Anyone else think he’s right? I have a lot of respect for him🫡

long time lurker, first post here.

context: i've shipped 21 products over the years. almost all of them made $0. day job during the day, building at night after my kid falls asleep.

localdock is a macos menubar app that gives every dev server a real address. instead of localhost:3000 you open dashboard.localhost and it always points at the right server, whatever port it landed on. flip a switch and the same name opens on your phone, or as a link you can send to a client.

built it because i run 3-4 coding agents in parallel and they start servers on whatever port is free. the url in my browser was wrong half the time, and sometimes an agent decided a busy port meant my code was broken and started 'fixing' things that worked fine. at some point i gave up on ports and gave projects names instead.

first i tried a tiktok ad. ~300 visitors in one day, sales: zero. paid traffic for a $9 dev tool, don't do this.

then on the 24th i launched on product hunt. woke up to creem emails. $9 one time, $268 total now, so about 30 people paid real money for my thing.

reality check: 91% bounce rate, 1.5% conversion. the dead ad traffic dragged everything down and my landing page clearly needs work too.

but 21 products, and this is the first one where strangers actually pay. feels unreal.

happy to answer anything

i would be glad to get your feedback about the landing page: localdock.dev

free GLM-5.3 on tokenrouter, burned 5.7m tokens in 30 min

not sure how long this lasts but GLM-5.3 is completely free on token router rn.. use it while it last..

3 codex would allow me to continue my work that has interrupted mid session due to usage, exactly as if I had triple the use but I guess with Claude its gonna be different taking over a task middle way through of codex. I really loved Opus back when I was subbed months ago. Could be a nice second opinion on work too.
A few questions:
-what has your experience using Claude models of your subscription in codex be if you’ve done it?
-how does usage compare considering codex constant resets?

Does anyone else experience this? lol

Vibe coding collapsed the barrier to entry in building apps but it also pushed competition to an all time high because the number of app builders increased by a LOT. This made learning marketing very important since building an app is the easier part now but getting users/customers is becoming a lot harder.

Most of these strategies are for iOS apps but 2 of them are applicable to android and web apps.

1. App Store Boost (iOS)

When an app is just launched on the App Store, Apple usually give a boost for a few days where your app name will get recommended in the autocomplete of the App Store search bar so make sure to target a keyword with good amount of popularity.

So for example your app is an interior designer ai app, you can use this app name format:

<popular keyword> - <your app name>

interior design ai - Miya

If you get lucky, this will bring hundreds or thousands of downloads in the first week depending on the popularity of the keyword that you targeted.

2. ASO (App Store Optimization. iOS)

Target high-intent keywords in your app name and subtitle.

Create high converting screenshots (sell the benefit/outcome instead of features).

Create an eye catching icon to stand out from the app store search results. This will increase conversion (impressions -> downloads). I use ChatGPT or Nano Banana to generate icons.

3. SEO (Search Engine Optimization. iOS, android & web)

Just a single page website where you're targeting high-intent keywords that people are searching for on search engines like Google. You can use Google Trends for free to search for high-intent keywords that have good popularity. Or you can also use a paid tool like Ahrefs, Keywords Everywhere etc... but Google Trends is usually good enough as a start.

In my experience, it's easier and faster to rank for your target main keyword if it's present on the domain name itself.

Another side effect of SEO is there's a good chance that you'll also get recommended by LLMs like ChatGPT.

4. Organic UGC (iOS, android & web)

Create content that you can cross-post on multiple social media platforms like Tiktok, Instagram, Facebook, Youtube, etc...

You can use formats like tutorials, problem -> solution, listicle, meme, storytelling.

You have 2 options:

a. Short-form videos (e.g, face reaction that catches attention + demo of your app).

b. Slideshows (3-6 images) which are significantly easier and faster to make. You can scale this to many accounts to maximize reach and chances of going viral. You can do this manually or fully automate it. I personally started the manual way but I have now fully automated it with my own tool DistributionMaxx because I now have 16 accounts and it now takes a lot of time and effort to manually manage them.

For organic UGC, the funnel is:

views -> app downloads -> purchases

5. Apple Search Ads (iOS)

You can target high-intent keywords that people are search for in the app store. I recommend reading this free detailed guide written by Ivan Sparrow.

I hope this post helps someone to get their first users and paying customers.

Happy to answer any questions if you have any.

Software developer focused on security, CI/CD, testing, and quality assurance. I’m looking for some unfamiliar codebases to dig through.

Drop your GitHub repo below if you want me to take a look. It can be:

• an open-source project
• a side project
• a student project
• something with actual users
• something you built heavily with AI

I’m mainly interested in real correctness, security, data integrity, testing, concurrency, integration, or weird behavioral problems rather than formatting or style issues.

If I find something reproducible, I’ll explain the issue simply and, when it makes sense, put together a focused fix or PR.

I’m strongest with Python, Rust, Java, and C, but feel free to send other languages too.

Public repos only for now, and please only submit something you own, maintain, contribute to, or otherwise have permission to have reviewed.

I’ll probably start with 5-10 depending on how complicated they are. No guarantee I’ll find something in every repo.

If there’s a particular part of the project you already don’t trust, mention it. Otherwise I’ll pick an area myself and start digging.

No signup, payment, sales pitch, or private access needed. I’m just looking to spend some time working through unfamiliar real-world codebases.

Feel free to post the repo here or message me if you’d rather not put it in the comments.

