How I've used AI to improve my work

My adventures in vibe coding

The problem with AI being able to do ‘anything’ is that it is often hard to know where to start.

If you fire up ChatGPT, Google Gemini, or god forbid Grok, the only question you’re really asked is “What can I help with?”, with no other hints or suggestions about what the apps are capable of. So no wonder some people find Large Language Models (LLMs) unimpressive, or consider them at best a glorified search engine.

I think that’s why I received such a surprisingly big reaction to my post in December about how AI has made my life better. All I really did was offer some examples of ways in which I use the technology in my somewhat weird job as a freelance writer and journalist, but after it was published I received a bunch of messages from people whose eyes I’d opened to the possibilities of the technology.

That’s why today I want to go a little deeper – and nerdier.

The view amongst many programmers is that over the last month or so, something has really changed when it comes to coding. For example, Andrej Karpathy, who co-founded OpenAI argues that AI has “crossed some kind of threshold of coherence around December 2025 and caused a phase shift in software engineering”.

His comments followed the release of the most recent versions of Claude Code and ChatGPT Codex, which are both specialist apps that are optimised for writing code. The way they work is that instead of the LLM simply writing a response that you can read or paste elsewhere, you instead point them at a folder on your computer and then the AI writes code and manipulates the files on your computer all on its own.

And I can personally attest that the latest version of Codex is incredible.

Though I’m not a professional coder myself, I am a hobbyist, having trained myself to code a little as a teenager. In the two decades since, I’ve regularly built myself tools, web apps and command line scripts to do useful things like scrape data or manipulate datasets in useful ways.

I think this was one of the reasons I immediately saw the value in ChatGPT after it launched in November 2022. Instead of typing out code manually, I realised I could just say to ChatGPT “Write me a script that downloads data from here, and puts it into a database here”, or whatever. And it would get it mostly right.

But this, it turned out, was just the beginning. Over the past couple of weeks, I’ve spent some time working with Codex to build a bunch of tools that are specifically useful to me in my work and life. This isn’t hypothetical – these are real, useful applications that I am now using regularly. So I thought it might be interesting to tell you about them, as I can’t believe what I’ve been able to build.