Claude AI Review: What a Non-developer Actually Does With It
Claude, and what a non-developer actually does with it
I maintain a catalogue of several thousand products, run two websites, and build browser-based tools, working from a Chromebook with no development environment and no computer science background. Claude is the reason that arrangement is possible rather than absurd.
This is not a page about AI in general. It is an account of what one person with twenty-three years in health and safety training, and no formal technical training, gets out of it in a normal working week.
Why I use it
The change is not that writing got faster. It is that a category of work which used to require hiring someone became work I can do myself.
Backfilling a metafield across several hundred products used to be a scripting project, which meant either learning to script properly or paying a developer and waiting. Now it is a sentence, a query I read before it runs, and a check afterwards. The same is true of the interactive tools on this site: each one would have been a quote and a fortnight, and instead each was an afternoon.
What that removes is not effort. It removes the ceiling that technical skill used to place on what one person could operate, and after two decades of working around that ceiling it is the largest change to how I work that I have seen.
What I use it for
Running my catalogue through connectors
Connectors let it work directly against a live system rather than telling me what to type somewhere else. Mine reach my store and my SEO data, which turns a bulk operation from a scripting problem into a described intention with a query I can read first.
Building tools I could not otherwise build
The calculators and assessment tools on this site are single self-contained files, built in a browser with no local environment. I can specify precisely what they must do, because I know the domain. I could not have written them.
Arguing with a decision
The use I would least like to lose. Put a plan in and ask what is wrong with it. Working alone on a decision is how you talk yourself into things, and something that will push back on a weak argument is worth more than something that agrees with you fluently.
Drafting and rebuilding pages
Structure, first drafts and the tedious work of rebuilding old pages to a consistent standard. The judgement about what is true and what is worth saying stays mine, which is not a modest caveat but the whole job.
What I would flag
Everything below is from getting it wrong at some point, which is the only way I know any of it.
Fluent and wrong looks exactly like fluent and right
It will state something incorrect with the same confidence it states something correct, and there is no tell in the writing. Anything factual, legal or regulatory has to be checked against a real source before it goes anywhere near a customer. On a compliance page a confident invention is not an inconvenience, it is a liability.
Never let it touch production unsupervised
I run a control query before every bulk operation and read what is about to execute. I duplicate a theme before edits. Both rules exist because of a live incident, not from caution — an edit in the wrong file took down a working tool on a live site. The speed is real, and the speed is exactly what makes an unchecked mistake expensive.
It knows nothing about your business
Output quality tracks how much context you supply almost exactly. Vague question, generic answer. The people who find it disappointing are usually asking it things nobody could answer well without knowing the business, and the skill is largely in knowing what to include.
Left alone it writes like everything else
Unprompted output has a recognisable texture, and readers are getting quicker at spotting it. It is genuinely useful for structure and first drafts. Publishing what comes back without rewriting it in your own voice produces pages indistinguishable from everybody else's, which is the opposite of why you have a site.
Plans have limits, and the field moves quickly
There are usage limits on every tier, and a heavy session can reach them. Model names, capabilities and pricing also change often enough that I have deliberately not listed any here — check the current position rather than trusting a page that was accurate when it was written.
Who it suits
- ✓Operators who know their domain but hit a wall at the technical work
- ✓Anyone maintaining more records than one person reasonably can by hand
- ✓People working on locked-down machines who cannot install a development environment
- ✓Anyone making decisions without colleagues to test them against
- ✓Writers who want an editor and a structure, not a ghostwriter
Try it yourself
There is a free tier. Give it a real problem from your own work rather than a test question, because the difference between the two is most of what people get wrong about it.
Visit ClaudeNot an affiliate link. There is no referral programme here and I earn nothing if you sign up. I pay for my own subscription.
Learning to actually use it
Most people stall in the same place: they can hold a conversation with it but cannot get it to do work. I have written two short courses on closing that gap, one on the fundamentals and one on connecting it to tools like Shopify, which is where it stops being a chat window and starts being useful.
See the courses