Jamie Lord

This week I stood at a primary school careers fair, on what turned out to be one of the hottest days of the year so far, and tried to convince children between the ages of four and eleven that software development is the best job in the world. I had a stand, a looping deck of slides, a board of brand logos, and a game I had built for them to play. By the end of the day my shirt was stuck to my back, the hall felt like the inside of a kettle, and I had enjoyed myself more than I have at almost…

The Dutch government today blocked Kyndryl’s acquisition of Solvinity, the IT company that operates DigiD. DigiD is the digital identity system roughly fourteen million Dutch citizens use to file their tax returns, book GP appointments, and pay municipal bills. The block came under the Act on Undesirable Control in Telecommunications, WOZT in Dutch, on advice from the Investment Screening Bureau. The Authority for Consumers and Markets had cleared the deal in February on competition grounds.…

At about half past eleven one evening this week I noticed a US Navy E-6B Mercury orbiting over the North Sea. Not “noticed” in the way of someone who happened to look up — I was on the sofa with a laptop balanced on one knee, and the orbit was being drawn for me, in slow careful circles, by a dashboard I had been building, in evenings and weekends, for the previous eight days. The aircraft is a survivable airborne command post for the strategic nuclear force. It does not normally show up on a…

Cloudflare’s Project Glasswing write-up landed today and the Hacker News thread is mostly arguing about whether the prose was written by Mythos or by Opus. It is a fair complaint and an irrelevant one. The diagram halfway down the page is the actual deliverable, and almost nobody is talking about it. Cloudflare has published the reference architecture for doing vulnerability research with a frontier model at scale. The model in the headline is the easy part. The seven-stage agent pipeline…

Cloudflare laid off more than 1,100 people yesterday, around 20% of the company. The announcement, titled ‘Building for the Future’, explains the cuts by noting that internal AI use is up 600% in three months and the company needs to ‘architect itself for the agentic AI era’. The stock dropped 15-18% in after-hours trading. I work at a Cloudflare partner, build on the Developer Platform daily, and have spent the last few years arguing that the platform is the strongest place to put new edge…

Robert Glaser has a long post arguing that individual AI productivity gains do not become organisational gains, and that companies need a ‘Loop Intelligence Hub’ to capture which agentic workflows produce learning. The diagnosis is right. The fix would guarantee the failure it claims to solve. The article hit the front page of Hacker News, and the top-voted comment, from a developer called olsondv, refuted the proposal in real time. ‘There is simply no motivation to develop this sort of…

Sena Evren’s piece on who owns the code Claude Code writes was the most-discussed link on Hacker News this week, and it is a careful walk through human authorship, work-for-hire, GPL contamination, and the Bartz settlement. Every authority it cites is American. So is every authority cited in the 549 comments. The thread argues out US copyright doctrine to four decimal places without anyone mentioning that the UK has had a statutory answer to the same question since 1988. I work on UK contracts…

The Lightning PyPI compromise published on 30 April is being written up as another Shai-Hulud variant, which it is. Versions 2.6.2 and 2.6.3 of the lightning package shipped with a hidden _runtime directory, a 14.8 MB obfuscated JavaScript payload, and the usual exfiltration to AWS, Azure, GCP, GitHub Actions secrets, and any environment variable it could reach. Andy from Lightning has confirmed on Hacker News that the PyPI credentials were stolen via a compromised pl-ghost bot account, not a…

Denis Stetskov’s piece this week, ‘The West Forgot How to Build. Now It’s Forgetting Code’, does a superb job of opening with the Fogbank story and the EU’s failure to deliver a million artillery shells on time. The hook works. The argument that follows, that AI-assisted coding is the same pattern of optimisation eating the talent pipeline, is half right and half a category error, and the half that is wrong is doing most of the rhetorical work. Fogbank failed because the knowledge was physical.…

Most of what I have shipped this year has been built using Claude Code. That part is no longer interesting on its own — half the engineering timeline is talking about it. What is more interesting is that one of the projects I have been working on is also for Claude Code to use, written by Claude Code, to extend what Claude Code can do for me. The agent built its own tools, and now the agent uses them. I want to talk about that loop, because the second time you encounter it the implications…

Six weeks ago I started building something. A real commercial SaaS product, not a prototype — authentication, billing, scheduled background work, document ingestion, AI generation, distribution flows, an end-to-end test suite that actually passes, and the kind of architectural scaffolding that survives a code review. I am one person. The product sits roughly where you would expect a small team to be after a quarter, and the experience of getting it there felt different in a way I do not think…

A reader emailed after my Email Service post to flag something I had missed. John Graham-Cumming, Cloudflare’s CTO until March 2025 and now a board member, has a long pre-Cloudflare history in spam detection. The implication is that the deliverability question I raised, whether Cloudflare can survive the reputation game given their existing abuse posture, has a more credible answer at the top of the company than I credited. The reader is right that I missed it, and the credentials are more…

A proof of concept I built for the College of Policing is included in the techUK ‘From Pilots to Practice’ report this month as the Justice and Policing case study. It was a short engagement against representative data rather than a production integration. The specific details of the work sit with the client; what’s worth writing up is the pattern, because it applies to a lot of public sector organisations. The shape of the problem is familiar. A department or agency has valuable data, but it…

Anthropic published a postmortem yesterday covering three separate changes that degraded Claude Code output between 4 March and 20 April. Two were deliberate product decisions. One was an actual bug. The postmortem treats all three as the same category of problem, and that framing is doing a lot of work. The bug is the interesting part only because it made the other two visible. On 26 March, Anthropic shipped a change that cleared older thinking blocks from sessions idle for more than an hour,…

The fastest-growing category of Claude Code add-ons solves a problem Anthropic has already mostly solved. Native CLAUDE.md, auto memory, plan mode, hooks, and skills cover roughly 80% of what third-party memory and planning tools promise. Before spending a weekend wiring up vector databases and agent orchestration layers, technical leaders should know that the evidence points the other way: more context and more instructions measurably degrade Claude’s output.

Anthropic does not publish exact token counts for any Claude subscription tier. This makes direct comparison between Team Premium seats and individual Max plans unnecessarily difficult, and most online guidance either conflates the two or invents precision that does not exist. What Anthropic does publish are usage multipliers relative to the Pro plan, and those multipliers, combined with the different limit structures each plan uses, are what actually determine which option suits a given work...

I have been writing about Cloudflare’s Developer Platform on this blog for a while now. Posts about Workers, Containers, Analytics Engine, the platform as a whole. Each one started the same way: I would encounter something that surprised me, spend a few days pulling it apart, and write up what I found. It was never planned as a series. I just kept finding things worth writing about.