How to Use AI and Not Become Dependent On It
Quick Points
- AI can be an amazing teacher if you let it
- Ask it for explanations and clarifications instead of just “build this”
- Ask it to compare and contrast different approaches so you can learn other ways to solve problems
- Don’t let AI’s existence blunt your problem solving skills
Outline
- Dependence on tools without understanding the underlying tech or techniques is a liability
- Applies to new engineers as well as senior
- AI can be your best teacher if you let it.
- Ask for different approaches
- Ask for compiler/syntax error explanations
- Don’t be a passenger in your own career
Dependence on tools without understanding the underlying tech or technique is a liability.
Am I saying you should always do things the hard way? Not at all! Improvements in tooling and techniques is what allows us to advance and make even more advanced products. But I do believe a basic understanding of the underlying tech or technique is important. A favorite example of mine from being a Kubernetes administrator is the program Lens. It’s a great tool that makes managing clusters easier, faster, and in many ways, more fun. But I purposefully haven’t used it much because I wanted to solidify my grasp on kubectl, it’s commands, and it’s syntax. Does that arguably slow me down? Maybe, but I also know that, if i’m ever in a position where the only tool I have is kubectl, i’d be able to manage just fine. And therefore, if I do use lens, I know the things it should be capable of since it uses kubectl under the hood. This idea applies to both new and senior engineers.
AI can be your best teacher if you let it
Let’s be clear, A.I is phenomenal, you describe what you want and it will build it for you. This in and of itself is a great exercise because, in order for it to build exactly what you want, you need to describe it well and give it specific rules. That means you have to describe everything, happy path, edge cases, what to do with conflicts, etc. This is a great exercise because, lets face it, lots of engineers struggle with thinking through these things and so being forced to do so is amazing. But you can take it one step further. As addicting as agent mode is, I challenge you to do this. Put your A.I tool of choice in ask mode and change your prompts from “write xxx” or “implement xxx” to “how would you suggest I do xxx”. Ask it questions, ask it follow ups. Even a simple question like “why did you pick this approach” can be very insightful. Other good questions are:
- Will this scale well with lots of requests?
- Does this use lots of system resources compared to other approaches?
- Does this result in multiple round trips to the databases?
If you don’t know what to ask as a follow up, how about “I’m trying to understand this code. What are some questions I can ask you to understand it better?” This puts you back in the driver seat instead of just a passenger on your own career journey.
Ask for different approaches
Another approach is to ask “what are 3 ways you would solve xxx and give me the pros and cons of each”. This again, puts you in the driver seat and lets you learn. This is also amazing if you don’t know enough about your current project to even know what questions to ask. Let the pros and cons trigger your curiosity and begin learning again. Once you pick one, rinse and repeat with anything you want to know more about in your project.
Ask it to explain compiler errors or syntax errors.
I know I cant be the only one who has gone around in circles with an A.I agent about fixing an error just for it to give me a “fix” that changes the error but doesn’t work. Instead, ask the agent to explain the error and why it is happening. Ask it for how it would fix it and what this change does. This builds your knowledge of syntax and approaches while you continue to build your project. It also means that, the next time your agent is “certain” of it’s fix and it ends in an error, you can manually intervene, fix it yourself, and move on.
Don’t be a passenger in your own career
With all the hype, it’s easy to feel like you are falling behind if you try to learn these things and aren’t just churning out another $40k/month application. I can’t promise you this knowledge will pay off. But I can almost promise you it will. Murphy remains alive and well in this age of AI agents and fast-moving tech. There will be a time when you need to solve a production issue and your LLM of choice is down, your agent is going in circles, hallucinating, or just straight lying to you. It’s those times that having this knowledge and slowing down just a little will make all the difference in the world.