The grunt work got faster. Learning the craft still takes reps.

Part 1 of a running series on what it’s like to join the workforce as an early-career professional, as so much of the routine work is getting streamlined.

The judgment new hires used to build on research, spreadsheets and first drafts hasn’t left the job. It has moved from doing the work by hand to steering it the whole way through. Working with AI, I decide what to ask for, check what comes back, dig where it’s thin, and decide what to keep. Each of those is a call, and each call is a rep. People who learned their trade before AI was a part of the job are right to ask where someone like me gets that practice. This is where.

The research that taught me this

The partners asked me to research social media for the firm’s new social plan. Which content types work, on which platforms, how often to post, with real examples pulled from the platforms and broken down in spreadsheets. By hand, I’d estimate that is a week of research or more. With Claude it took me a day or two. When Peter, one of the partners, sent over the brand template, Claude rebuilt my finished deck in it in five or ten minutes. The speed is the part everyone already knows about. The part that taught me something came after.

I had asked for every claim to come with a linked source, and a lot of what came back either had no source or had one that didn’t match. My first run leaned on sources that turned out to be outdated, so I ran it again. Then I hit a recommendation for how many times a week to post on X with nothing behind it. When I dug in, Claude hadn’t researched it at all. It had given me what it generated as the best practice, and no source anywhere backed it up. So I pointed the research at that one question, checked every source that came back, and built the recommendation from what held up.

I didn’t catch that because I already knew social media. I caught it because the claim had no source and I had asked for one. That habit is where it starts: nothing goes to my boss until I can see where it came from. Moments like that are a big part of how I learn.

The loop is the rep

That whole process is one rep. Some of those reps are small and specific. I learned early to ask for linked sources in the prompt itself. I learned later to have it check its own work, which I hadn’t thought to do. I learned to tell it to ask me questions first, so it gathers what it needs before it starts. None of that came from a manual. It came from doing the loop on real work, day after day.

I didn’t arrive fluent, either. Through most of college my professors were against AI, so I barely used it for school. When I interned at Dual Logic in the summer of 2025, I came in knowing very little and started out confused. I also expected intern busy work. Instead I was put on work the partners were going to act on, and I learned the tools by trial and error, every day, on that work. Two months as a part of Dual Logic, and the tools that confused me as an intern are how I do my job. In fencing, I didn’t get better by just watching others fence. I got better by trying an action, seeing whether it worked, and adjusting the next time. The loop at my desk works the same way.

The objection worth taking seriously

The strongest pushback is that these are reps with the tool, not with the craft. An editor who has never written can’t tell a good draft from one that only sounds good, and knowing what good looks like is exactly what the slow years used to teach.

Part of that is true. I bring in what school research and earlier work taught me, and that counts for a lot. But the social media project is my answer to the rest. Some of what I know about it now, the AI handed me. I checked all of it, and the parts it couldn’t support I had to find myself. I could use that knowledge tomorrow without Claude. Checking it is what made it mine.

Most of what people call taste lives in the last step of the loop, deciding what to keep and what to do yourself. You get better at that step the same way you get better at anything: by taking it many times, on work where the call matters.

What to do with a new hire

I’m two months in, so I don’t have this figured out. I can only speak to what’s helped me. If you’re training or leading someone in a seat like mine, here’s what has made a difference:

• Work where AI is actually useful. The social media research was the kind of task AI is built to speed up, so I learned it on work where it made sense. Being handed AI for a task that doesn’t need it mostly just leads to confusion.
• A rule about sources. Asking for a source on every claim is what caught the X recommendation. It’s a small habit that does a lot of work.
• Room to learn by doing. I figured out the tools by trial and error, every day, on work the partners were going to act on.

The people who learned before AI was part of the job were right that reps build judgment. From where I sit, they still do.

Lev