Steve | Systems & AI
Region: US
Saturday 21 February 2026 13:20:16 GMT
16581
598
42
69
Music
Download
Comments
Sloth :
This is one reason companies will start to seriously consider breaking up their monoliths. Before it was "that's a tech debt problem. Engineers make the same pay regardless of how long they have to fix things". Now it's "hey, get them to break up all this code into small maintainable modules so ai can efficiently handle each piece and we can get rid of 90% of engineers"
2026-02-21 14:27:04
12
Brian 🇺🇸 :
Openspec with Grounded Docs is great for larger codebases.
2026-02-21 17:48:44
0
gamedevmicah :
Based
2026-02-22 15:28:46
0
William :
Whats small? Medium? Large? In regards to lines of code. I’m up to 100k lines of functional code in my SaaS and anything I do now seems 10x more fragile or capable of screwing something up - in regards to adding more features especially. All AI built no previous coding or software experience for context
2026-02-23 01:15:10
1
damir🇭🇷🇪🇺 :
Gemini 3.1 helps 2 Milion token window
2026-02-22 08:58:11
0
bamaham93 :
Microservices getting a new lease in life!
2026-03-17 04:28:27
0
🫧 :
This why we gotta learn language so people can point ai in the direction it shall go
2026-02-22 10:52:45
0
Grumpy IT PM :
That’s why you have documentation. You need it for humans too.
2026-02-22 20:20:37
0
popup580 :
quality over quantity
2026-02-22 20:40:14
0
robo.trick :
Then program it yourself; it's faster.
2026-02-22 03:10:51
0
Undefined :
100% very valid 🔥. Struggling with that now. Feels like I’ve changed from dev to project man and it’s eating soo much time and and killing my execution loop!!
2026-02-22 21:48:29
0
Todor Popov :
This! The other day I had to update around 10 e2e tests because a new required column was added to a table. That particular app uses a custom Playwright testing framework, that takes in specific JSON files as steps of what the framework needs to validate. Most of these files are a couple thousand lines long. Copilot had a seizure trying to understand what was going on in the files. It literally took a couple minutes for it to (try) update a single file. It made so many mistakes that after the first few test runs I just rolled back the changes and did everything manually in like 30 minutes. Keep in mind I spent a lot more than that trying to explain the task, let the agent run, and run the test only to see the million mistakes.
2026-02-22 12:36:40
0
SpinNexus :
This hasn't been my experience
2026-02-22 11:03:44
0
Lu :
could you just put an md in every folder but slightly modified for each folder
2026-02-21 13:41:14
1
Teycircoder1 :
Yes. It forgets what it already did. Does forget even rules. Currently you need a lot of modularity in the code to still be able to use it.
2026-02-23 17:13:30
1
KaMabui :
facts facts 💯
2026-02-28 14:30:01
0
Giacomo :
small patches even in large code base are ok; for brand new features i usually do ad standalone prototype, after that i shape it like a library, and finally i work in the original repo but just for integrations
2026-02-21 14:46:46
1
noO_Oon :
Memory….baaaaank! Got a cloud software here that was built by multiple hundred people over 10 years. Spent 2 weeks explaining a basic concept in that repo just to do a very fundamental task + branches and pull requests in our CI/CD. Got it running now, let’s see how deterministic it is. 😎 So far, LLMs are performing like enthusiastic interns.
2026-02-21 13:37:10
0
Ekow Gravity :
I learnt this the hard way..you have to clearly define your trade off and patterns so the model can stay on course and the important part is to allow another model or agent review large refactor or implementation for gaps and edge cases missed
2026-02-21 15:47:00
0
💻🅹🅾🅷🅽🤖 :
This is great information for me! I’m learning so much tidbits I need to know
2026-03-26 16:06:48
0
masterplanner :
Currently building and you're definitely not wrong. I've found it helps to have those.md's like you said
2026-02-22 02:23:11
0
boombanger :
You need a SOLID principal model and I can see summary and API models being held in context as they get bigger.
2026-02-21 22:33:57
0
liam_mcpoyle0 :
Your code needs structure and that structure should be well documented in readme.md files at the base of the repo and each subdirectory and each feature etc. In addition to your coding style and guides in rules. The more structured the easier it is for the LLM.
2026-02-21 21:00:23
0
DaisyNova :
Why I switched about a year ago and do all ai coding In elixir. It will be the winner
2026-02-21 17:22:08
0
cezarbarca :
Yeah, I found my way to work in a big repo. Instead of providing just a feature description, I often provided one module for structure and this work great
2026-02-21 13:46:58
0
To see more videos from user @sd141414, please go to the Tikwm
homepage.
