Love of coding vs love of building (LLMs + agents edition)

I’ve always thought there are two kinds of joy in software.

One is the joy of coding: the craft, the clean abstraction, the quiet satisfaction of making something right. The other is the joy of building: shipping a real thing that exists outside your editor, even if it’s imperfect, as long as it’s moving forward.

Most of my career I could pretend those were just preferences. Lately, with LLMs and agents, it feels more like a fork in the road. Not because coding stopped mattering, but because building just became dramatically cheaper.

The tradeoff we used to accept

If you’ve shipped anything serious-especially as a dev or a founder-you know the deal we’ve all been making for years. You can move fast, or you can do it properly. You can ship this week, or you can build the foundation that won’t punish you next month.

And the annoying part is that “properly” rarely loses because we don’t care. It loses because we run out of time, energy, and attention. So the boring-but-important work gets pushed out of scope: documentation, analytics, cleanup, tests, automation, deployment polish, all the little decisions that make a project feel coherent. You promise you’ll come back to it later, and later turns into “never” more often than anyone wants to admit.

That’s the real cost of the old tradeoff: not just technical debt, but a steady stream of good ideas that never get a chance to exist.

What changed with LLMs + agents

The shift isn’t that an LLM can write code. That part is impressive, but it’s not the thing that changes your behavior day-to-day. The real change is what happens when you combine LLMs with agents and give them a workflow: context in, outcomes out.

When I say “agents,” I mean something more practical than a chat window that answers questions. I mean a system that can read the repo, propose a plan, make coordinated edits, run scripts, commit changes, deploy, and then iterate with you until it feels right. The model is important, but the loop is the product.

This is where “plan mode” becomes a superpower. It’s not about generating output faster. It’s about having a cheap, repeatable process that takes an idea from vague to real without requiring you to be in hero mode the whole time. You’re not outsourcing taste. You’re outsourcing the grind.

And once the grind gets cheaper, something weird happens: the old quality-versus-speed tradeoff starts to loosen. Not because the first draft is perfect-it won’t be-but because iteration is. You can afford to do a second pass. You can afford to make it cleaner. You can afford to fix the part you’d normally ignore because it’s “not worth it.” Suddenly, you can ship fast and keep standards, simply because the cost of getting from “works” to “good” collapsed.

That’s why I think builders are going to love this era. The tools reward momentum and feedback. They make the “nice to have” things feel realistic from day one.

The builder mindset I’m trying to practice

I don’t think this is about choosing sides and declaring that coding is inferior. Coding is still a craft, and craft still matters. But if you’re coding-first-if your identity is deeply tied to being the person who writes the perfect solution-this shift can feel uncomfortable. Agents will write things you wouldn’t write. They’ll take shortcuts you wouldn’t take. They won’t “feel” the design in their hands the way you do.

If you treat that mismatch as an insult, you’ll fight the tools instead of using them.

The way out, for me, is to relocate where I spend my standards. The goal isn’t to love every line of generated code. The goal is to build a system that reliably produces outcomes that meet your bar, with your taste guiding the shape of the work.

So my current rules are simple:

I try to be explicit about constraints and direction instead of hoping the first draft lands. I use plan mode on purpose-approach first, execution second-because it turns quality into a normal step rather than an afterthought. I judge the workflow, not the initial output, because the real win is a loop that keeps shipping without requiring perfect conditions. And I save my pickiness for the last 10%, where it actually changes the feel of the thing.

If you love coding, keep loving it. The craft isn’t going anywhere.

But if you love building-if you care more about the thing existing than the purity of how it was created-LLMs and agents are going to feel less like a tool and more like a permanent advantage.