a claude code review (from building this site with it)
i didn't read about claude code to write this. i built and shipped this entire site with it, from an empty repo.
so this is the review from the other side of that - what it actually does when you point it at real work, not what the demo promises.
tl;drclaude code is the closest thing yet to a tireless junior engineer that ships. it's excellent at bounded, well-specified work and at the plumbing nobody enjoys, and it still needs you for architecture, taste, and the final call. used with a clear standard, it out-builds you on speed and stamina.
what it is
an ai coding agent that works across your whole project - it reads the repo, writes and edits code, runs commands and tests, and ships changes, in a plan-act-check loop. not autocomplete; an agent that does, then checks its own work.
what it's genuinely great at
- features from a clear spec. describe the behavior well and it builds, wires, and tests it.
- the boring necessary plumbing. config, sitemaps, feeds, headers, the stuff you forget until it bites.
- bugs and refactors. point it at a failing test or a mess and it works patiently through it.
- stamina. it doesn't get bored, tired, or sloppy on the tedious tenth pass.
most of shipping this site from an empty repo was exactly this.
where it still needs you
- architecture and big trade-offs. it'll build the wrong thing very well if you let it. the shape is your call.
- taste. it can place every pixel and not feel the page is off until you say so - then it fixes it in one pass.
- the standard. it holds your bar only as high as you set it. vague in, competent-but-not-yours out.
an agent that decides for you is a liability. one aimed at a clear standard and reviewed at the end is leverage. the full picture of that is in ai coding agents in 2026.
who it's for
builders who can specify and review - solo founders, indie devs, anyone who wants to ship more than their own hands allow. less useful if you want it to decide what to build for you; that part is still the human's job.
the verdict
it changed where the work happens. the bottleneck moved from typing to deciding, and that's a real gain. point it at a clear goal, check the output, and one person ships like a small team.
faq
is claude code worth it in 2026?
if you can write a clear spec and review output, yes - it ships real features and handles the tedious work fast. it's less useful if you expect it to decide what to build; direction and judgment stay with you.
can claude code build a whole website?
largely, with you directing - this site was built and shipped with it from an empty repo. it handles features, plumbing, and fixes; you own architecture, taste, and the final call.
does claude code replace developers?
no. it replaces the typing and the tedious parts, not the deciding. the work shifts to specifying, reviewing, and making the calls an agent can't - which is most of real engineering.
more: ai agent tools for solo builders and ai coding agents in 2026. more in the gear.