AI and Imposter Syndrome: A Paradox
I’m far from the only person who has felt the dreaded Imposter Syndrome in their career. In high school and college, I studied, took hard classes, and got pretty good grades. Did pretty well in coding interviews.
But then, I got my first job out of college. Suddenly it didn’t matter that I knew how to write an algorithm to optimally schedule tasks or that I remembered the finer details of Dijkstra’s Algorithm. Instead, in my first job, I found myself assigned the project of coding up a whole new webapp with websockets using Ruby on Rails. I had to do it in one month. I definitely panicked. I remembered learning a little bit of Ruby for my programming languages class in college, and I thought I had read a blog post somewhere about websockets, but I didn’t know much else.
I found myself struggling. Other people I worked with had been coding seemingly their entire lives. Lots of them participated in Hackathons regularly, they seemingly had the stamina to code until the wee hours of the night, and none of them had to look up the Ruby on Rails documentation as frequently as I did.
And although I struggled, I was learning, and I did start to get better. After my first company, I joined another that used Python and Django instead of Ruby on Rails, and I was able to generalize the concepts I learned in my previous job to pick up my new company’s tech stack pretty quickly. And since then it’s only gotten easier - time and practice had definitely made these things feel like second nature, so I didn’t feel like a fish out of water as often.
Until AI coding hit the scene.
For years, my history as a software engineer taught me that the harder the problem, the higher the value. Struggling through problems was considered good, because we were doing hard work. A few hours optimizing a db query? Great. Rabbitholing down AWS documentation? A good use of time most certainly.
But I remember the first time Claude was able to spot a database performance problem before I did: I felt that Imposter Syndrome coming back. Was I even the author anymore? How did it catch bugs before I did? I submitted a bugfix, but felt more like this was Claude’s success, not mine. And I wasn’t just comparing myself to engineers who had coded their entire lives, I was comparing myself to a robot that crawled the entire internet. But eventually, something changed. The AI bot that first ruined confidence as a coder became the thing that helped rebuild my confidence as a software engineer.
How I learned to stop worrying and love the AI
After the initial shock, confusion, and uncertainty of how to attribute a developer’s coding success in the era of AI, I’ve come to embrace the confidence AI coding has given me in surprising and unexpected ways.
No more stupid questions
In my previous post, I lamented the ways AI creates knowledge silos. But on the flip side, there’s no more stupid questions anymore, no more fear of feeling like we’re wasting someone else’s time on a bad question. I’m making a concerted effort to not silo knowledge further, because I know that usually if one person has a question about something, they’re probably not the only one with that question.
But AI has helped me become more confident in how to frame questions, how to search for documentation, and when to identify when a piece of knowledge is likely old and outdated. I no longer have this underlying fear that my question is bad, poorly-researched, or confusingly worded. I don’t have a nagging voice in the back of my mind telling me I’m disrespectful of a more senior engineer’s time.
Practice time with the bots
I used to have so much anxiety anytime I’d have to present an RFC, a big code change, or a demo to the company. Too often I’d run through presentations in my head and freeze up at all the possible ways they could go badly. Now, I rehearse out loud. I have Claude poke holes in presentations before anyone else can.
For me now, AI is a low-stakes way to hash out presentation issues before I do the real thing.
AI has blindspots that I can still correct
In my previous post, I wrote about how “taste” is the big new topic in how we think about differentiating ourselves from the robots. The reality is, often Claude Code does generate things I wouldn’t have, often it makes tests overly complex or misses some edge cases. I know I still have taste and a point-of-view, and that makes me feel confident that I can continually correct and improve Claude as I go. And as a bonus side effect, it’s helped me tighten up how I think about giving feedback to others in code reviews.
Becoming a learner again
Maybe the best part of the new era of AI is that it’s given me the chance to learn again. I’ve gotten to learn and experiment with new workflows on my own terms. I’ve used Claude to teach myself about Computer Vision algorithms, and I’ve asked Claude lots of questions along the way. It’s been a while since I’ve had a totally open sandbox with a new technology and seemingly limitless potential of ways I can take it to learn and grow.
All that to say, in this era, I hope we come out of using AI coding tools feeling more confident, not more siloed. For my Imposter Syndrome, it’s helped me practice, rehearse, and refine so I’m more confident in what I’m doing.