I am a hands-on software engineer with more than 17 years of experience, and I no longer write code. Yet, I still deliver software on a daily basis. I fix bugs, do analysis, design systems, implement new features, and test my changes.
For the actual coding, I have become the guide for my agentic coding tools. My responsibility is to give them the necessary context to make the changes. I define the acceptance criteria using deterministic verification, and I review the evidence to check the completeness and accuracy of the results.
Guiding like a junior programmer, but forever
Some people say that LLMs are like junior programmers because you have to give them proper guidance about everything. This analogy is true up to a certain point. The main difference is that a human junior developer becomes an expert over time. They make better decisions after making several mistakes, they generate trust within the team, they develop pride in their work, and eventually, they mentor and coach other developers.
But LLMs are in a perpetual juniorhood. You need to make sure their agents always have access to the necessary context, and that this context is real and updated. The agent needs access to the right tools in a secure environment, and it might need your assistance at any minute during the task.
