Tom Oram

Claude Code estimates work in the opposite way to how humans do it:

> Would you like to proceed with option 1, it will take 2-3 hours?

Yes

..5 mins later

> Done

Have had Claude Code generate quite a few authorisation rules that look great on first glance but contain significant flaws when studied. I certainly foresee a lot of companies ending up with these kinds of scenarios in the near future.

What’s the thinking on comments in code when using agents manipulate the codebase? The agents add comments and I’m inclined to remove the ones that don’t seem to add value (i.e. most), but I do wander if the context helps future agent interactions?

Today has been an extended example of why it’s important to run your infrastructure deployment code from scratch on a regular basis 😬

Refactoring Legacy Code with Alessandro Di Gioia: Don't miss the Episode!

Trips to the office are a great chance to get a bit of fun coding.

Today I’m pairing with Claude Code on a bit of Idris. Claude definitely knows Idris better than me but we are both struggling a bit.

Thanks to Claude, I can now hit the wall where a realise what I’m trying to do with type-level Haskell is impossible much quicker.

There is definitely a correlation between the number of shitty snacks I eat in a day and the amount of documentation I write in Confluence.

One powerful aspect of mob programming that I don’t see mentioned much is that if the whole team is working on one thing then they must choose (or be given) the most important thing to work on - prioritisation is naturally pushed front and centre.

YouTube video by Kai Lentit Interview with Cloud Architect in 2025

“Every year we get personal thank you letter from cloud provider for hitting Q1 revenue target…their Q1 revenue target”

It’s funny, but it’s also seems tragic to me how much many larger companies are willing to spend unnecessarily on cloud resources.

youtu.be/xIk0_uFV-rU?...

Technology migrations can be quick, simple and straightforward if you discount all the other things that are integrated with it and the people who use it.

IntelliJ IDEA Moves to the Unified Distribution | The IntelliJ IDEA Blog We are excited to announce the next step for IntelliJ IDEA: we are moving to a single, unified distribution. And yes, before you ask, our commitment to open source remains as strong as ever. There

Good move by JetBrains: blog.jetbrains.com/idea/2025/07...

This will benefit experienced developers, beginners, and students who get introduced to the tool early on as well.

The basics of OpenTelemetry are nice clean and straightforward, however, wading through the piles of information around which specific attributes to include isn’t a whole lot of fun (not least because 90%+ of them are marked as being in development).

Today has been a day where the LLMs have been a hinderance rather than a help.

OTLP: Open Telemetry Protocol

OLTP: Online Transaction Processing

My brain, tongue, and typing fingers dislike the fact that both of these abbreviations exist.

Giving Claude code a try. I tried to be ambitious and create an iOS app for my first attempt, it’s not gone too well so far.

Often Haskell projects have code which can look quite daunting, but Hadolint has done a great job of being clean, simple and accessible.

Two parenting tips for people with a 4 year old:

1. Get construction toys with real tools like screwdrivers. Real tools are so much more fun and engaging than plastic toy ones.

2. Don’t let your child have a real screwdriver, they will take things apart!

Not personally, but I’ve seen posts on Reddit from people claiming they have been pulled up by management (who have been tracking their usage) for not using their LLMs enough.

My general rule is that if there's a certificate in it, it's not Agile. The very notion of a certificate is based on a faulty premise: that Agile is a fixed process, and if you learn and do that process, you will be Agile. That's just wrong.
1/5

Quitting programming as a career right now because of LLMs would be like quitting carpentry as a career thanks to the invention of the table saw.

I extracted two sub packages out of a larger one today. I only did it to improve code organisation, however, it had a very positive effect on the overall design because the new boundaries made it much clearer where things weren’t quite right.

I had to bail on Podman and revert to Docker because its performance (on Windows with WSL) was much worse when using the Azure EventHubs Emulator. It may have been fixable but I don’t have the time to look into it atm.

Time to look at the Manning MEAP page and make a FOMO induced purchase.

Trying to install Podman Desktop on Windows for the first time. The onboarding process is pretty nice so far, with clear guidance on what you need to do to enable required features.

Lost an hour today bashing my head against why the Flink Kubernetes Operator was failing to create a deployment; it turned out that I had deployed a newer version of Kubernetes which I guess isn’t currently supported.

Having event properties in C# that throw when you try and add more than one handler with += seems like a weird choice.

Being an event suggests that you can add multiple handlers but then you can get caught out unexpectedly at run time.

Playing with data streaming technologies with a focus on exploring different architectures to build real-time data products.

(and because it’s fun 😉)

A day of intense redesigning of an Azure EventHubs integration has left me with no energy left for tinkering with my own stuff.

A challenging problem is bait for a software engineer, however, when a book says:

“Here is a big long list of all the complicated problems you might eventually need to solve, but first start with a simple solution and get it working first.”

That last part is the most important.