The Uneasy Tradeoff of Agentic AI Tools

This is 2025, the year of AI, and more specifically, agentic AI. Over the past year, coding tools that don’t just autocomplete snippets but actively write, refactor, and even manage codebases have generated both excitement and skepticism. On paper, the promise is compelling: an AI teammate that can take care of boilerplate, track down bugs, and even propose architectural changes. Yet despite the hype, many seasoned software engineers remain hesitant.

Living in San Francisco, I’m surrounded by software engineers. I’ve asked many friends and colleagues what they think about these tools. One of them, ironically, an AI engineer, gave me an answer that I think captures why many experienced engineers are still reluctant. He said:

“The problem with these tools is that they take away the fun part of engineering, solving problems, designing, and architecting patterns, and they leave us with the annoying part: reading through inconsistent, verbose, and often buggy code. Instead of writing code we are proud of, we end up spending more time fixing code we do not love.”

What struck me about his answer is that he did not mention the usual suspects e.g. hallucinations, data limitations, or lack of context. His concern was not even technical. It was emotional. For him, and for many of us, engineering is not just about output, it is about the craft and the joy of creating something elegant. When AI takes away that part, what is left can feel more like cleanup duty than creative work.

I personally believe these tools will get exponentially better over the next five years. They will write cleaner code, hallucinate less, and most of the current technical issues with foundation models will be addressed. Will they replace software engineers completely? I doubt it. Will they impact the demand for software engineers? Very likely. What I am more certain about is that they will not erase the human side of engineering. The creativity, intuition, and finesse that great engineers bring to their craft will continue to matter, and the best engineers will still choose to write the most creative code by hand.

I plan to revisit this topic in a year or two to see where I was wrong and where I was right, and I will write a follow-up post to reflect on how far agentic AI has come.