Emil Atanasov blog

AI Agents are new developers

AI is the new modern! Everything with AI is cool. (Well, at least catches your attention.) Too many tools, too much information. If everything wasn’t fast enough now it’s at least x10 times faster. The volume of code that’s generated daily is greater than before. Developer (and wannabe ones) are using AI tools to build services, apps, sites and other software. No one is talking about quality, everyone is just trying to invent something new, that can attract more and more people.

There are too many services based on LLM (Large Language Models). Here are a few - ChatGPT (by OpenAI), Claude (by Anthropic), Gemini (by Google), LLama (by Facebook/Meta) and many others. I’m using Copilot (by Github/Microsoft) to speed up my development. Recently, the “agent mode” has changed the way I work daily.

Before that I would say the AI (or LLMs) were really helpful, but I would compare them with a nice tailored tutorials, which you have to manually follow step by step. This is not the case with the agents. They are “smart” and “autonomous”. In short, they do few steps before they succeed/fail. This saves your time (developers time).

The quality of the code varies a lot, based on the language and how much “XP” a LLM has with it.

I haven’t used MCP in my daily work. If I find a nice way to do it, I’ll share it.

If you haven’t tried the agents, then you have to check them. This is probably how the development will look in few months. Most senior developers won’t write much code. They will read and validate everything, that has been generated by some agents.