The Code Was Never the Product.
AI didn’t come for your job. It came for the part of your job you were secretly tired of anyway — and it’s about to make the rest of it worth ten times more.
Read my blog on medium.
https://t.co/9eQZ5K6VeM
Is anyone even using OpenAI models now?
In my circle, most of the people used claude but they are slowly moving away from claude to cheaper open source alternatives like deepseek, mini max and hy4.
I think, using claude is waste of money especially when you don't require advance level of reasoning. There are cheaper models who can do the coding at a really high level when executed under predefined rules set by developers.
Just give the cheaper models rules amd criteria to operate under and they would easily complete your day to day coding tasks.
If you are interested about learning AI and in depth concepts related to it, check out my medium blog.
80% of AI Leverage Comes From 5 Mechanics:
https://t.co/fox45I5PMA
It then compared the same for other providers and provided a comparison table with data of cost and what all providers scale to zero and then agreed. It also gave me the data if we get traffic and as the traffic was increasing the gap between gcp cloud run and other providers was becoming narrow, and eventually hetzner beats gcp but at a very high traffic count. SoIt was kinda useful. Based on this data it agreed that gcp cloud run is the right choice now and we should deploy to gcp now and if traffic grows to that number, we should switch to hetzner, and we should make our deployment pipeline in a way that it is very easy to switch.
Which AI models are you using for building software?
I have completely switched from claude to Deepseek via opencode Go, even for planning and architecture.
I create an architecture and asks deepseek to challenge me and then we both work on refining it and later finalize it.
Able to save a lot of money and deepseek is one of the best models ever, can even call it a revolutionary model.
Can do everything to a good level if you direct it properly with appropriate checks.
@babykswanson LLM is used for planning how the video should be rendered, what sbould come first, what should come at end etc. but not no video model is used.
Hey Guys, 👋
I am Ahmed, Senior Software Engineer with 10+ years of exp.
I am taking up a new challenge to build solo and create SaaS, I would be posting about my entire journey here on X
Looking to connect with fellow founders, builders and people in tech.
Let's connect.
I give it context on which things it must argue with me on.
For example, recently I was having a discussion on where we should deploy our workers on, the parameter was price for a product which is brand new and unpredictable traffic. It could most likely not receive much traffic early on. I suggested gcp cloud run but deepseek suggested cloudflare or hetzner. I argued why cloud run is better for our use case as it can scale down to zero when idle and gives a generous free tier which would most likely make our running cost 0 until we get some actual users and deepseek agreed with it after a few back and forth.