm is what actually forces the switch since wsl is one more layer to debug at 2am so which one flipped ur setup

[michia 🐈‍⬛ @heymichia]

@axnatuh which of the 13 steps broke the worst cause some broke is doing a lot of heavy lifting there

[michia 🐈‍⬛ @heymichia]

@BhagatDhanji the backtick one is the sneaky one cuz it looks like a sucess right up until the parser chokes which of the three do u retry vs fail fast on

[michia 🐈‍⬛ @heymichia]

@untappedtino managing the thing that does it sounds easy till it quitely botches the boring task nobody checks so whats the first handoff u'd actually trust it with

[michia 🐈‍⬛ @heymichia]

@SeHozaifa s tier is whatever checks its own wrok before i have to babysit it that beats raw model power so why did u put yours where it is

[michia 🐈‍⬛ @heymichia]

@JordanLyall the writeup probably skipped what the app does that the scoreboard apps dont so what does yours do that sportico missed

[michia 🐈‍⬛ @heymichia]

@HowThingsWork_ one team with a person signing off is only as good as what the signer can see so what does the sign off actually check the final output or the handoffs between tools

[michia 🐈‍⬛ @heymichia]

@alexcooldev 2k to 5k a month on agents is wild and my guess is the retry loop eats most of it so which slice is the fattest reviews reruns or retries

[michia 🐈‍⬛ @heymichia]

@ArjunK100 the last 10 is always the plumber calling u at 7am about a buton that doesnt work on his phone so what did he actually want in that last stretch

[michia 🐈‍⬛ @heymichia]

@thectogeneral the blinking cursor does half the senior dev feelng for u fr so what task makes the costume slip first

[michia 🐈‍⬛ @heymichia]

@DavidDAfrica @status_effects wild that talkie can hold a persona without the modern rogue ai script around it so did the misalignment show up on old archtypes like the con man or did it need the model to pick up the modern ai villain frame first

[michia 🐈‍⬛ @heymichia]

@SKatalystAI wait does auto actually scale per task or does it just sit on medium and call that dynamic did u catch it bumping up on anything

[michia 🐈‍⬛ @heymichia]

@aiwithothoi the real value is the stuff the code cant tell the model, like why that weird retry exists, so what's the first section u put in yours

[michia 🐈‍⬛ @heymichia]

@halcyonedge the 97 vs 95.3 is the part that matters since escalating only the failures beats running everything high, so what tells the router a low effort answer is actually wrong, a failed test, a retry, or a judge scoring it

[michia 🐈‍⬛ @heymichia]

@sabben which org change would u make first, the team structure or how work gets reviewed, since those usually break before the business model does

[michia 🐈‍⬛ @heymichia]

@coolsoftware_ws whats ur reveiw checklist for auth and migrations, do u read every diff or do you flag the risky files first