How Do You Use Gen AI?
Here's my approach to using, and discussing my use, of these tools
I’m seeing interviewees being asked about how they use AI tools as Software Engineers in nearly every interview, and nobody seems to know how to approach the conversation. My take is that it’s important to go beyond the mere tactical use of AI tools by creating and explaining a systematic framework.
Here’s how I leverage AI in my day-to-day as a software engineer for coding, communication, and document writing.
But first, before I dive in, I want to mention that I struggled with how to frame this piece, because I think there are three different ways to use this information:
How to talk about your use of AI in an interview
How to actually use AI better
How to leverage AI as a job seeker
Might be helpful to keep those perspectives in mind as you read through this post. I wrote this as if an interviewer were asking me how I use Gen AI in my day-to-day, but it could also be useful if you’re struggling with incorporating these tools into your workflow as well.
My Approach To Leveraging AI Tools
1. Use AI for manual tasks, not creative ones.
This is the core of my personal philosophy- Gen AI is Not Creative. It is your baby-brained intern that’s really good at performing manual tasks when given a detailed description of how to do so. I theorize that the most successful users of agentic generative AI are the same employees who are best at delegating work by writing well-formed tasks in their ticketing systems.
The takeaway is that you should find opportunities to automate manual work using Gen AI to give yourself more time to focus on creative aspects of your projects. Design your own data model and have Claude write the SQL or ORM code to generate the actual implementation. Write your own API contract and have an agent implement it for you. Describe in detail each stage of a data pipeline, and let AI implement the tedious parsing code. Just make sure you know how to verify the outcome- just as you would with an intern.
2. Conversations over queries.
When you’re working with Agentic AI, remember that it only has the context you give it. Just like with a newbie engineer, it’s usually helpful to find out what they don’t know. I like to add the following to the end of my prompts, turning the response from a randomly selected reply to a conversation that helps the agent zero in on what I actually want it to do:
Before you get started, what questions can I answer to help you understand how to approach this task?
3. Never Copy Paste. We all know to never copy-paste code because if you had a bug in the code, now you have two bugs. I’ll take that tenet one step further- if you copy-paste code, you miss out on the opportunity to improve it. I apply this to my use of AI as well. I avoid copying agentic responses directly into my workspace, whether it’s code, comms or documentation. By retyping the outputs I accomplish the following:
I catch bugs. We all know that AI is unreliable- just like the stack overflow posts and reddit content it was trained on. It’s better to catch that before code review.
I keep my voice. I can rewrite the code to be in my style, make sure the email has my charmingly groan inducing wit, and keep things consistent across the entire product.
I understand everything that goes into my code or document. This is mostly for code, but sometimes there will be creative coding, or uses of libraries, that I’m not familiar with. I don’t want to try and extricate meaning from a complex nested list comprehension when trying to hunt down a bug months after it was written.
5. Ask For Options.
Here’s a pretty simple one. The first time I ever really used Gen AI was to edit my resume when I realized my job sucked. I realized that ChatGPT was really effective if you just treated it like an over-eager thesaurus. When I noticed that I had used the same word three times across my bulletpoints, I fed each one into the agent and asked it to “**Rewrite this sentence in 10 different ways**”. The best part was that, as this was earlier days, that only counted as a single query and helped me get more out of the tool’s free tier.
After it gave me a handful of different wordings, I was able to read through them and determine how I wanted to write my resume content. Now I use this approach the majority of the time I work with Agentic tools in every setting.
5. Teach It How To Teach You.
If you want to use AI to learn how to complete a task, give it an example of what you already know. Claude has the concept of a **Project**, where you can give it a set of context to always refer to. For example, if you’re learning a new framework:
“I am building a web application in nest.js. I have built similar applications in Django, Flask, and Ruby On Rails. When I ask about concepts in this new framework, explain them using parallels from my experience.”
One More Thing
As for how you discuss using AI in an interview, they want more than just the tactical response of your use cases. They want a framework that describes how you would approach a new tool or use case.
This means putting time and experimentation into it- try things that feel uncomfortable just to see what it’s actually like. You don’t get brownie points for adopting the opinions of others as your own without trying it yourself, and to be frank, I wouldn’t want to hire anyone who isn’t open to new things.
My Guiding Philosophy In Life: I’ll try anything twice.
What do you think? Do you have your own approaches that are helpful? Did any of these work for you? Let me know in the comments- bonus points if there aren’t any emdashes or correctio in your response.