After a few weeks with Claude Code, it has completely replaced my default workflow. It’s really good, and weirdly addictive! The productivity gains are undeniable. You ship faster, explore more ideas, and iterate constantly. But once the initial magic wears off, you start to feel a trade-off: you are trading peace for pace.
Coding used to have a sense of calmness. Uninterrupted stretches where you’d just be in the zone. Shipping a feature felt like a milestone. Now, because execution is so easy, that emotional reward has moved. The joy shifts from writing code to getting the tool to do what you wanted, cleanly and autonomously. You need more volume to feel productive. There is less deep focus and more context switching. You orchestrate agents, shape prompts, and review output. You get more done, but the work is fragmented, reactive and parallel.
This new pace introduces a subtle pressure: if nothing is running in the background, it feels like wasted potential. Like, agents must always be running! (yes, I realize how absurd that sounds.) You see people on X running multiple agents at once, showcasing what they built and the FOMO is inescapable.
There’s an irony here. AI is supposed to free up time so we can do more with less, but in a culture that glorifies shipping, we just use the freed up time to build even more. It’s also incredibly easy to spend hours optimizing your setup. Refining workflows, writing new skills, installing the new trending plugins to make things just a little better. When the tool does the heavy lifting, you spend a lot of time sharpening the tool itself.
The shift is real though and the leverage is incredibly empowering. We have moved from writing (and reviewing) code to expressing intent. You can take on more ambitious projects and build bespoke solutions for every “I wish this existed” thought. Because the cost of building has dropped so drastically, software is becoming disposable. Permanence starts to matter less. Fulfillment comes less from the effort of execution and more from taste, judgment, and deciding what is actually worth building.
Agents are getting incredibly good at execution. They write clean code and make reasonable choices, often following the exact patterns I would. But as I readily accept their output and offload the thinking, I can’t help but wonder about the risk of my own coding skills atrophying.
Coding agents aren’t a novelty anymore. They are the new baseline expectation for software engineers. And it takes deliberate practice to steer them effectively.