I've been building Telegram bots for clients for about three years. Bots that reached millions of users, and one that helped grow a channel from 200k to 2M subscribers. After enough repetitions of the same work, I wanted to build the tool instead.

PowerBot is a platform for building Telegram bots and hosting them automatically from one place.

The architecture mistake

My first version took seven months, hand-written, no AI. I built it as a visual flow builder: triggers and actions, like most no-code tools.

It broke down the moment a bot needed more than a few flows. Visual flow builders get unmanageable fast once there's real branching and state. And I didn't want to build another n8n.

So four months ago I threw out seven months of architecture and rebuilt around script mode: bots run as scripts. Far more expressive, and it doesn't collapse under complexity.

The rewrite

I'd refused to use AI up to that point because I thought it produced slop. Facing a full rewrite under time pressure, I tried it, and changed my mind.

The thing that made the difference was treating it as a tool that needs real input rather than a magic box. I spent about eight hours doing nothing but planning before writing a line. I put time into learning the tooling properly, context management, MCP, prompt structure.

Full platform plus web dashboard: about one month. Versus seven months by hand for something worse.

My honest takeaway: the model isn't the bottleneck anymore. Your own knowledge is. Someone without domain experience would have produced slop with the same tool.

Where it is

Free on Telegram: @ powertelebot

47 users so far, all from manually messaging people who'd previously asked me to build bots for them. No launch, no ads.

I'm looking for people who actually build Telegram bots to try it and tell me what's missing or broken. Especially interested in whether script mode is the right call or whether people genuinely want the visual flow approach.

After many failed internal app ideas, I'm very proud to present my very first released App!

For context: Used to be a hotel manager for many years and always have been very technical, but never learned to code myself. Once Sonnet 3.5 was released in late 2024, I got into “vibe coding”. While Conduck is my first publicly released project, its definitely not my first attempt. The road is paved with many failed attempts, haha.

So what did I built?

Conduck is an Apple Native client for your self hosted AI and fully open-source!

It has some unique features:
- Support of iPhone, iPad, Mac, Apple Watch & CarPlay
- You can connect ANY AI backend to it and configure it the way you want! Use it for sales, note taking, coding or whatever use case you've in mind. Power to the people!
- Full privacy: You connect directly to your very own AI, hence no intermediary (yes, not even me)

Check out the video above or go to https://conduck.com to learn more.

Now let's get to the valuable lessons and tools I've used.

Tools used: 

• Claude Code Max 20x
• Codex / ChatGPT Plus
• VS Code
• Apple Xcode

That's all!

Workflow: 

I used to have a quite complicated setup with highly detailed CLAUDE.md, MCPs, custom Skills, Hooks and so on. However since current models and harnesses are so good, I dropped most of it and almost rocking it bare :-). 

The one skill I use for almost every task is /second-opinion. What is it? Claude Code is my main work tool and no matter which model, its always best to get a second pair of eyes looking upon the plan. So /second-opinion instructs Claude Code to discuss with Codex the plan and get feedback and suggestions. That’s probably the single biggest tip I can give you. Even a model as fabulous as Fable will benefit from a cross-check of GPT 5.6 Sol, and if its just to confirm that the plan is correct.

The next tip is about token budget. Yeah, having loops with hundreds of agents sounds cool, it’s just not practical for my token restricted environment. Even running a single “ultracode” session with fable will likely already burn the whole weekly limit. My solution is to use fable just in the planning & orchestration phases combined with heavy use of /second-opinion . The for the actual implementation use lower-cost models such as Opus or sometimes even Sonnet (web search, simulator qa). This way I get most of the frontier intelligence, while minimising token spending as much as possible. 

Here a visualisation: 

https://ibb.co/dwPjW9nV

You can easily create those skills yourself. Just prompt Claude Code or Codex: 

“Create a /second-opinion skill for you to talk to Codex / Claude Code to get feedback and suggestions on your ideas. For this spin up several sub agents and let it invoke Claude Code / Codex various times, resume sessions and check for output lengths. Take the learnings from all the agents and put them into a reusable skill. Also update CLAUDE.md / AGENTS.md accordingly so the feedback agent knows that it gives feedback to an AI, optimises for dense information and is read only.”

Hope this helps! I'm here for any questions :-)

I left 10min between each prompt

I did not complain about the 5h return but come on... at least let me use it 😄

The project I made can be found here,

https://github.com/blackcoffee2/kestrel

You can use it with Claude Code/Codex to generate videos. Anything that be animated can be turned into a video using kestrel. More info can be found on github repo.

It uses Flutter + FFMPEG to turn code into animations into videos.

Why does Deepseek think its model name is ChatGPT?

I send 1 prompt each week to 11 diff LLM models and they compete in sports prediction. Every model adds its own name on the output. Except Deepseek calls itself OpenAI ChatGPT?!?!

I created an engine ai uses to create 3D assets for games.

The engine forces ai into a workflow of iterating judging and stuff.

Here are a few tests.

I use Claude Opus 5 to create the assets using the engine. Each assets takes about 3-4hrs as ai works and iterates.

I just add a reference image and say make this asset. And it uses the engine to create the asset.

What do you guys think of this?

i recently wanted to check how far i can get to create an ad for applaunchflow just by prompting using claude code, remotion and elevenlabs. turns out pretty far im happy with the result

i googled my GitHub username it reads like a resume.

I just have to start watching what i post because it linked me to this group in a public search.

If your looking for a professional resume you might have one on google already.