How Agentic AI Changed the Way I Work
My honest take on building software with AI tooling after years of rolling my eyes at autocomplete.
The way I get work done as a product engineer looks a lot different these days than it did two years ago. I think most of my peers would say the same.
I’ll be honest, I initially wrote off some of the early AI-assisted tooling like GitHub’s Copilot. They felt gimmicky to me and slowed me down more than they helped.
But by late 2024, Cursor was beginning to surface in my feed, becoming too popular to ignore. I decided to try it out and immediately saw the value in its ability to anticipate my intentions, autocompleting the exact code I would have written by hand. It was an obvious unlock moment. Cursor quickly became my main IDE, convincing me to ditch my beloved VS Code that had been my daily driver for years.
And then, sometime around the middle of 2025 I tried Claude Code. The magic of describing a bug or a feature in plain English and then watching Claude write the code in real time was intoxicating. The more I used it, the more I found myself wanting to use it. Agentic AI had arrived and it was obvious that it was the future.
I’ve been fortunate to work for a company whose engineering leadership has fully embraced agentic engineering. Many on the engineering team had already adopted these new tools, but at the beginning of 2026, they began a company-wide initiative to explore and learn how to use AI. They’ve provided our team access to Claude, ChatGPT, Cursor and pretty much any other AI tool we wanted to try.
The pace at which these tools have changed the way I work is simply incredible. At this point, I typically spend my days bouncing between Claude Code and Codex/ChatGPT for work that needs my attention or judgment to steer to completion.
But for other tasks our team built a cloud-based agent that we typically interact with in Slack. Need to triage a support request? Ping the bot and it will investigate, document, and even open a PR to fix any issues it finds. Have a well defined Linear issue? Just ping the bot in Slack and tell it to get started. A short time later, it will respond with a PR for review. 🤯
“Clicky”, as we call it, does a whole lot more, too. Some of my teammates have written more about our software factory here if you want to learn more.
The rate of change we’re going through as an industry is unlike anything I’ve ever experienced and it’s hard to keep up at times. But I’m so hopeful and optimistic for the future.
The unlock isn’t that AI writes code. It’s that I can spend more of my day on the parts that actually need a human: taste, tradeoffs, and knowing when something’s wrong.
