I’m 50 Now, and the Thing I Loved Has Changed

JAMESRANDALL.COM

… and the changes are coming in waves.

Over the Christmas period we observed an uptick in people sharing their amazement at the power of the latest models. People experiencing first hand the power of agentic loops and long-running agent execution. Learning that the way to be most productive in this new world requires adaption, and a fundamental change in their role.

But boy were they productive.

Now we seem to be experiencing a post-Christmas slump. I’ve recently shared numerous posts from experienced engineers who are on the one hand AI optimists. They have adapted, they are running faster. They are doing things they never would have imagined. But at the same time they have lost …. something?

Something that is hard to articulate.

James Randall has a similar journey to myself (and so many others), starting life programming on computers where you could easily understand the full stack, right down to the schematic.

Technology has evolved, and become infinitely more complicated. But the skills needed to be effective have stayed much more constant. And the joy of exploring, learning, creating and problem solving has endured.

But this time, things feel different. And it is deeply unsettling.

We mourn our craft - I didn’t ask for this and neither did you

NOLANLAWSON.COM

AI is amazing, what it can create is amazing, the speed at which it can create it is amazing. But it isn’t all upside. It is almost certainly going to have an impact of the craft of software engineering.

And I am quite deliberate in my use of the word ‘craft’. Most people who work in software not only enjoy the products they create, but also enjoy the creation process itself. In fact, a lot of us enjoy the creation process the most (users are a pain!)

“I didn’t ask for a robot to consume every blog post and piece of code I ever wrote and parrot it back so that some hack could make money off of it.”

This isn’t an easy time for our industry, we’re faced with greater productivity, but at the same time it feels like we are losing something.

“We’ll miss the feeling of holding code in our hands and molding it like clay in the caress of a master sculptor.”

This blog post is short, but gets right to the point. Nolan isn’t denying that the world is changing fast, he isn’t actively resisting. But the pain is real.

“Now is the time to mourn the passing of our craft.”

As someone who has spent countless hours crafting software, writing code just for the fun of it, I feel the loss too.

Building a C compiler with a team of parallel Claudes

ANTHROPIC.COM

Claude Opus 2.6 has just been released, with a focus on agentic coding and its increased ability to undertake complex software tasks over long timescales. The benchmark scores have of course increased, but I’m much more interested in ‘real world’ applications of this technology.

In this post an Anthropic engineer shares a story of how coordinating a team of agents in order to create a C compiler (written in Rust) that can compile Linux 6.9 on x86, ARM, and RISC-V. A pretty ambitious project.

Two weeks later, after 2,000 Claude Code sessions (at a cost of $20k), the 100,000 lines-of-code project was functional. Not a perfect compiler, a bit slow, and a 99% benchmark score, but good enough to compile QEMU, FFmpeg, SQlite, postgres, redis and of course Doom.

This post is quite honest about this being an ideal problem for an agent, with a near perfect evaluation and feedback mechanism. But it is sill deeply impressive and there is much we can learn from those who are pushing agentic coding to the very limit.

Introducing GPT‑5.3‑Codex‑Spark

OPENAI.COM

An ultra-fast model for real-time coding in Codex …

OpenAI has launched GPT-5.3-Codex-Spark, a real-time coding variant of its Codex family optimised for ultra-fast feedback loops inside editors and terminals. This marks a bit of a departure from other recent model releases which have been focussed on overall capability (benchmark scores, long-running task execution etc). Running on specialised hardware with a >1,000 tokens/sec inference speed and a 128k context window, Spark trades some deep reasoning for responsiveness, making incremental edits and rapid iteration feel almost synchronous.

The initial buzz on Hacker News reflects this duality: many developers are excited by the speed and real-time interaction, seeing it as a way to keep “in the flow” when sketching code or refining logic, while others highlight that its depth and consistency can feel lighter compared with the full GPT-5.3-Codex in more complex tasks.