Being An Intern In The Age Of AI

There was an occasion early in my internship where I picked up a ticket that stretched beyond my current experience, and I let Claude rip. Code was generated. Specs passed. The MR went up. Then I found myself in a code review unable to defend a choice I'd made an hour earlier, because I hadn't made it. The model did.

It’s easy to get caught in this trap. The model hands you a convincing illusion of competence, and you don't notice until someone asks you to defend it.

But I'm glad it happened early on, because it quickly shifted how I use these tools. When I don't understand a problem, I lean on AI to help me learn it, not to write the solution for me. Once I actually understand the shape of the thing, I'm in a position to attempt it myself and let AI review.

That's my goal, anyway. There are times I really have to fight the urge: I'll catch myself reaching for the agent too quickly, skipping the thinking I should be doing myself. When I notice, I pull back. It's less of a rule I'm enforcing on myself and more a habit I'm building in the new ways of working. 

I joined Fullscript at the start of this year, and AI helped greatly with navigating the codebase. The rules baked into the repo guided the model on our conventions, so the answers I got back weren't generic; they were Fullscript-flavoured. That mattered most when I was asking for feedback, because the model would steer me toward Fullscript patterns rather than whatever a generic Rails or React tutorial recommends. For me, the greatest benefit was time saved, as I managed to pick up our patterns and conventions quickly.

The same is true for the questions that aren't Fullscript-specific. I rarely find myself leaving the editor for those. Before AI, any lookup questions would have gone to Google, then Stack Overflow, and finally to a teammate. Now they go to AI. What I bring to the team are the questions only they can answer. Who owns this domain? What's the intended behaviour here? Should this be behind a flipper? Asking AI is efficient, less distracting for the people around me, and frankly easier on the part of me that doesn't want to be the intern bombarding senior engineers all day. The tradeoff, though, is that some of the most valuable learning opportunities tend to come from conversations that start with a small question and drift into something bigger.

What I do think about is the desk-leaning question that can spark a five or ten minute conversation and ends somewhere I didn't expect. I enjoy those tangents. I find I usually come away with something useful: a piece of advice, a story, a way of thinking about a problem you hadn't considered. This is the kind of thing no model can offer. It's part of why I enjoy going into the office a few times a week; sitting around at lunch, listening to my peers talk about what they're working on, the problems they're solving, what they're building, and of course their opinions. That's the stuff I don't get from my editor.

There's a big part of the job that still feels like apprenticeship in the old sense. Reading other people's MRs and sitting in on design discussions. Watching how teammates react to a proposed change, what they push back on, what they ask a follow-up question about. You start to notice the small stuff, why one approach gets approved and the other gets three rounds of comments.

The role of software engineer is shifting, and interns are learning the craft right as it happens. The upside is that we get to learn these new tools alongside our peers. Fullscript runs a handful of events I've gotten a lot out of, including the AI workshops, where we pick up new workflows and ways to get more out of any model, and the bi-weekly AI office hours, where we share what's working, what isn't, and where we're stuck. What I appreciate about both is that they're built around the assumption that nobody has this fully figured out yet.

I think a lot about what it means to be early in a career right now. The tools are new, but the job underneath them isn't. You still have to understand what you're shipping. You still have to learn from the people around you. The shortcuts are tempting because they work, until they don't, and you're caught in a code review with nothing to say.