Start with a problem, not a tool. Build something small you’d actually use, with whichever agent you’ll actually open tomorrow, Claude Code, Cursor or anything else, on the most basic deployment setup. Then test it with every person who could tell you whether the idea is worth anything.
When something breaks, ask why before you ask the agent to fix it. The only difference you have from expert programmers when they hit bugs is that they just know how to ask better questions and recognize how it should work by examining the parts that were built. You can develop that sense of what is broken with practice. You learn faster with every feedback loop you complete while you do it.
You don’t need to question yourself if you know enough to solve the problem. A reliable agent will keep hammering the details until it just works. Every feedback loop is becoming faster and cheaper. New models improve more rapidly than any group of humans could. The option left is to build with them and develop a new form of craft.
Previously, I’d get too invested in the input of various designs because it would be a lot of time and effort to throw things away and the work had a time limit to be shipped. Now I can explore any idea without the sunk cost.
Don’t spend your learning on comparing tools. Spend it on getting every feedback loop faster.
The real unlock isn’t the tools you use. It’s the mindset: you’re not learning to code. You’re learning to direct systems that code for you.
Also in January: How I Write Code in 2026 · What I Actually Built with Agents · A New Kind of Technical Career
Continues in February with The Death of Front-end Development.
Daniel Arevalo is Product Experience Lead at Omny Security. He builds AI-native tools and writes about design, product leadership, and what it means to direct systems that execute at a speed humans never had before.