Hey makers!
I've been deep in prompt engineering lately while building an AI tool, and I'm genuinely curious about how others approach this.
A few questions:
1. Do you save your best prompts somewhere? Notion, text files, dedicated app, or just copy-paste from chat history?
2. How do you iterate? Do you have a systematic approach or just tweak until it works?
3. Different prompts for different models? Or do you use the same prompt for ChatGPT, Claude, Gemini?
4. Text vs image prompts do you treat them completely differently?
I've noticed I was doing the same optimizations over and over (adding role, being more specific, structuring output format), which made me wonder if everyone has their own "prompt formula."
Would love to hear your workflows!

Here's my hacked-together, messy, voice-based dev environment:

1. Voice-driven loop with screen-shotting so the LLM in the loop can see what's in my terminal and editor. The prompt varies depending on what I'm trying to drive with this loop.

2. A few tool definitions that give read access to files and URLs.

3. A tool the LLM can send a block of output to that generates keyboard events, so the LLM can drive any editor/terminal.

4. A separate process watching a directory and constantly making LLM-driven git commits. (git autosave).

I have some pieces of this running most of the time. But I'm lazy, and doing other stuff, and I also try to use a variety of editors and tools, to see what's good lately. Which ... no stability, so my hacked-together stuff is always broken.

I don't want to replace @Windsurf / @Cursor / Claude code. A seriously good agent and expert-system dev toolkit is a lot of work.

Two hours into a refactor yesterday, my agent wrote a helper function the exact same one it wrote 90 minutes earlier in a file it created itself. Then it apologized. It always apologizes.

The easy diagnosis is "it ran out of context". Except my session was sitting at 120K in a 200K window. Nothing overflowed. The context didn't run out it rotted.

You have two budgets, not one:

• Hard Budget: Token limit. You notice it when the API errors out.

• Soft Budget: Attention quality. It drains silently long before you hit the limit.

Hey everyone. My name is Talha Masood and I am a second time founder. Previously I built Remotebase[dot]com and raised $2.7 million for my startup.

I am now building an AI editor to help interviewers assess vibe coding skills in software engineers. I myself vibe code every single day, even though I have more than 10 years of experience writing code. I believe that software teams can now move 10x faster if they're using the right AI workflows and tools.
Here are a few facts about me:

• I have launched close to 10 times on Product Hunt

• I organize a lot of IRL events in San Francisco. My latest event is happening on April 24th. Here's the link to it if you're interested - https://lu.ma/5s9vc24j

• I can write code 10x faster than an average software engineer by using the right Vibe Coding tools. Please challenge me. P.S. I came up with the number 10x out of thin air lol. But I do a lot of Vibe Coding

• I am also looking for a technical co-founder right now.

I do frontend for a living, and lately i can tell almost instantly when an app was vibe-coded. same component library, same gradient cards, same spacing that's slightly off everywhere.

the code works, it just all converges on one generic look.

vibe coding killed the barrier to building, which is great. but it also killed the friction that used to make products look different from each other. the models learned from the same patterns, so everyone lands in the same place.

I started using Replit about 3 weeks ago. First on a trial basis and then decided to pay the quite ok monthly fee of $25. What I didn't realize back then was that once they got my credit card info they just keep charging for use even if cancel my sub.

The thing is that Replit charges 'checkpoints' a 25 cents each no matter what it does. Implementing Google Auth with 2FA or failing over and over again to format an output i Typescript, same result. The checkpoints just runs away. When the AI fails to solve a very basic variable naming problem for the Xth time you tend to get a bit frustrated - to say the least. Tell the AI to mind the difference between CamelCase and snake_case and it remembers it for less than 5 minutes. Putting it in writing in the manifest Instructions.md makes no difference.

After about a week I decided to make a first deployment. The AI ran all sorts of test and then deployed. Successfully in its own words. Except not a single endpoint worked in deployment. Not even logging in to the Express backend as admin. So I asked the AI if it thought it was ok that code that worked in the Replit dev env did not work in their own deployment env (based on Google Cloud). It wasn't. So I sent a mail to support. Which yielded absolutely nothing but a boilerplats BS answer. You see, the support too is an AI. How clever is that!

We've been getting the same request over and over from our users: "My AI gives great answers, but it can't actually do anything."

It got us thinking most AI integrations today are still essentially fancy search boxes. The AI talks, the human acts. But the real unlock is when the AI can close the loop itself query the database, send the email, update the record without a human in the middle.

The hard part isn't the action itself. It's the non-determinism. How do you build a system where the AI decides when to act, which action to take, and what parameters to pass based purely on context without it going off the rails?

A few things we've learned building this:

We started routing the easy stuff to a small model to save money. Cost per request dropped about 40% and the dashboard looked great. What the dashboard didn't show was people re-sending the same message, so once you count the retry and the follow up clarification, the cheap path burned more tokens than just sending it to the big model once.

The mistake wasn't the routing. It was measuring cost per request instead of cost per finished task. A request that has to be asked twice isn't cheap, it's a request plus a support ticket forming.

Hey everyone,

I don't actually like using the term "vibe coding". We've been software developers for over a decade ,are not one-shotting features, and have a very opinionated and strict dev process.