There's a growing sentiment in tech right now that I think is worth paying attention to. People are getting tired of AI.

Not tired of the concept. Tired of the experience. Tired of wrestling with ChatGPT to get it to write code that actually works. Tired of Copilot suggesting things that look right but break in production. Tired of spending more time prompting than building.

A thread titled "LLMs can be exhausting" hit the front page of Hacker News recently, and the comments were full of developers nodding along. The frustration isn't that AI is bad. It's that using general-purpose AI for specific tasks feels like hiring a brilliant generalist who needs constant supervision.

The supervision tax

Every time you use a general-purpose LLM to build something, you're paying a hidden cost. You write a prompt. You review the output. You spot something wrong. You re-prompt with corrections. The AI apologizes and changes its answer, sometimes breaking things that were already correct. You go back and forth. Twenty minutes later, you've got something that mostly works but you're not quite sure you trust it.

Some people call this the "are you sure?" problem. LLMs are so eager to be helpful that they'll flip a correct answer if you push back even slightly. They're optimized for agreeableness, not accuracy in any specific domain.

For developers, this is annoying but manageable. They can read the code, spot the bugs, and fix things manually. But for non-developers trying to build apps with AI? It's a dead end. If you can't evaluate whether the AI's output is correct, the whole promise of "anyone can build software" falls apart.

The fix isn't better prompts

A lot of AI discourse right now focuses on prompt engineering. Learn to write better prompts and you'll get better results. That's true, but it misses the bigger picture. If you need to become a prompt engineering expert to build a simple web app, we haven't actually made software development more accessible. We've just replaced one technical skill with another.

The real solution is specialization. And we're seeing this play out across the industry. General-purpose AI tools are being replaced, or at least supplemented, by domain-specific ones. Cursor for code editing. Harvey for legal. Elicit for research. The pattern is clear: AI gets dramatically better when it's trained and optimized for one thing.

This is the bet we made early at DontCode. Instead of wrapping a generic LLM in a nice UI and calling it a day, we fine-tuned our AI specifically for application building. It knows what a SaaS dashboard looks like. It understands how auth flows work. It generates database schemas that actually make sense for what you're building. And because everything comes pre-configured, there's no gap between "the AI generated some code" and "I have a working app with a database, login system, and live URL."

The difference shows up in first-try accuracy. When your AI has seen thousands of app architectures and deeply understands the patterns, it doesn't need five rounds of corrections to get the basics right.

Where this is heading

I think AI fatigue will accelerate the shift toward specialized tools. People who tried building apps with ChatGPT and gave up aren't going to give up on AI entirely. They're going to look for tools that work for their specific use case without the exhausting prompt-debug-reprompt cycle.

For builders in Korea, this shift matters even more. General-purpose LLMs barely understand Korean business context, Korean payment systems, or Korean user expectations. They'll generate a Stripe integration when you need KakaoPay. They'll build English-first UIs that feel like afterthought translations in Korean. Specialization isn't just nice to have. It's necessary.

We're building DontCode for this exact moment. The DontCoders on our platform aren't prompt engineers. They're entrepreneurs, product managers, and small business owners who want to build real apps and get back to running their businesses. They shouldn't need to babysit an AI through twenty rounds of revisions.

If you're feeling the AI fatigue too, maybe it's worth trying a tool that was purpose-built for what you're actually trying to do. Check out DontCode and see what it feels like when the AI knows what it's doing from the start.