Claude and Gemini crush system design and high-level modelling, but the moment you drop into actual function logic, they fall apart. It s like pairing with someone who writes clean diagrams but chaotic code.
And if you ask them to improve it, they vanish into SOLID-theory rabbit holes and over-engineered OOP madness.
These days, since I vibecode more, I have found myself refactoring what the robots are writing ever so often that it has got to a point where I straight up delete function logic and write it myself.

Curious if others are seeing the same?

Hello everyone, recently I am refactoring old AI codes in hot spots. I used both Claude Code and Codex to quickly implement the feature at that time. It works but when I look into the logic today, it introduces too much unnecessary complexity (a lot of helper/manager/try-except). Although I have manipulated CLAUDE.md (emphasizing KISS principle, introduce Linus https://gist.github.com/iiiyu/4c... ), the code agent still try to add entropy on the whole project. I can understand the code LLM is trained to program defensely, but if I do not review carefully and really understand the logic, the project quickly becomes hard to maintain. Now every week I will leave 1 day to write codes without AI to clean the whole project for longer future.
Do you have similar experiences or solutions to share?

As devs, we all know the struggle with code security scanners:

• Switching between IDE and clunky dashboards to check vulnerabilities

• Wasting hours sifting through false positives

• Getting vulnerability alerts with zero guidance on how to actually fix them

I stumbled on an IDE extension (ZAST Express for VS Code/Cursor) last week that s amazing for my workflow. The Proof of Concept (PoC) feature is what sold me instead of just flagging issues, it gives runnable snippets to validate the problem, plus clear fixes right in my editor. No context switching, no guesswork, and it s free to start.

I hear people liken the AI rush to the blockchain rush of 2017-2021. 

I used to work in the blockchain world, and having had an intimate view of both worlds, I think these couldn t be more different. 

Hey Product Hunt community!
What do you guys think are the best startup incubators in the world?

This is my current list, would love to get your input:

1. FR8 (Helsinki, Finland)

2. Founders House (Stockholm, Sweden)

3. Founders Hub (Oslo, Norway)

4. Founders Inc (San Francisco, California)

5. Idealab (Pasadena, California)

6. Rocket Internet (Berlin, Germany)

7. Science Inc. (Los Angeles, California)

8. Station F (Paris, France)

9. 1871 (Chicago, Illinois)

10. Cambridge Innovation Center (CIC) (Boston + worldwide)

I ve been talking with a lot of founders and support leaders lately, and something interesting keeps coming up. Many teams want to introduce AI into customer support, but the onboarding and setup process often becomes the biggest hurdle. Not because the tech isn t good enough, but because teams worry about training, accuracy, and maintaining a smooth customer experience during the transition.

I m curious how others are navigating this. Are you rolling AI out gradually? Testing it on internal queries first? Letting it assist agents before going customer-facing? Something else entirely?

Learning a new tech stack can be overwhelming. Too many tutorials, too little structure. That s why we built Edulavo.

It s an AI-powered skill coach that generates personalized learning roadmaps for you instantly. Whether you want to master React, Python, or Productivity hacks Edulavo guides you step-by-step.

Hey folks Ryan here

I ve been chatting with a bunch of vibe-coding builders lately, and everyone keeps saying the same thing:

Prompts get you UI fast but the moment things get real, everything starts to fall apart.

For more than a year now I have been consistently working in a vibecoding format. During this time I have created several unusual solutions and finally decided to start showing them publicly.

These are small research pieces that can inspire and at the same time remain utilitarian.

Hey everyone

Taskade started as a real-time collaboration tool for planning and productivity. Then we added memory, agents, and automations. Soon it stopped feeling like a static tool and started acting like a real living workspace that could handle parts of the work on its own.

I ve hit a few demo nights in SF with some gems and some straight-up misses. I m chasing the high-signal ones where you leave with something you can ship next week.

What s the best demo you ran or attended recently, and what made it great? If there s a series or event worth following, drop it (city or online both welcome).

I keep seeing the same problems over and over in job applications. Across LinkedIn, emails, ATS systems, referrals, and technical interviews:

For candidates:

• different recruiters ask for CVs in different formats

• no clear expectations for the process

• ghosting after interviews or coding tests

• positions closed with no notification

I ve been building full-stack applications for about 14 years mostly with Python, a few with Node, PHP, and Ruby.

Lately, my workflow has changed drastically.

With tools like Cursor, Claude Code, Codex, and Gemini CLI, generating reliable, production-ready code has become incredibly fast. These AI coding agents don t just accelerate development they often remove the friction entirely.

TikTok, you must stop?

Chinese tech regulators won t let ByteDance use Nvidia s chips in its new data centers. The TikTok owner needs a lot of computing power to keep its 1 billion+ users watching lip-syncing videos. But Nvidia is a U.S. company, and China wants data centers made with homegrown chips. (Reuters)

Just in time for Black Friday 

Location, Location, Location

Robotaxi company Waymo is expanding across California. Already active in SF, Silicon Valley, and LA, it now has a green light from California regulators to cruise into Napa, Sacramento, San Diego, and many of the bedroom communities surrounding those areas. 

It s easy to see how Google Maps and Waze, both owned by Alphabet, will fit into the plans of a self-driving car company also owned by Alphabet. But as self-driving cars take off and vehicles become something of a third space, how might app usage change for passengers? 

Will you still want to listen to a podcast if your hands are freed up? Would you rather have apps built into the center console? Or will it just be more of the same: scrolling Instagram and checking email on your phone until you arrive?

Lately I ve been talking to other founders who struggle to bring structure to growth: everyone s testing ideas, but few track experiments systematically

I ended up building a Notion system to keep everything in one place (diagnostic sprint plan hypothesis scoring go/kill decisions)

Curious what you all use - Notion? Airtable? Sheets?

Here s mine if you want to peek: https://lake-skateboard-ad2.noti...

Hey! I'm an AI vibe coder like many others here.

I mainly use ChatGPT, Claude, and Gemini.

What I have seen so far is a split. Some power users go all in on agents like Claude Code using Skills, Hooks, Subagents, MCPs and other advanced techniques

Others only use lighter tools such as autocomplete or Cursor. Across most teams, adoption seems inconsistent

Hey Product Hunters and fellow builders!

There's a significant development coming out of China's AI scene that's worth our attention: Volcengine (ByteDance's cloud service) has officially launched its first dedicated coding Large Language Model, the Doubao-Seed-Code.

While the AI code generation space is crowded, this launch is specifically noteworthy for its aggressive positioning on both performance and, crucially, cost.

When Android & iOS arrived in 2007 08, it kicked off a golden era for indie developers. From simple utilities to quirky experimental apps, anything could thrive, even without high-end graphics or complex systems.

That window kept expanding till ~2017.

But post-2021, everything changed.