Replying to Avatar Satosha

The conversation led to a new light bulb . I am sure most devs have already figured it out !

# New Dev /Test cycle with AI

## Specs

Describe well what you want - Let AI create the specs - Read the specs - ask for changes - let AI rewrite . Run the specs with experts. Ask questions. Narrow the scope of first version - Spend at least couple of days .

## Build / Deploy / Test

All these are rolled into one because build and deploy is pretty much ten minutes job. The key thing is TEST. This is where you want to spend time - Test well.

## Spec Rewrite

Every time you encounter a bug , don't run back to AI to improve your code . NO ! Instead append every bug in Spces . Basically testing and spec rewrite go head in hand . Each time , have your model rewrite the specs with a separate section of say POETNTIAL BUGS.

## LOOP

Basically run the cycle at least 7 times . Because it took GOD 7 days to build the world. It was not separate things being built daily.. It was prompt getting updated - 7 times :-)

## Backup

Only thing to back up in this cycle is the Final Specs .. That is where #nostr comes in - store your specs as a long text event !

The question is how do you learn !

This is where version management comes in ..

Don't waste your time , reading the first code . Rather have git give you a diff between successive version . Now that you have a baseline of code , and have a vantage of bugs , check how AI fixed those bugs ... You don't need to know the coding language , or architecture ... The knowledge is built into the change . Read the comments and add more comments ..

Git diff is your path to knowledge .. once you gone through seven cycles of diffs ... I think , you will be able to maintain the code .. without having to run back to AI ..

nostr:nevent1qqs9fuv3yg3z93ax6qjtcz2l9ey03r0ua4jh6s735sa7u7nas7leskgpz4mhxue69uhkummnw3ezummcw3ezuer9wchsygz67pu5vp4ptdtyrcj65g7sft6vkrtate5tz89vk3l9dfrf3l9gcspsgqqqqqqszummk0

AI Driven Life Cycle

First ask #Gemini to write the specs ... Limit to say #Rust and #CLI - minimum feature - Build code .. resolve bugs .. #Prototype

Then ask it to write the Tech Specs - insist to include all the bugs from prototyping phase - modularize code - test and squish more bugs - update tech specs #Dev

Then ask to write test cases - Squish more bugs #Test

Put the code on #github - open source - and deploy where ever you want . I recommend building only #CLI apps for #Linux #Debian - cuz that is the only place test - #Deploy

But here is the key - develop ONLY for yourself . Which means the app MUST be a part of your work flow ...

Why Rust ? - Cuz Rust wont let you compile code until all bugs are fixed . Which means you are free from AI misgivings .. besides Rust being the the fastest modern language .. that actually works !

nostr:nevent1qqsdz9m23dzm7nwp4v6fkpfyuxy98hgm8zlat8nxmf60j855znhjavqpzamhxue69uhhyetvv9ujuurjd9kkzmpwdejhgtczypd0q72xq6s4k4jpufd2y0gy4axtp474u693rjktgljk535clj5vgqcyqqqqqqgj7353h

Reply to this note

Please Login to reply.

Discussion

I like #Rust but the problem is its steep learning curve. Its fine for side hobby project but only few companies are actually adopting it, because it is much cheaper to write and maintain good quality code in #python or #golang than the whole dev team learn and migrate to rust. Just my experience.

Steep learning curve is the very problem that AI solves :-) .. Writing rust is hard .. but reading isn't !

But the key thing is RUST's compiler is THE GRAND GATING FACTOR. It wont let AI sneak in random bugs .. which is often does no matter how much you test ..

So in a way , the rigor of writing code yourself is transferred into resolving the compile time bugs .. btw , if you invest time ... Those bugs are the real learning opportunity .. that is your opportunity to dive deep into Rust .. assuming you have read first few chapters of THE BOOK !

Btw ... With AI .. the whole idea of writing code in a COMPANY/ Group setting is going to be disrupted .. I mean , no one writes a BOOK as a collective ... Writing a song is a personal thing ... Same way ... Future of code is to write all that you need yourself .. mostly for personal use but shared with close knit groups .. some ideas will bubble up just like some good poems do ...