Agents have collapsed the cost of building native apps twice. But I think they will also make formal verification the default way software gets written. The social media discourse is ablaze with people porting apps from slower dynamic languages to faster statically compiled languages. And we've seen both Shopify and Coinbase start switching away from React Native back to native mobile apps. What's

OpenAI has cut Luna prices 90% in two months, DeepSeek is still king on cached input, and even Anthropic is now cutting Opus prices. What that means for the frontier labs. I've written a lot about the economics of inference - especially with the advent of "good enough" open weights models. We're starting to see real price competition from all the frontier labs, even Anthropic who have been reticen

The recent agent sandbox escapes at OpenAI and Anthropic look less like a technical failure and more like a philosophical one - treating security controls as though they only need to work most of the time. 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 th

A simple technique for keeping up with a codebase that coding agents are changing faster than you can read: ask the agent to quiz you on it. 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. This cognitive debt builds up quietly - nothing is broken, but you gradually have this feeling of n

Z.AI ran a full model release on domestically manufactured Chinese chips. Impressive, but EUV lithography is a hard wall and the gap with Western inference hardware is more likely to widen than close. Z.AI confirmed that their most recent model release was running all inference on Chinese manufactured hardware. While no doubt an impressive feat, Western companies still have a huge advantage that I

Winter 2025 was the tipping point for coding agents. Summer 2026 is shaping up to be the same for open weights - and compute supply, not model quality, is now setting the price. Over the winter of 2025, after the release of Opus 4.5, coding agents grew tremendously and usage exploded. I think this summer is proving itself to be a similar tipping point for open weight models. The compute crunch and

A practical framework for reducing AI costs - auditing token spend, killing legacy models, moving workloads to open weights providers, and fixing the agent and tool inefficiencies that quietly burn tokens. AI inference costs are something I've been writing about for a while . It's clear that for many companies this is becoming a huge problem: AI spend per employee per month. Source: Ramp AI Index

For long agentic sessions cache reads are most of the bill, not input or output tokens. KV caches have shrunk dramatically while cache read pricing has barely moved. I know I'm guilty of just scanning OpenRouter's pricing tables and looking at input and output costs per million token. I've realised that's the wrong number to be focused on these days and cache read costs are actually far more impor

For the first time I'm choosing my daily driver models on tokens per second rather than raw intelligence. Why ~100tok/s might be the new 100ms, where the speed gains stop mattering, and the price war that's coming. For the first time I can remember, I'm not choosing my daily driver models on raw intelligence. I'm choosing them on speed. Have we reached an intelligence tipping point? This is probab

Breaking down the Hugging Face security incident caused by OpenAI's own models during a benchmark run - the sandbox escape, the package proxy, and whether it's really a marketing stunt. In the past few days Hugging Face announced a security incident, which transpired to be a "runaway" agent from OpenAI. This is likely the first known autonomous offensive agent working like this, and certainly the

Part two: as AI inference margins collapse, who actually captures the value? The hardware supply chain and consumers win, model inference commoditises, and the frontier labs' escape routes are managed agents and staying ahead. 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. If you haven't read it yet, I'd re

GLM 5.2 is the first open weights model I'd call a genuine competitor to Opus and GPT for agentic work - at ~15-20% of the price. Part one of why AI inference margins are about to collapse. This is a two part series focusing on what I believe is perhaps the least understood upcoming shift in AI economics. If you've enjoyed this, part two is now out here . And to be notified about future posts, ple

Profiling a MoE model to find which experts matter for a specific task, then quantising the cold ones hard. The result: near-Q4 quality at near-Q2 size for local models. 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 b

How KV cache compression - from MQA and GQA to MLA and linear-attention hybrids - quietly unlocked the long context windows that make modern agentic LLMs possible. While much is focused on the improvement of models , there's been radical improvements in the efficiency of KV cache compression. I was curious to figure out just how big the improvements are and why I think it matters so much. The head

xAI is renting huge amounts of GPU capacity to Anthropic and Google. Financial engineering ahead of the SpaceX IPO, a real compute shortage, or a genuine datacentre advantage? Probably all three. 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. It's worth remembering that xAI is now part of SpaceX,

The UK is obsessed with building AI datacentres at home. But the arguments for sovereignty - latency, tax, control - mostly don't hold up. 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'm not convinced this is the case, and it risks reall

I joined Radical's Built for Turbulence podcast to talk about what AI agents are doing to the economics of software, the Figma Trap, and why running human-written code without AI audit is going to start looking reckless. 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. We got into whether 5 people with agen

Google's Gemini 3.5 Flash was the headline model at I/O - fast, but expensive and middling at coding. Why it makes more sense as a model built for Google itself, the TPU advantage, and Google's real weakness in coding agents. Google is in a strange spot right now. They've got arguably the deepest research bench in the industry, their own custom silicon, and effectively unlimited money - and yet mo

Managed agents (cloud-hosted agent harnesses) are powerful, but locking yourself into a frontier lab's platform now is risky - here's why and what to do instead. Managed agents (cloud-hosted agents) are the next big push from the frontier labs. They're genuinely incredible. They're also going to be the AWS Lambda of this cycle - powerful, sticky, and an absolute nightmare to migrate off once you'r

Open weights models keep frontier labs honest on price. If they disappear, we end up with a handful of oligopolists extracting consumer surplus. It's been a bit of a given in the LLM world that there will be somewhat competitive open weights models. I'm not sure that's a good assumption anymore. A short history of LLMs In the relatively brief history of LLMs, there's been two [1] types of LLMs - c