The “father of the iPod” says the first generation of AI gadgets failed to solve real problems — and the next wave will need to earn consumers’ trust.

Personal AI agents promise to shop, book flights, and make reservations for you. But deliberate blocks and anti-bot defenses are getting in the way, leaving consumers caught in the middle. A new standard aims to help.

hardware device

open-source audio media server

Lotus Improv 2.0 For Windows ( 2.0) ( 1993 02) [ English] ( 2.5'' 1.44 MB)Lotus released a set of three update files which are included in this archive. ...

spreadsheet program

Gathering my thoughts on the use of LLMs for writing prose, and how I’ve changed my workflow following some feedback and reflection on my own writing.

The man said they discovered that their parents’ coffee maker had generated far more traffic than they expected.

Four billion phones ask one text file what time it is. One lecturer keeps it in his spare time.

Feeding random numbers to image generators: the smart new models try to turn nonsense into sense, while plain diffusion models still produce pleasant surprises.

I've been trying to write a positive blog post about using LLMs as a senior developer. I've made a couple of starts: one looking at how...

A drawing of Eliezer Yudkowsky as the Paperclip Maximizer that I commissioned back in 2016 and have had an alarming number of chances to use since This is a...

Bot traffic passed human traffic on the internet in May, according to Cloudflare

How many times have I had to migrate from Microsoft SQL to Oracle or to any other database, or switch the Cloud Data Warehouse or Cloud Providers, sometimes for reasons, and...

Will software development jobs survive the future?

From NeoBook to VisualNEO Win: 30+ Years of Keeping a Development Tool Alive

Chapter XXI, an intermezzo where the rules matter and I stray into the world of AI

The more we focus on a particular technology, the more we lose sight of what normal people actually care about.

Hi friends!

Well over a decade ago, hundreds of conversations over a period of years crystalized into a book called The New Kingmakers. The core thesis of the book was simple: a constituency long thought to be…

On 10 August 2024 Debian bug 1078208 reported that the fail2ban SSH filter no longer worked with OpenSSH 9.8: part of the daemon had moved into a process called sshd-session, and with the changed name on its log lines the patterns stopped matching without reporting any error. Text has been the universal interface since Doug McIlroy wrote in 1978 that the output of every program should be expected to become the input to another, and it works between people and machines because the person at the keyboard notices when a sentence changes. fail2ban removed that person long ago, and on FreeBSD 15 its sshd filter carries 27 regular expressions where the first sshd support in October 2004 matched two phrases. Agent pipelines rebuild the same arrangement with models, one writing prose and another interpreting it, with no agreement about the words. A UC Berkeley team annotated 1,642 traces from seven multi-agent frameworks for NeurIPS 2025 and found task failure rates between 41 and 86.7 per cent, with 32.3 per cent of failures attributed to inter-agent misalignment. OpenAI reported in August 2024 that gpt-4-0613 followed a complex JSON schema less than 40 per cent of the time when asked, against 100 per cent once the platform enforced it. The strongest objection is granted in full: natural language is why these systems compose at all, and many teams already use structured outputs and the Model Context Protocol. The FreeBSD base system met a version of the problem earlier with libxo, which first appeared in FreeBSD 11.0 and lets 25 base programs print text for people and JSON for programs from the same code, while jc needs 220 parsers in 57,683 lines of Python to convert output after the fact. The limit starts at home: ifconfig does not speak libxo, and jc parses it with 728 lines and 31 regular expressions.

Measured on FreeBSD 15.1-RELEASE-p3 with helm 3.21.0: helm create writes twelve files with 539 lines for the scaffolding of one web service, 214 template directives sit across nine of those files, and helm template renders 100 lines of YAML from them. Every programming language in common use was designed by a mathematician, from Turing and Church in the thirties through Backus with FORTRAN in 1957 and McCarthy with Lisp in 1958 to Hoare and the null reference, and their metaphor was the equation. Three devices of human language never made it into that notation: inflection, contextual scope, and mutual agreement between parts. All three came back one layer above the language, in configuration files: inflection as templating, contextual scope as a stack of overrides four deep, and agreement as schemas, in a format whose 1.1 specification reads the bare word NO as the boolean false. A values file with region: NO renders as region: false in Helm 3.21.0, and a key moved two spaces to the right renders silently into the wrong parent. The same service in two environments comes to seventeen lines of jail configuration in one notation on FreeBSD. The piece argues for a surface designed by people who study how meaning is read, with the formal machinery kept underneath where mathematics excels, and credits KYAML, stable in Kubernetes 1.37, as the platform correcting itself.

What I learned from building a less-is-more self-hosted rig in the most normie way possible.

What is a developer to do when they need something more tangible than a chat box? Enter canvases.

A member-governed cooperative for AI inference. Direct the models, the values, and the direction — together.

A follow-up to Donut Lab: a study in deception in venture-backed technology, published 25 March 2026. If you haven’t read it already, go and have a look (please do, it’ll make what’s below make a lot more sense) then come back here. There is a German company called CT Coatings whose patent portfolio covers screen-printed […]