I stopped writing code. Sort of.

For months, I watched the agentic AI wave from the shore. I’d been using generative LLMs daily: ChatGPT, Claude, the usual suspects, for everything from drafting docs to rubber-ducking tricky problems. But the agentic side of things? Letting an AI actually write code, run commands, and open pull requests? That felt different. That felt like it was coming for something I cared about.

I kept hearing about Claude Code. Simon Willison was talking about prompting Claude from his phone to create PRs on his open-source projects. People were shipping features without opening an editor. Part of me was fascinated. A bigger part of me wasn’t sure what this meant for my identity as a software developer.

I’d spent years building that identity. Learning how the sausage is made. Understanding parsers and ASTs and control flow. And now someone could describe what they wanted in plain English and get working code back? It stung a little, if I’m being honest.

Separating signal from noise

The AI world moves fast enough that it’s almost impossible to distinguish hype from reality in real time. For a long time, I filtered out most of the noise. CEOs had been predicting the end of software engineering for over a year and I was skeptical. Most of the loudest voices had something to sell: a product, a narrative, engagement.

But then the people I actually respected started shifting in tone. Kent Beck, the creator of TDD and someone who has been programming for over five decades, was talking about being re-energized by AI agents. He was calling them “unpredictable genies” but still coding with them every day. Martin Fowler was on the Pragmatic Engineer podcast talking about how AI changes software architecture. Gergely Orosz was dedicating entire issues to how agentic coding would reshape the industry. Addy Osmani had written an entire book about it, Beyond Vibe Coding, framing AI not as a replacement but as a multiplier.

These weren’t hype merchants. These were people who had shaped how I think about software. When they started paying serious attention, I figured it was time to actually try it.

The dive

The timing worked out in a way I didn’t plan. My dive into Claude Code coincided with joining Instacart, a company that genuinely surprised me with how far ahead it was on agentic AI. This wasn’t lip service. Within my first few weeks, I noticed most engineers had already moved away from Cursor. My onboarding buddy was direct about it: ditch Cursor and go all-in on Claude Code. That matched what I’d been seeing from the people I follow. The tide had already turned, and Instacart was already riding it.

So I dove in.

Going back to the terminal was a strange feeling. It reminded me of what programming must have looked like before my generation got comfortable with GUIs. The hardcore coders of the 90s. There was something oddly satisfying about it.

What I didn’t expect

The first real win was exploration. That probably sounds underwhelming until you understand what I was exploring: a deeply complex, domain-driven Ruby codebase with dozens of active contributors and years of history behind it. I’ve joined new codebases before. This was different. I was able to orient myself faster than I ever had, not just skimming files but actually understanding how pieces fit together.

It cut my onboarding time significantly. Simple Jira tickets my mentor handed me early on could be handled in a single prompt. That part was almost jarring.

What I didn’t expect was where the difficulty actually showed up. It wasn’t the code. It was making sure I understood what Claude had done so I could genuinely own the PR I was putting out. That’s a different kind of challenge than I was used to. The bottleneck shifted from writing to understanding and directing.

The identity question, revisited

The fear I’d had was rooted in a narrow definition of what it means to be a developer. I’d been equating my value with my ability to type code, remember syntax, hold a mental model of a codebase in my head. Some of that is less important now.

But the things that actually matter: understanding what to build, knowing when something is wrong, designing systems that hold up under real-world pressure, asking the right questions. None of that went away. If anything, it got more important. Kent Beck put it well: the frontier of AI-assisted development isn’t about maximizing code production. It’s about maximizing human learning.

Within two months, I went from copy-pasting between ChatGPT and GitHub Copilot and my IDE, inefficiently and reactively, to functioning more like an orchestrator. I’m not just writing code anymore. I’m directing it. That shift changed how I think about what a software engineer actually is.

What I’ve landed on

My workflow looks nothing like what it did before. The terminal is my orchestration layer now. I run tmux with nested sessions: each outer pane is a project, and inside each one I have Claude Code running in one pane and everything else in another. lazygit for version control. Vim for when I need to read or edit something quickly without context-switching. The IDE is still there, but I’ll sometimes go an entire day without opening it. MCP-connected to everything I can.

Addy Osmani wrote about this shift in Death of the IDE? The argument is that the center of developer work is moving away from the editor and toward supervising agents. The IDE becomes one instrument underneath a control plane, not the control plane itself. That matches what I’ve been living.

But I’m also skeptical about whether this is the right setup indefinitely. I’ve studied enough attentional psychology and read enough Cal Newport to know that humans are genuinely bad at multitasking, even when we think we’re not. Orchestration fatigue is real. There’s a sweet spot for how many sessions you can actually manage in parallel, and that number isn’t infinity. I haven’t found mine yet. That’s still TBD.

The transition wasn’t smooth. There were days where I felt like I was losing something. More days where I felt like I’d unlocked a new gear. But the honest answer to “where I’ve landed” is that I’m still landing.