J.

future multi gajillionaire. stay a while and listen.

[J. @munchivelo]

@steipete @kenwheeler @SimiStern any other suggestions besides twilio?

[J. @munchivelo]

The math is insane on its own.

But what’s crazier is how many VCs take 10%+ of a company with supposed 100B–trillion dollar potential, then basically just sit back and hope it works.

If VC's truly believe something can be that big, what real operational weight, network, or hands-on leverage do they actually bringing to accelerate it?

Capital is easy. Getting the right enterprise meetings or marketing links, and converting them from day one — that’s where most VCs fall short.

[J. @munchivelo]

when i run out of usage, i switch to codex spark.

which basically does everything just as well anyway.

lesson in there somewhere.

[J. @munchivelo]

@aarondfrancis 'the ai can give out it's own tasks, just one more loop refinement bro, i can perfect this, just give me more time'

[J. @munchivelo]

@pmitu nah - just a literal plant shop.

[J. @munchivelo]

@thsottiaux @YashHustle_22 Codex Mobile in bed just before a 3 hour sleep.

[J. @munchivelo]

@CharlesFerqlrm @ApplyWiseAi your atlas cloud spam bots need work.

i mean, cmon guys. exact same time?

rookies.

[J. @munchivelo]

track back to just over a year to now.

i'd built an automated ecommerce flow that took a whole store end to end.

seo would research trends, products, and map those into .js scripts which would launch prompts that read those research files. that would feed an image gen prompt which created designs, set to specific standard. i'd generate them, and then ANOTHER prompt, would check the images, score them with a criteria, and either move them to an accepted folder, or move them to an archive folder. the accepted folders, would automatically fire a script which would open photoshop, map the image to smart layers, in a 'product shot' template i'd made, and then export all of the final product shots to another folder, and then exported the flat designs which would be used for the products.

another script took the product images, did visual lookups, generated all product descriptions, renamed the images and generated the seo text. it ran optimizations locally via a jpegoptim and oxipng script. it then uploaded them to dropbox, and via API, would generate a dropbox link map.

i had one barebones csv template, which i'd run a ps1 script through to map json files into the csv rows, and insert the dropbox link map. all my images, links, followed the exact same slugs, so it turned 2 hours of manual work into a 5 second bulk rename and insert.

it then converted that csv into json, which then itself converted that json into ld-json for product rich listings.

ai would write the product description based on a dataseo keywords, and googletrends json file that would run on every product type. collecting keywords for that specific product. it also formed it around brand profiles, copy guides and other things. this was sonnet 3 days, GPT 4.0 days, and it STILL wrote great copy when it had the right guidance. in the .js file, i'd replace all em dashes with a hyphen if they ever appeared.

i built a custom product uploader, built my own php plugin which synced to local .js files and connected via rest. it was (and still is) one of the best wc product uploaders that exist, as it completely resets filterlookups only for that product, and is lightning fast because i upload it directly into woocommerce rows from json. no importers, no wordpress malarkey, or WC rest needed. it was 50x faster than wc's own CSV import. the images would be uploaded via ftp, and then on detection, would sync those to the media library, and i'd upload the image meta from the seo run, so they all had captions/alt text etc.

it took what would be 3-5 hours of manual work per product, and congested it into a 2 minute image to fully live product system.

after that, i'd export sales data, the ai was constantly learning, sales data feeding back to files, which would then teach the ai what products work, what doesn't. what copy worked, what copy didn't.

that would then flow back into the original source files which told the ai what images to gen and what products to launch.

all of it was local on my pc. i wasn't selling an saas. it was just something that worked for my very particular setup.

the thing about it is; i built that mostly with GPT 4.0 and a little bit of 3.5! mostly copy and pasting code manually from the chats in chatGPT. all the plugins, the php, everything. then some of it got improved inside vscode back on the old original copilot plans, when $10 used to last you an entire month of none stop coding.

this was before n8n, before agents were even a thing.

all of that I built very specifically for myself, local, syncing folder to folder, json file to json file. python scripts watching files, and .ps1 files that would follow up with other .ps1 files, which launched .js files which contained prompts for AI, and hitting the openAI API's whenever I needed the AI layer.

eventually i built a terminal tool, which would allow me to run the scripts from the terminal, and i'd manually type in the slugs for which products i wanted processed. all files would sit in specific folders, and scripts would do the rest. i was so excited about that, giving my terminal app a shortcut icon and putting it onto my taskbar.

that was a year ago.

fast forward to now.

the game has changed so much. ANYTHING and i mean anything is possible now. people 'new' to codex, and CC etc don't know how good they have it.

my advantage is that i have a year of scripts, a year of tools. i've laid the SYSTEMS in place, to fully map out entire features, precisely, and organized, and build out projects, in one hour, and have it implemented within the next. entire saas features - mousework.

