Claude: A Working Review After a Year in the Trenches

A note before we start

The interview for this review was conducted by Claude. Yes, that Claude. The one being reviewed.

I could pretend that is not worth mentioning, but it is actually the most interesting thing about this piece. I asked Claude to interview me about my experience using it, and it did what a good interviewer should: it asked follow-up questions, pushed back when my answers were too generous, and asked directly about the things that frustrate me. It even asked me how I reconcile using a closed-source American AI product with my values around open source and data sovereignty. That is not a question a marketing department would suggest.

The result is a review shaped by a conversation between a user and the tool itself. You should read it with that context in mind. But you should also consider this: if an AI can conduct a critical interview about its own shortcomings without flinching, that tells you something about the tool that a star rating never could.

Make of that what you will.

The setup

Most AI tool reviews are written after a week of testing. This one is different. It is based on a conversation with someone who has used Claude daily for over a year, across software development, business strategy, data analysis, compliance management, and elected council work. Not a demo. Not a controlled experiment. A year of real, unglamorous, Tuesday-morning use.

What follows is an interview-style review: honest answers to direct questions, covering how Claude earned its place in a complex working life, where it falls short, and whether it is worth the money.

How it started

How did you first come to use Claude?

I had been using both ChatGPT and Google Gemini. I was building an email parsing and categorising tool for personal use and spent many hours going in circles trying to get it working. I had heard of Claude via a podcast but had not investigated, so I gave it a try. It did not just solve the issues. The Python code it helped me write still runs now, over a year later, completely unchanged. But what really hooked me was the desktop app, which at the time was not a feature most other AI tools offered.

That email parser still running unchanged after a year is a more meaningful endorsement than any benchmark score.

Where Claude sits in the workflow

Where does Claude fit in your actual working day?

It is now an integral part of my workflow. Not just for system development and coding, but for data analysis and strategy too. I love having the mobile app. I often dictate to it whilst doing other life tasks, early morning or late at night, and set it off on a challenge, ready to pick up when I sit down at the laptop.

That is not just a feature preference. That is a different relationship with the tool. You are thinking through problems while walking or running, speaking them aloud, and coming back to a worked response.

Exactly. It changes when and how ideas develop. The thinking does not have to wait until you are at a desk.

How does Claude handle real-world code? Not clean-slate demos, but the messy reality of existing codebases with business logic and legacy decisions?

It has become revolutionary for that. I manage some code that was written by third-party developers. Debugging and replacing that code has become quick and efficient. Starting from scratch on a project, the step from idea to working prototype can be hours, not weeks. Sometimes a quick frontend demo designed in a meeting can be the difference between people understanding the idea and it not landing.

Where it falls short

Every tool has rough edges. Where does Claude frustrate you?

Sometimes it has off days. I think it is fair to say that. This is often when a new model is released but can happen during other system issues too. Just like any team member, the code, workflow, and understanding needs checking. It can be frustrating when the same mistakes occur on multiple projects.

The "off days" observation is one that does not get enough attention in reviews. Most reviews are based on a single session. You are describing the experience of relying on a tool day after day across months, where consistency matters as much as peak capability.

That is exactly it. When you depend on something daily, the occasional dip in quality is more noticeable than it would be in a one-off test. You learn to recognise when it is not quite right, the same way you would notice a colleague having an off morning.

The open-source question

You care about open source, data sovereignty, and who controls the technology. Claude is a closed-source product from a well-funded American AI company. How do you square that?

Being closed source is problematic. However, I have tools to ensure my learnings and experiences are backed up elsewhere, so I can transport outside of Claude if I need to. I would love to see a usable open-source model which has the slick integration of Claude, but we are not there yet.

Compared to other AI tools, I suppose it is the lesser of the evils.

"Lesser of the evils" is an honest position, and I think a lot of people with similar values share it. The pragmatic choice, made with eyes open, while waiting for something better on the open-source side.

Value for money

You are on the Max plan at $100 a month. Does it pay for itself?

That $1,200 cost per year is very outweighed by the time I save and the additional work I am able to produce. Less tangible is the brain dump and offloading benefits it brings. I am able to outsource much of the mundane processing, the stuff that would otherwise sit in my head taking up space. When you are managing multiple domains of responsibility, that mental offloading is worth more than the time savings alone.

Who is it for

Who would you recommend Claude to?

I would recommend it to anyone, certainly with the free version to try it out. I think it particularly fits for those of us who straddle multiple domains of work, personal life, and other activities. If you are someone whose day involves switching between code, strategy, compliance, community work, and everything in between, having a tool that can context-switch as fast as you do is genuinely useful.

It is not about Claude being good at one thing. It is about it being good enough at many things that it becomes a constant in an otherwise chaotic working life.

The verdict

Claude is not perfect. It has inconsistent days. It repeats mistakes you have already corrected. It is closed source, which matters if you care about who controls your tools. The Max plan is a real cost that not everyone can justify.

But after a year of daily use across software development, business strategy, data analysis, compliance, and public service, it has earned its place. The code it helped build still runs. The thinking it helped develop led to real decisions. The time it saved was reinvested in work that matters.

The fact that the strongest endorsement comes not from a feature list but from a working routine that now depends on it says more than any specification sheet. Claude is the AI tool that survived contact with a real, messy, multi-domain working life. That is harder to achieve than passing a benchmark.

The lesser of the evils, perhaps. But also, for now, the best tool for the job.