What Building With Claude Code Actually Takes

I've been building with Claude Code for a couple of weeks now. Three projects so far (RE Newsletter, OpenGate, and Legal Tale), and every one of them has been fascinating. The age of the one-person company has truly arrived.

Along the way I've started to notice which skills actually make someone a good builder with a tool like this. Claude Code is powerful enough to let you create real software without years of engineering training, but a handful of traits will make you dramatically more efficient and productive.

Think deeply

Claude Code is an eager, capable tool. Give it a one-line prompt like “I want to build X,” and it will go off and do it as best it can, making assumptions along the way and occasionally stopping to ask. Left to its own devices, it behaves like an eager-to-please employee racing to hand you a finished product.

But to produce a genuinely good codebase, and to build anything sophisticated, you have to think through the application yourself. You need to reason through every part of it, anticipate the edge cases and outliers, and decide in advance how they should be handled. You need to think in systems, and break a complex process down to its smallest pieces.

The single most important part of building something good is the planning. Not by asking Claude Code to plan for you, but by working with it to write a detailed CLAUDE.md.

Absorb a lot of information

Claude Code makes the coding easier, but it makes your mind work harder. It makes a lot of assumptions, and however good it is at reaching the goal, it will set things up in ways you didn't intend. The steps are all laid out for you, but you have to be able to read through a wall of code, design notes, and status output to actually understand how your application works. Skipping that is always a mistake.

A session with Claude Code is easy and hard at the same time. I find myself constantly on guard, reviewing large amounts of text to judge whether it's really building what I asked for. The ability to ingest and process a lot of information quickly isn't optional here. It's a requirement for working with the tool well.

Human creativity still matters

A common assumption is that AI spells the death of human creativity: that it can produce content in seconds where a person needs months, and that an ocean of “good enough” AI output will drown out the “amazing” work of humans.

Building with Claude Code has convinced me that's wrong. It is excellent at working within the parameters you give it, at following your instructions and building the thing you described. But how good the result is comes down entirely to you. It will make decisions on your behalf, and it often falls short of the genuinely good, efficient, optimized solution.

Take OpenGate Intel, a newsletter that surfaces the latest government procurement opportunities for Canadian businesses. When I described the idea, Claude Code's first instinct was to pull the list of past bidders and email them tender updates sorted by the government's own categories (IT & engineering, construction, logistics, and so on). My approach was the opposite: start from a specific tender, then go find the businesses that can actually deliver that good or service in that region. Better results, and fewer tokens to get there.

AI can make the small decisions. The strategic ones should still be made by people.

Learn to slow down

It's not hard to foresee a flood of new apps, software, and content pouring into the world. After two weeks with Claude Code, the feeling I keep coming back to is a strange mix of drained, overstimulated, and exhilarated, all at once.

The tool is exciting. The potential is a little scary. Keeping up with it is a grind.

Which is why the ability to step away and switch off might be the most underrated skill of this moment. Protecting your attention and your mental health, going for a walk, working out, being with people you like, isn't a break from the work. It's part of doing it well.