Warp, Beyond Cursor: Trusting the Machine

A note on AI-assisted programming

Since I started working on Anantys, an investment tracker platform, I’ve been rediscovering my craft as a programmer. I’ve already talked about it a bit here. Today, nearly nine months after the project began, I hardly write any code “by hand.” I generate it with Cursor. That doesn’t mean I don’t code—quite the opposite. It’s easy to take the shortcut and cry blasphemy, but that misses the point. Code generation (production) is only one part of the job. What truly matters for long-term project durability—handling load, security, good design, and the ability to evolve—is not the code itself, but how we conceive it, organize it, and “see” it. This is exactly where my expertise has shifted, my role as a “developer,” because the machine can’t do that—due to its amnesia and its inability to take real distance.

Because precisely: that big-picture sense, that taste, is the mark of an experienced developer. It’s what ultimately makes the difference between a weekend project that won’t survive its fourth feature and a multi-year codebase with hundreds of thousands (millions) of lines running in production.

Cursor

Cursor has been my development tool for months. And by the way—is it still a tool? There’s debate. Jensen Huang, Nvidia’s CEO, says no: AI isn’t a tool but “work,” because AI itself uses.

A philosophical debate that serves his point.

As for me, I’ll stay on the “tool” side. Because I, a human (with consciousness!), use AI for my own purpose. So yes—it is my tool. But I digress. Cursor, then. A tool (as I said) that multiplies the firepower of a motivated, cheerful developer. Those who still say AI is a mirage and that productivity gains remain to be proven should look more closely. That massive impact, that phenomenal (inhuman) change of scale that happens inside a development team that embraces this way of working is truly a revolution.

We move faster—much, much faster—but also, we get more done. The barrier between idea and implementation is almost gone. I like Cursor because I offload all the heavy work to it: I brief it, it codes, and I review. Important: I review all its code, all the time. I understand it, I amend it, I reject it, etc.

It’s that control that keeps me confident. It’s fast and efficient, but I keep my hands on the wheel.

Warp: Trust me.

Warp proposes something different.

My cofounder brought up Warp. “You should try it, seriously. You don’t see the code by default, and the result is impressive.”

Perplexed at first—held back by the (universal?) urge not to change habits—I dragged my feet a bit. Then, this morning, I jumped in. I had an annoying UI bug in our Next.js app (the future Anantys product we’re aiming for in Q1 2026). I figured it would be a good test.

Disorienting at first (the machine takes care of everything, the UI no longer shows the code—a bold, deliberate choice). Impressive, yes: the work produced is good. The code is clean and well organized. The result is fully comparable to what you’d expect from an experienced developer assigned that ticket.

By the way, you can literally talk to Warp, by clicking the microphone icon. You talk to it. It goes off to work, and you come back to read its code and approve. It’s slower, but the result is very good.

I was even surprised: while fixing a bug, it literally stepped out of the frame and proposed improvements that weren’t in the original request. Powerful: that’s usually where LLMs fall short—they stay trapped inside our prompt bubble. Here, Warp took a step aside: “your Momentum average—it might be interesting to make it a value‑weighted average by asset allocation. Want me to add that?” Wow.

Adopted?

I don’t plan to replace Cursor. But I now have another tool in my toolkit. For bug-fix sessions, Warp is probably better. Maybe (likely) for bootstrapping features too. I’ll need to try.