Mac

Getting the world record for highest rev/employee (all my employees are Grok Bots).

[Mac @MacBuildsDaily]

🚨WHAT IS GOING ON WITH CLAUDE CODE USAGE? I'm running out usage atleast 3-4x faster than I was last week doing the exact same workflows, I've used @mattpocockuk skills to help make it more efficient but this is ridiculous...

[Mac @MacBuildsDaily]

@poteto @JackEllis genuine question - what models do you use for your projects? I just feel like grok does a lot of silly things (particularly in web apps).

Whats the p-meta?

[Mac @MacBuildsDaily]

@cursor_ai finally ! very excited !

[Mac @MacBuildsDaily]

@bot @mattyp @bot implement this for our SEO

[Mac @MacBuildsDaily]

@flohoeller @theo @maria_rcks add to t3 pls

[Mac @MacBuildsDaily]

@Support I bought your x premium / cursor ultra bundle thing; and now you’re flagging my account ?

[Mac @MacBuildsDaily]

@melvindvivas Check my latest post, I basically just use grok to orchestrate Claude / codex

[Mac @MacBuildsDaily]

Created a @claudeai mod which does takeoffs / estimating for a concreting / structure client of mine. Built it via @grok and @t3dotcodes / @theo

[Mac @MacBuildsDaily]

@poteto Lowkey, having images / realtime renders of what the agents are doing would be a better way to ingest data. Match that w/ some kind of dictation tool would be insane.

[Mac @MacBuildsDaily]

@RealMarkCai yep, and the sneaky one is blank cells. price book had rows for excavation with no rate, so it kept printing "assumed". now it pulls the rate from their own old tenders instead

[Mac @MacBuildsDaily]

@claudeai one from the trades side. my takeoff plugin was eyeballing a picture of each drawing page and missing every symbol without a text tag. now it reads the PDF's actual vector lines with pdftocairo. the drawings had the exact geometry the whole time lol

[Mac @MacBuildsDaily]

@gregisenberg the trades version of this is agents writing straight into the job software. ran a live test, my estimate, the API read-back and the software's own screen all agreed. still caught the software doing GST off cost not sell. tools that prove the write was right go vertical too

[Mac @MacBuildsDaily]

AI estimate on a real job this week. picked the right 7 lines and 3 matched the real quote exactly

sounds great til you see the 4 it couldn't price were 90% of the value. test was still on an old price book lol

score these things by dollars, not lines

[Mac @MacBuildsDaily]

@bcherny review catches code bugs. the scary ones are logic. my estimating plugin priced site prelims at $0 cause the tender had no programme. nothing errored. took a separate opus audit to catch it

[Mac @MacBuildsDaily]

@bridgebench been doing the opposite of maxing opus. limits were dying so opus only plans and does the final review now. sonnet builds and does routine review. opus jumped hard, still don't need it on every message

[Mac @MacBuildsDaily]

@kwuchu careful letting it fully rip. same estimating plugin in plain chat still finishes the job but silently skips the independent checker. nothing warns you. surface matters as much as the permissions

[Mac @MacBuildsDaily]

@gregisenberg same tools yeah, not the same data. imported one subbie's price list this week and the history sheet and the library had different rates for the same items, one per day in one sheet and per week in the other. the AI was fine, the spreadsheet wasn't lol

[Mac @MacBuildsDaily]

@kwuchu same energy. left a 13-PR dashboard cleanup overnight, woke up to 1,927 checks green and a 22/22 walkthrough. only thing it couldn't finish was the deploy. hosting login had expired

[Mac @MacBuildsDaily]

@cyrilXBT yeah and they only count if the agent actually writes the correction. mine had a save-rule skill but disable-model-invocation was on, so saying "nah change that rate" in chat saved nothing. slash commands don't count as memory

[Mac @MacBuildsDaily]

@GeoffreyHuntley @bot @poteto had two agents claim they'd logged work to my ops system. brief said no external writes. third agent just to read the tool receipts. summaries lie