Always wanted to use this, seems really convinent.

https://thevalleyofcode.com/git-worktrees/?ref=dailydev

I don't think I could have learnt anything if there was an easy way out like this. I'm starting to think that AI is going to dominate because it will degrade human intellect to the ground.


https://x.com/karpathy/status/1992655330002817095?t=QEBwGj-JuNy5tS8e5JG58w&s=19

On other news, AI is ruining another market.

not going to lie, the workflow of making detailed plans with step by step architecture things you need to decide on with the ai before building anything, and then storing it in an actual live architecture.md file is pretty fire

so have you ever been using an LLM chat like chatgpt and it suddenly became like 80% dumber when the conversation gets a bit longer?

it's because the llm has a finite context window, 270K tokens for gpt 5.1, and when it approaches it, it automatically summarizes your whole conversation into a summary text and starts a new conversation from that tiny summary. basically forgetting a lot of the minor details

in opencode you can see the context window and manually compact it when you are at a stopping point of a problem or something. better to compact and summarize it on your own terms than it cutting off randomly in the middle of something important.

making it write the important things inside of a text file is also a good habit.

One of my favourite videos I've seen in months.

https://youtu.be/mfv0V1SxbNA

i just realized that i am using opencode with wraps codex-cli which wraps gpt-5.1 to build an ai wrapper over mistral ai

it's litterally

ai_wrapper(ai_wrapper(ai_wrapper(ai_wrapper( ))))

i'm cooked and i have a week left for me to live with 40% when i speant 60% of it in one day

guess it's back to dumb models.

Happy that I am working in the greenfield side.

A greenfield project is starting a software system from scratch with no existing code or infrastructure constraints.
A brownfield project is developing new software or features within an existing, operational system, requiring integration with legacy code and infrastructure.


No Vibes Allowed: Solving Hard Problems in Complex Codebases - Dex Horthy, HumanLayer

Apparently when we are in a loop of the ai does something wrong and we yell at it to fix it, and it does something wrong and we yell at it... All an LLM is is a next word predictor looking at the previous words. So it looks at the previous fight and predicts the next thing to be it doing something wrong and you yelling at it. So it gives you the prediction of it doing something wrong.

Damn, I've never thought about it this way.

https://youtu.be/rmvDxxNubIg?t=5m31s

how did i not know that react and react native are coming under the linux foundation????

https://x.com/fb_engineering/status/1975935750509343006

lowkey i really like opensource projects, like people were bugging the opencode team to add oauth for providers like atlassian last week and today it is merged and added to the docs.

btw, the jira and confluence MCP is apparently really cool, you can summerize tickets and search through whole company documents stored in confluence to understand (dig up secrets) about your company and projects.

just asked it if what we've been doing for the past couple of hours is the wrong way to do things and it replied this. I hate AI coding, like not gonna lie. now that i have built the incredibly weak foundations of the project really fast, editing one thing takes so long and is just a pointless spiral. you don't know what is wrong till it blows up in your face at the end.

I need to start using tab complete instead of agents.

https://youtu.be/MNvciiZ5x9A

This is cool

https://youtu.be/UALxgn1MnZo

And the cycle goes on, gonna test it out inside of codex.