Skip to content
Back to posts

My AI Coding Assistant Quietly Became My Product Manager

Notes by SDG ·

Eighteen months ago, if you had told me I would be building multiple production apps without being a full-time software engineer, I would have smiled politely and moved on.

Today, my workflow looks completely different. It’s not just because AI can write code it’s because AI has quietly evolved into part engineer, part debugger, and, surprisingly, part product manager.

The Early Days: High Energy, High Friction

The first few months were exhausting. Every new feature felt like a gamble fix one bug, introduce two more. While the AI was helpful, it was also spectacularly confident when it was dead wrong. I spent hours chasing down solutions that sounded completely logical on paper but had zero connection to the actual problem.

I learned an important lesson pretty quickly: AI accelerates coding, but it doesn’t replace thinking.

Some of my biggest breakthroughs came simply from refusing to take the first answer as gospel. Sometimes the fourth explanation was the right one. Sometimes none of them were.

Learning to Work With the Tool

Over time, our dynamic changed. Instead of asking the AI to just "build a feature," I started treating it like another engineer on the team. I’d bring it in to review architecture, challenge my assumptions, hunt for edge cases, or explain why a bug might exist rather than jumping straight to a fix.

The conversations shifted from raw syntax to underlying logic. Ironically, the better I got at understanding software development, the less time I actually spent typing out code.

Production Over Perfection

One of my biggest mental hurdles was learning what not to care about. It’s way too easy to fall down rabbit holes obsessing over perfect architecture, premature optimization, or tweaking clean code that ultimately solves zero customer problems.

I had to become ruthless about one simple question: Does this actually move the product forward? If the answer was yes, we shipped it. If not, it waited. Building real software is rarely about writing elegant code; it’s about making thousands of practical trade-offs.

A Better Debugging Partner

My approach to troubleshooting evolved, too. When something breaks now, I almost never ask for a direct fix. Instead, I ask questions like:

  • "What are the five most likely causes for this?"
  • "What evidence supports each one?"
  • "What would you check first?"

That tiny shift changed everything. The AI stopped acting like a glorified autocomplete tool and started acting like a seasoned dev thinking through a problem. It still hallucinates sometimes, but it helps me eliminate wrong turns infinitely faster than I could on my own.

The Unexpected Product Manager

The biggest surprise came recently. After finishing a feature, the AI didnt just stop at reviewing the pull request it started critiquing the product itself.

It pointed out UX inconsistencies, suggested simplifying workflows, flagged confusing UI elements, and even proposed features I hadn't thought of. At first, I brushed a lot of it off. But once I actually tried a few suggestions, the product got noticeably better. Some of those ideas are live in production right now; others are sitting at the top of my backlog.

Some of the best product calls I’ve made this year weren't part of my original roadmap they just emerged from conversations that started with a snippet of code.

A New Role Entirely

When I started this journey, I thought my job was to write software. Now, I realize my job is something else entirely: understanding users, prioritizing ruthlessly, questioning assumptions, and connecting the dots across features.

The AI handles more of the heavy lifting implementation than ever, but the final judgment is always mine. I spend far less time worrying about how to build something and much more time deciding what deserves to be built in the first place”and honestly, that's made the whole process a lot more fun.

Tools are going to keep improving. Today’s s coding assistant will probably morph into tomorrow’s designer, QA engineer, and release manager. But the real edge won't belong to the people who use AI to spit out the most code. It'll belong to the people who know how to hold the best conversations with it.

Because at this point, building software feels less like traditional programming and a lot more like managing a very capable, very opinionated teammate.