Martin Alderson

I've always thought of systemd as a fairly boring piece of plumbing, that boots your system and runs background services.

With the Anthropic IPO gathering pace - and chip stocks just posting their longest losing streak since 2008 amid worries about AI debt, I thought it'd be worth writing up some of my thoughts I've put together while thinking through what may happen.

The social media discourse is ablaze with people porting apps from slower dynamic languages to faster statically compiled languages.

I've written a lot about the economics of inference - especially with the advent of "good enough" open weights models.

The highly publicised sandbox agent escapes have certainly made news, and I wrote about the issues with sandboxing agents back in January - though I certainly didn't foresee they would escape the frontier labs.

Like many, when I'm building with coding agents, one of the "worst" problems is the codebase feels like it is often evolving faster than I can keep up with it.

Z.AI confirmed that their most recent model release was running all inference on Chinese manufactured hardware.

Over the winter of 2025, after the release of Opus 4.5, coding agents grew tremendously and usage exploded.

AI inference costs are something I've been writing about for a while.

I know I'm guilty of just scanning OpenRouter's pricing tables and looking at input and output costs per million token.

For the first time I can remember, I'm not choosing my daily driver models on raw intelligence.

In the past few days Hugging Face announced a security incident, which transpired to be a "runaway" agent from OpenAI.

This is the second article in a two part series focusing on what I believe is perhaps the least understood upcoming shift in AI economics.

This is a two part series focusing on what I believe is perhaps the least understood upcoming shift in AI economics.

While researching and writing my last article on the history of KV cache compression, it occurred to me while there has been so much implemented research on KV cache efficiency, actual model weights quantisation is still pretty blunt.

While much is focused on the improvement of models, there's been radical improvements in the efficiency of KV cache compression.

An unexpected development over the past few weeks is xAI's new partnerships with Anthropic and Google, providing them with a huge amount of capacity.

In the UK (and I'm sure elsewhere) politicians and commentators are falling over themselves to suggest that without huge fleets of datacentres built in the UK that we are going to be hopelessly left behind.

I joined Radical's "Built for Turbulence" podcast recently for a wide-ranging chat on what AI agents are doing to the economics of software.

Google is in a strange spot right now.

Managed agents (cloud-hosted agents) are the next big push from the frontier labs.

It's been a bit of a given in the LLM world that there will be somewhat competitive open weights models.

On 29 April 2026, a Korean security firm called Theori published 732 bytes of Python that breaks Linux container isolation.

I think Figma is increasingly becoming a go-to case study in the victims of the so-called "SaaSpocalypse".

I've been curious for a while about what's actually happening inside Mixture of Experts models when they generate tokens.

For nearly 20 years the deal has been simple: you click a link, arbitrary code runs on your device, and a stack of sandboxes keeps that code from doing anything nasty.

I thought it'd be a good time to continue on the same theme as my previous two articles The Coming AI Compute Crunch and Is the AI Compute Crunch Here?

While the world's been watching physical supply chains, a different kind of supply chain attack has been escalating in the open source ecosystem.

I don't think people have fully internalised how good agents are at reverse engineering code.

Last Friday Anthropic released a new (production at least - has been in beta for a while) 1M context window variant of Opus 4.6 and Sonnet 4.6.

I've always been really impressed with how well the Gemini models do OCR of difficult PDFs - not nicely formatted PDFs, but badly scanned images in a PDF file.

My LinkedIn and Twitter feeds are full of screenshots from the recent Forbes article on Cursor claiming that Anthropic's $200/month Claude Code Max plan can consume $5,000 in compute.

In January I wrote about the coming AI compute crunch.

There's a growing narrative that on-device AI is about to free us from the cloud - the pitch is compelling.

Most existing AI code review tools require you to grant them access to your GitHub or GitLab repositories.

I wrote an article a couple of months ago about which languages are the most token efficient.

Anthropic's red team released research showing that Claude Opus 4.6 can find critical vulnerabilities in established open source projects.

I cloned most of Linear's core functionality in 20 prompts using Claude Code.

One of the biggest things to improve how agentic tools like Claude Code/Cowork and Codex work is by using CLAUDE.md or AGENTS.md files[1] - which give the agent context on the project.

Every organisation has brand guidelines that nobody follows for internal documents.

The "SaaSpocalypse" began on the 3rd of February 2026 - with $285bn wiped off technology companies on the public markets.

It still shocks me how much difference there is between AI users.

I've been a TDD sceptic for most of my career.

While I've been blown away by the development in coding agents, I'm starting to get worried there are some quite serious security risks coming from them.

There has been so much written about the "unsustainable" AI capex recently.

I've been trying to think through what happens to programming languages and tooling if humans are increasingly no longer writing it.

Over the holidays I saw a link to the original Photoshop 1.0 source code from 1990.

Over the past few months I've found my enthusiasm for MCP somewhat wane.

Travel agents are the go-to example of an industry killed by the internet.

You've heard the new narrative: AI scaling hit a wall, the capex is insane, the returns aren't there.