What blocks you build products in AI Era?

Context

As a software developer, when I browse the internet I see breakthroughs in AI for software happening every day. LLM providers and other companies and individuals keep launching inspiring products and features, trying to reshape the rules of the world. Software development has become a driving force that is already affecting the real world and the people connected to it.

I am a software developer too, maybe just a small one with limited influence. Still, I try my best to catch the AI wave and subscribe to AI tools and plans. They help me work more effectively, but when I try to build an entire product by myself with AI, I find it is not as easy as I thought. At work, when my intent is clear and I ask agents to do specific tasks, it often works. But when I am unsure what to do next or what the real requirements are, the agents do not give satisfying results. I have tried many times and still have not launched a product. It makes me sad when I see so many people sharing what they built on Twitter or GitHub. So I want to figure out what is blocking me from building products with powerful AI agents.

Advantages and limitations of AI agents

After Claude Opus 4.6 launched, it feels like AI can outperform 95% of software developers. If you try models from Anthropic or OpenAI in tools like Claude Code, Codex, or Cursor, you will find you can finish some development tasks in minutes instead of days. It feels like being on a rocket, when before you were just on foot. The speed difference between humans and AI is huge, and no one can ignore it. I used to enjoy coding, but now I barely write code line by line. Most IDEs are also shifting toward a chat-first experience.

So the advantages of AI agents:

• Much faster than humans
• Know a huge amount about software development
• Never get tired
• Never get bored with repetitive work

In a way, agents are like superheroes with superpowers. So why can’t I build anything I want with them? Maybe we cannot ignore their limitations.

The limitations of AI agents:

• Context window limits
• Lack of visual understanding
• Token usage limits

Agents have context windows, so their memory within a single chat is limited. If they do not have the right context, their output can become guesswork. Especially on the web, you cannot rely on prompts alone to produce good UI or UX. When working with agents, you cannot act only as a developer. You also need to think like a product manager and a UI designer. You have to clarify what problem the product solves and what style will work for the people you want to serve. The real blocks are often ourselves. Our limitations can prevent agents from building what we want.

In the AI era, people should reconsider their careers. If you only have one narrow skill, it may be replaced by AI very soon.

Builders in AI Era

In the AI era, I think there will no longer be developers, product managers, UI designers, or test engineers. Instead, these roles will merge into one: builders. Builders set constraints and ask AI to clarify requirements, design prototypes, implement code, and pass tests. They then review the results and iterate with agents in the loop. Software engineering will become a true industry rather than manual manufacturing industry.

How to become a builder in AI Era?

I think the most useful approach is to keep trying and to turn what you learn into your own skills. Agents can do a lot, but if you do not use them, they stop helping. Learn from other people’s tips and methods, but do not just read or watch them. Practice them yourself. Practice is the best way to learn.