going to present on AI Engineer SF next week the work of my whole last year, first time this work would be in public and first time in last 5 years i am going to present something from the scene. the work is presenting Amazon Privacy Preserving Intelligence, answer to apple private cloud ai.
A simplest server + skill + static web page for hosting.
Inside it is a sqlite per site + static hosting + SSE. You can oneshot almost any small app and make it interactive and coordinate between isntances in realtime!
It is very performant, very heavily tested by multiple models and API and devex is exactly what i needed. Vibecoded test with a chat and it works almost perfectly
1) coding assistants, coding harnesses (now could have been one of the top easily or at least have a name) 2) vector databases (it turns out most of them were indeed shit except turbopuffer) 3) compact nuclear reactors (guess what’s the next way in energy in us?) 4) home management - literally take 3d scan of a home enumerate everything autistically and vendor contractors and sell as subscription. (guess what was recently launched in bay area?)
There is an infinite list of apps i want reimplemented: 1) Plex - slow, overheats Apple TV, stutters, not smart enought to classify movies, super weird closed source software 2) Todo apps - most of them are not focused for thought where you sort stuff around and highlights important ect etc, still havent found good one 3) Better notion - simpler, but not lacking and high perfromant, including the native mobile apps not html abomination 4) Calendar app - all of them was acquired and killed nothing on the market at all 5) "company/home brain" - not some half-assed implementation that everyone have 6) Car Tracking apps, RaceCrono is good but it haven't changed in last 10 years 7) GoPro software is dog shit to sync 8.) Personal dropbox or something that doesnt make me depend on random cloud and wont be wiped using some worm 9) Home automation, including parsing video feeds and smart rules, can something detect that i woke up instantly? 10) Managed AI where all this is interconnected, not running claude in the loop and not insanity of openclaw 11) Simple document app, like quip (that would be turned off soon!) - simpler than google docs with comments etc etc, it is like medium but slightly better 12) Finance apps - how would you build and project and estimate your personal/family financials? there are only pre-ai tools like monarch which is barely useful 13) Self-organizing document app (mix of 3 and 8 and 11?) - upload doc and it would be stored somewhere and can't be deleted, extracted structurally and ready to use by me and agents. 14) Process automation - something happens - i want AI to react and prepare
Something else probably needed too.
Where the fuck vibecoders that would build all this that is plesant to use?
I did two very long experiments with GPT-5.5 xhigh fast, one failed completely, another one succeeded wondefully.
I had a task 2 years ago and it was very hard task - i needed to boot android emulators as EC2 instances or as some VM, but i needed it to be safe and contrained, not some forked QEMU with disabled security. There is still no solution for this - you have to build something yourself, i have asked GPT to boot android in Apple Hypervisor (write swift app to use this api!) and show me screenshot of it and give me a web ui to control it remotely. It implemented it perfectly, it took vanila AOSP build, created a 30+ patch files, build a dockerfile with all scripts to build an image and then booted it on mac. Then i asked to use vanila QEMU and it succeeded too. Then i asked it to benchmark GPU against host, it succeeded too. Then i asked it to make it secure, thats where my token limit ended, but i assume it would be still successful.
Second task is to build an AI runtime. I need a some unified way to run agents that WONT require me writing code, but i still can configure system nicely. That would work with API and with Subscriptions, that would have different sandboxing mechanisms, etc etc. This task failed. It failed because it did work on non-latest main (i forgot to pull) and it built some very very simplified version and completely wrong. I have asked then to pull the main and thats where everything become a huge mess.
So far i am impressed, didn do similar test for Opus tho, curious how it will perform.
i am still looking to find what’s wrong in my reasoning about building products/companies. i learned some in recent days:
1) i reject many ideas because i can build them in a day, but absolutely no one can or want to do the same, even simplest things 2) i feel that challenging ideas is not a way to make friends - people in sf are consistently agreeing to each other and not sharing anything direct with strangers, while this is usually a reason something collapsed spectacularly - it is not a basis for Californian people
