The two key skills junior engineers need for effective AI coding

It has never been easier or faster for a junior engineer to get help with their work than at any time in the history of software engineering. LLMs have put the insights from massive collections of codebases at the fingertips of software engineers.

A wealth of LLMs have sprung up to help engineers in their work but how should a junior engineer approach the use of these powerful tools?

There are two things to bear in mind when using LLMs to support you in your coding:

• The quality of the answers you get is directly related to the quality of the prompt you use. If you don't accurately define what you are trying to achieve you may get a misleading response.
• LLMs can be confidently wrong. The answers they give may seem convincing but could be completely wrong or have a bug in it.

So, how can we avoid these pitfalls whilst being productive?

Asking the right question

If you don't specify the question well then you can get a confidently wrong answer which could mislead you. It can be difficult to articulate the problem or question as a junior because you may have a limited understand of the language, tools and libraries you are using.

When you prompt an LLM, provide as much context as you can. Simple questions can easily give misleading results. Mention the language, libraries and frameworks you are using. Explain what your code is supposed to be doing. Be specific about the expected inputs and outputs. As you gain more experience with the language, tools, frameworks and libraries that you use, the easier this will become. You will have a better grasp of what is going on with your code and will be able to more tightly describe the issue or task to an LLM.

Understanding what you get back

I have often seen engineers copy-paste a solution from the Internet that they do not understand. In the past this might been a line or two from a forum post or a Stack Overflow answer. With an LLM, this has the potential to be a whole module of copy-pasted code.

Engineers adding chunks of poorly understood, AI-generated code to a codebase is the biggest risk that software teams have when adopting AI in their workflow. Poorly understood code is difficult to maintain. It can be difficult for an engineer to understand code that they have written themself just 3 months after they wrote it. Imagine how much more difficult that would be if they didn't write it in the first place.

It is important to understand what the LLM is recommending you. Don't just blindly use it without taking the time to sense check it. The code could turn out to be wrong or there is a bug in it. You could then waste time trying to debug it and the LLM then becomes a hindrance rather than a help.

Cross reference the results with official documentation to sense check what the LLM is recommending. Sometimes it's answers may refer to an older version of a framework or library or it could use a function that does not exist. The documentation for a language or library is the authoritative resource on what functionality exists and how it works.

If you are confused by an answer, then don't just blindly copy paste. Apply the 30 minute rule. This would be a good point to ask for help from a colleague.

Key skills for using AI

There are then two key skills for using AI to support you in your software development work:

1. Prompting effectively
2. Code comprehension

Prompting effectively

Try to give as much context as you can to get the best results. The more knowledgable you become in the languages, tools and libraries you use, the better you will become at explaining problems and questions to an LLM.

Code comprehension

Code comprehension is key in helping you understand the code that an LLM generates. The better you get at this the quicker you can sense-check / verify answers that the LLM gives you. See my article on this for advice.

Happy coding!

Photo by Lisheng Chang on Unsplash