Remember what I was saying the other day?
A few days ago I posted about how I was reluctant to fork out money for access to the popular mainstream hosted LLM models… yesterday I forked out for 12 months of access to Claude Code. So, what changed?
Interviewing
It’s June 2026 and I’m on the market for a new job as a software developer, having had my previous role somewhat taken away by AI.
The explanation of the interview process for an interview made particular mention of Claude Code, so it makes sense for me to build up some familiarity with this particular system.
Initial impressions
I have only had a day or so to have a play around with Claude Code, so this is going to be a shallow and brief blog post.
I started with a mostly empty Java project set up with maven for dependency management and tried out prompting Claude Code to create some classes to achieve particular functionality.
A few things surprised me, as I requested some setup to align with how I like to utilise the most up to date testing library. When I saw that the test set up had only brought in JUnit 5 I requested to set up JUnit 6. The model responded that as far as it was aware JUnit 6 had not yet been released – because the model information was based on information available sometime back in 2025.
Similarly, Java 25 – now the most up to date long term support version – also did not exist when the model was built.
Next steps
I’m going to have a look around for efficient and effective ways to bring Claude up to date.
Based on my limited experience and know-how so far, I will want to ensure that the set up will not involve inefficient context and token expenditure.
My expectation is that this will be such a common concern that there will be a trivial way to achieve it.
Why not just ask an LLM?
I enjoy jotting down my experiences, perhaps sharing my naïvety, so if I knew the answer I wouldn’t be able to hit “publish”.
Thinking aloud is still allowed.