HamadMalikKhan

1. The keys are on server side. Every section has a separate file on the server with full access to the context (variables), so it renders on the runtime.

Here's how it works Browser -> Request for section -> Backend (with full context and keys) -> Generate it using AI. We are using Deepseek 4.1 flash, so latency is minimal.

2. So far the ad spend is super bad and it's not improving. It tries to correct things too quickly without giving it a time to settle down. When I tried to force it to do long horizons, it stopped spending for days because it was too focused on one grand plan.

The prompt is literally "You have $1,700 and access to these 4 platforms. Do your best to get more people try the demo (core KPI).". We are using Fable 5.1 for Ads.

I wrote under 5% of the code by hand. It was mainly telling the AI to build x and how to build it. With the new Fable 5.1, even the how part is getting replaced.

For this project, the maintaince is not too big of a pain, but we are working on another project where it's an absolute pain. Every PR is like 100k lines change, simple bug fix needs 600k tokens. We are probably doing a choppy job here.

The size of the codebase definitely plays a role.

Software engineer here, so I am unsure if it counts but we built a product that's doing ~$500k/year and ~95% of it is vibe coded.

The difference would be, we knew what we were doing (This makes me sleep better at night.), because of pre LLM era experience.

Where's the benchmark? I am afraid the model will be nerfed and replaced with an orange after the hype.

Anything that can't be solved with a regex and LLM feels like an overkill. There are surprisingly tons of things that falls in this category. E.g Filter, classification, preprocessing micro decisions, etc.

I am starting to like Typesafe more and more lately. It's plainly simple but the next time I see the same problem, Jev just fits right.

Super interesting! I can see 273,229 records from the network requests. Out of curiousity, how do you plan to keep it updated?

I can feel like a sultan for the first time. Unfortunately, it's Bayezid.

Can you specify the Stack or things you are trying to build? With Claude Code, the dev tool space is getting exciting and blurry at the same time.

This, plus the implementation on the AI providers' side is still choppy. I remember one of our implementations had a $500 limit on OpenAI with auto-recharge enabled. It burned through $600 despite the lower limit.

You can be down serious money even with a limit set.