context, quickly: I've been building a thing where a technician takes a photo, says what he found, and a standardized report writes itself before he leaves the site. this week I connected it to ElevenLabs' agent API. so now it actually holds a conversation – it asks, you answer, it asks the next thing, you can talk over it. no buttons, no tapping. you'd think the voice was the hard part. it wasn't. it took an evening. the hard part was the opposite of talking. an agent is built to be helpful, and helpful means it will happily fill in "alvorlighedsgrad: kritisk" because the sentence sounded serious. nobody said it. it inferred it. and that is the one thing a report must never do – because a human signs that document, and a plausible guess is worse than an empty box. so now it can do exactly two things: ask about a field that's still empty, and read a part number back to check it heard right. it cannot propose a value. ever. and everything it claims still goes through my own check afterwards, where anything you didn't actually say gets emptied out again. that took three times longer than making it speak. it's not perfect - but it works! ⚙️ - AI voice is fun to work with.. what's the document in your week that you fill in afterwards, from memory? (potential solution is right here...) @Amy Zinserling – could you imagine this approach for your case? your experience, multiplied 💛

I've been building the next classroom course this week, and it kept circling back to one question. turns out it's the hardest one to answer and the most valuable. not "what do you do." what must never happen in your work? here's why that matters more than it sounds. an agent can look up almost anything about your field. the standards, the formulas, the vocabulary, the best practice – all of it is a search away. what it cannot look up is the rule you learned by watching something go badly wrong. so it will break that rule. cheerfully, confidently, probably in the first week. and you're the one who signs off on the output. the fix is almost stupidly simple: write the rule down once, somewhere it reads every session. most of these rules are under ten words. the one I keep coming back to – never present an estimate as a quotation – is seven, and it prevents an entire category of expensive mistake. no amount of general intelligence gets you that. you only get it from someone who has watched it go wrong. so, one line in the comments 👇 what must never happen in your work? mine: never a number without its source. unverified figures don't go into research, and a sourced one takes ten seconds to check instead of an afternoon. and a few of you I'd genuinely like to hear from, because I have a guess: @Uzair Nazeer – LN app creation?. what's the reading you'd never act on without checking twice? @Amy Zinserling – what would you never say in a session, no matter where a client takes it? @Tim Adams – what would you never put in writing to a prospect? @Monosriz Dutta – twenty years of transformation work. what's the shortcut that always costs more than it saves? @Joao Carvalho – you publish weekly now. what would you never let go out under your name? wrong guesses welcome – correcting me counts as an answer 😄

ten days ago @Joao Carvalho lho dropped one line in here – the part of his blog process he doesn't enjoy doing. a day later he shipped it: bloghelper-phi.vercel.app – a publishing checklist, a pattern read-out of his own posts, a formatting desk. his voice untouched, the boring parts gone. go look, it's genuinely good. and I realized – that's how everything good in this room has started so far. Katerina asked one question about brand images. Aaron asked about a video twin. one honest line, then a build. so I want to try something this week. call it the board 🧱 drop ONE line below: the task in your week you'd hand off first. that's the whole ask. "the friday report." "chasing suppliers for spec sheets." "the same email for the tenth time." if it annoys you, it qualifies. my side of the deal: I'll reply to every single one with the exact agent brief I'd write for it – in the thread, free, for anyone to steal. a few guesses I'm curious about 👇 @Uzair Nazeer – the factory-data report nobody reads, or alarm triage? @Tim Adams – the notes and follow-ups after every call? @Amy Zinserling g – session notes → the follow-up message? @Monosriz Dutta – the discovery doc that's 80% the same every time? @Katerina Bakunina – your consistent-image question IS this. put it on the board and we'll build it properly. wrong guesses welcome – correcting me counts as an answer 😄 and if you're reading this thinking "mine's too small to mention" – that's exactly the one I want. small is where every good agent starts. if this catches on, the board opens every monday. your experience, multiplied 💛

that's @Joris Peels so before I show this to anyone else in here, I want his eyes on it – and yours. the thing: drop one photo. ~40 seconds later you get an STL you can send to a slicer – with a printability score next to it, so you know BEFORE you print whether it's actually a part or just a pretty render. 👉 romanmartins.com/forkable-factory/photo-to-print the score isn't "does it look nice." it's the stuff that decides whether it prints: is the mesh watertight, are the walls thicker than a nozzle, how bad are the overhangs, does it still look like the photo. every AI-mesh benchmark I found grades on textures. none grade on whether the slicer accepts the file. so I built one that does. there's also a compare mode – the same photo through three open-source image-to-3D models (TRELLIS, Hunyuan3D-2, TripoSR) side by side. they disagree a lot more than you'd expect. that's half the fun. I've loaded some API credit into it, so go break it with something real from your desk. a bracket, a knob, a broken clip you've been meaning to remake. it's on me. @Joris Peels – my guess is you'll say one of two things: "the walls are still fantasy for FDM" or "the geometry is fine, the tolerances are the problem." which one – or is it a third thing I haven't thought of? tear it apart, that's exactly what I want. everyone else: ONE line below – what did you drop in, and what score did it get? screenshot welcome. your experience, multiplied 💛