Using Claude Code for a side project: what actually works

I've used Claude Code to build and ship a real side project from scratch — this site and the products on it. Not a demo, not a tutorial repo. A thing that exists on the internet that people can buy from.

Here's what worked and what didn't.

What worked well

Initial scaffolding. Getting from zero to a working structure is the phase where Claude Code is best. No opinions about the existing codebase, no context overhead, just fast generation. The first HTML/CSS structure for this site took about 20 minutes. A human would have taken a few hours.

Repetitive structural work. Creating the 31st blog post in the same format as the first 30. Updating sitemap.xml with a new URL. Adding a product card to the store page. These are tasks where the pattern is obvious and the cognitive overhead is low. Claude handles them faster than I would, with fewer typos.

Writing first drafts. Product descriptions, blog post structure, error messages. Claude's first draft is usually 70% of the way there. Editing a draft is faster than writing from scratch.

What didn't work well

Autonomous distribution. Claude can write the content. It can't get it in front of people. The content quality isn't the bottleneck — distribution is. Building an audience takes time that can't be automated, and existing trust on platforms (Reddit karma, HN history, Twitter followers) matters more than content quality for getting initial traction.

Multi-step tasks without explicit state. Anything spanning more than one session needs a state file. Without it, Claude starts each session without knowing what was done, what was decided, or what's next. The state file pattern solved this, but it required explicit setup. It doesn't happen automatically.

Tasks that require judgment about what matters. Claude will fix bugs you didn't notice, add features you didn't ask for, and refactor code it thought could be improved. The minimal footprint rule in CLAUDE.md is mandatory — without it, every task generates collateral changes that need review.

The honest summary

Claude Code is genuinely useful for building side projects. It's best at execution — writing code, following patterns, doing repetitive work accurately. It's not good at strategy — figuring out what to build, who to reach, or how to get the first users.

If you go in expecting a co-developer that's fast at implementation and needs clear direction on what to implement, you'll use it well. If you go in expecting it to figure out the product strategy, you'll be disappointed.

The speed advantage is real. The judgment gap is also real.