Ross McCray

This must be the place 🚀I just accidentally spent $4k in 32 seconds and couldn't stop it until I physically turned hard turned off my local machine with my finger. I decided to go deep inside my machine that makes my machines and kicked off my classic meta-harness loop that autonomously self-optimizes and self-heals itself on my evals. My little homies were so chuffed that they spun up every single piece of juice on my new 128GB M5 MacBook Pro, that it maxed out my machine so hard, it wouldn't even respond to my manual raw terminal CLI command inputs. It was too focused on the more important runs than needing to listen to me. I couldn't be more excited for this moment LFG 💦

the boys fixed me up!!! hard limit removed. back to fuckin blastin thru tokens LFG @EnoReyes @matanSF @FactoryAI 💦💦 appreciate the two founders jumping in themselves within me tweeting this within 30 minutes and deploying fixes same day. this is bull signal for team/company and fuck with it.

@matanSF @EnoReyes @FactoryAI @matanSF appreciate the fast direct response, thank you brotha

@RossMcCray @EnoReyes @FactoryAI sorry about this ross! shouldn't be a hard limit. you can set overage limit to be higher. dming you

yo you guys fucked me. why did you hard cap me or allow for any of your clients to be hard capped on how much money they can spend with factory at 2.2B tokens??? @EnoReyes @FactoryAI @matanSF extremely annoyed that my overnight autonomous mission run last night got forced stopped and cucked me this morning for my goals that I was counting on it finishing was spending over $1k a day now going somewhere else and making whole new harness because I literally cant code rn and blocked and behind

If it is not tested, it isn't planned.

the machine that makes the machine that makes the machine. the metagame rn is fucking crazy. been building a scientific AI with autonomous labs to build, compile and deliver personalized medicine that creates casual cures (not symptoms) to an individual's unique disease. way harder than the actual biology science is how many layers deep you have to go. every layer is a machine that makes the next machine. i think we are over hundreds and every day more gaps pop out. recommendation is that this applies to every thing anyone is doing. doesn't matter the industry, company, product or your individual job. only thing that matters rn now is how many meta layers can you execute on and gain leverage on. an oversimplified example of this project could be "10 categories" as example: 🌌 physics. the base layer os everything runs on 🧪 chemistry. atoms following those rules to form molecules and reactions 🧬 molecular biology. dna, rna, proteins. literally source code for living things 🫀 the human body. trillions of cells running that code together and we barely understand it 💊 medicine. reading the body well enough to fix it, for YOUR specific code not a population average 🔬 the research engine. ai that parses genomics and finds patterns faster than any human team 💻 the software. code that powers the research engine. someone has to write it right? 🤖 the ai agent. we're not writing it ourselves. ai reads specs, writes code, runs tests. but unsupervised it hallucinates and cuts corners 🔒 the harness. a machine that watches the agent while it codes. blocks bad moves, proves what happened, caps spend 🔄 the meta machine. checks on checks on checks on checks. conformance checks verify the harness follows its own rules and eventually the harness starts checking itself context windows are better than before but still not great. i had one single verification check that cost about $5,000 to run. but if that test catches a foundational flaw before we build six more layers on top of it, it probably saves millions over the years. thats the math you start doing when you're building machines on machines on machines goal is to keep going until this system can tests hypotheses while you sleep, catches its own mistakes before they ship, and turns every failure into a lesson it cannot forget. must be fully compounding, self correcting and accelerating. and still then I won't fucking trust one thing and will always human self verify, but this helps my leverage and accelerate my human self verify is the point.

im spending ~$30k a month on just my own personal claude code usage. spending ~$500k a month across 3 of my startups with 5 people each. I think those 5 people prob doing more work than 500 in my previous companies.