but i've had this fucking idea for so long, to build a fully automated, self learning ecom business, that launches products end to end based on it's own research, writing, and growth, but the complexity of it previously , and being busy with life, it never got finalized. the secret is i sync it via etsy too, but they're API keys take FOREVER to aquire, but built my own etsy system, product uploader, which runs across 7 different stores.

however, now, i've finally been building the replacement for it. i'll be able to run that exact same system, except this time through a full app, with a canvas, and agent systems instead of .ps1 scripts. not to say i won't run scripts; they're an integral part of any automated workflow, but now it has superpowers, and it can do so so so much more. all the ideas I wanted to do, automated, fully, end to end.

not only that, but i moved away from woocommerce entirely. instead i just built my own website builder, which is also fully automated end to end. my brand profiles, my artwork system? i'm still using those, just for more things. now i can launch 50 brands just like it, running the same system, all in about 5 minutes.
whether it's saas, local service, or online ecom.

i also built an ai automated ad builder. it takes my brands images, or generates images. i've got background removers, and full skills and agents which fully generate the ads for me. it mixes all that into seedance videos, and posts in logos etc. now i take those image/videos, and build instagram, tiktok, facebook vids, generate descriptions, and upload them automatically. it has an every growing library to source from, templates to use, and the system derives right with the websites, so all themes/styles match precisely to the brand.

this is why it's so great building for yourself. the amount of reusability you get with it, the fact it's free forever, can never be beaten. none of these saas companies get it. and they're heading in the wrong direction. we could already DO half of what these companies are doing. my own personal SEO system, which i built for my automated web builder, is already 10x better than any yoast, rankmath etc. i skip expensive ahrefs, semrush, and just rebuild their services myself, using API, which is 100x cheaper. except this time it FEEDS my system, and i don't need to lay a finger on it.

nobody cares about these little one off apps that won't exist in a year. they're either failing to see the future, or they're hoping for an early exit before they know the dominos start falling. and they don't get it. their 'app' is just a little tiny module in something that thinks bigger.

people will want PRIVATE systems. all speaking to each other. not 1200 integrations and 1200 invoices to send to, that don't even have a fucking brain.

i'm not selling anything yet. but if you're interested in seeing how i think about automation, then stay a while and listen.

the tool i'm building will absolutely help you too. but i'll be honest. i'm actually quite scared to release it, solely down to how powerful it is.

not many people do it like i do, and i'm finally on here to tell the world.

if you're a cannabilistic, sick sadistic, son of bitch, 666, you're in pain but you sit and stick with it, in the midst of business, then drop a you know what.

[J. @munchivelo]

@benln i'll buy you a beer in tokyo.

[J. @munchivelo]

@tibo_maker whats the average warmed range?

[J. @munchivelo]

remember when we didn't have codex, and chatgpt had like 10 memory slots and couldn't even read or get context from other chats?

those were the days.

[GEOFF @geoffreywoo]

what should do with my X revenue share?

i think i will give away 200 bucks each to top 5 followers who RT, like, and comment something funny and/or useful 💸💸💸

[J. @munchivelo]

Interesting program.

I wonder what actually ends up counting as impressive or incredibly useful in practice. Are we meant to post here, or it's just an announcement?

Curious how much of the selection ends up being driven by visible signals like socials, documentation and conversation versus private usage data. Either way it might push people to be more deliberate about capturing the actual flows they're building, which could be useful on its own.

[J. @munchivelo]

@OpenAIDevs more importantly you fixed the scroll bar select when next to the github box thing.

[J. @munchivelo]

@dayonefoundry hilariously accurate

[J. @munchivelo]

at this point i don't think humans are even concious.

we think we might be. we think we know we exist. but given how we all basically think and act the same, how do we know that's not just training data? how do you 'know' what it's like to be you? your analysis of it matches everybody elses.

how do we even know what we exist in is even real?

being concious is different from being self aware, but the only real difference from machines is that we think we know that we can feel things.

[J. @munchivelo]

everybodies obsessed with agents.

nobodies obsessed with how you keep your brain aligned with the speed they work.

agents produce 100's of .md a second, and you're thinking in obsidian still. yea right, as if you're even reading half of them.

but guess what, neither is your AI.

then you end up with 100's of stale .md files that all say different things, because you forgot to build the system that keeps them aligned and were never given the tools to create the right context in the first place.

so congrats, now your architecture is fucked, and you've burned a billion tokens.

i'm obsessed with how you keep that accountable, streamlined, and tidy.

but more importantely, i'm obsessed with how the ai assists in preventing your brain from exploding, and how it helps align your focus through its swarm.