Alasdair Allan

@alasdairallan.com

Scientist, author, hacker, maker, and journalist. Writes, speaks, and builds. An accidental privacy advocate.

In 1988 a company that made record turntables built its own CPU, because it hated its factory software. It failed. Every one of those ideas ships today. Hardware memory safety. Garbage collection in silicon. Persistent objects. Chips shaped to a workload. negroniventurestudios.com/2026/07/18/t...

The computer at the bottom of a canal

Somewhere at the bottom of the Forth and Clyde Canal, buried in the silt, there is a box of custom silicon that was right about almost everything. Back in 1988 a Scottish hi-fi company shipped a proce...

negroniventurestudios.com

Announcing agentlanguages.dev, a community-edited catalogue of the 21 programming languages designed for AI agents to emerge since late 2025. The most interesting thing to emerge? They fall into three philosophical camps that disagree on the underlying problem. #AI #LLM

agentlanguages.dev — Programming languages designed for AI agents to write

A community-edited catalogue of programming languages designed for AI agents to author code, organised around three philosophical camps: syntactic, verification, and orchestration.

agentlanguages.dev

Expanded the Vera benchmark to 6 models, 3 providers. Kimi K2.5 got every Vera problem right while only managing 86% in Python and 91% in TypeScript. Flagship average Vera and Python level at 93%. Yet there is no Vera in their training data, no examples on GitHub or Stack Overflow. veralang.dev #AI

VeraBench v0.0.7 benchmark results across four panels. Flagship tier: Kimi K2.5 scores 100% run_correct on Vera versus 86% Python and 91% TypeScript; Claude Opus 4 scores 88% Vera, 96% Python, 96% TypeScript; GPT-4.1 scores 91% Vera, 96% Python, 96% TypeScript. Sonnet tier: Kimi K2 Turbo scores 83% Vera, 83% Python, 79% TypeScript; Claude Sonnet 4 scores 79% Vera, 96% Python, 88% TypeScript; GPT-4o scores 78% Vera, 93% Python, 83% TypeScript. A horizontal bar chart shows Kimi K2.5 is the only flagship model where Vera beats both Python (+14pp) and TypeScript (+9pp). A fourth panel breaks out all four modes including Vera spec-from-NL, showing GPT-4.1's spec-from-NL score dropping to 50% while Kimi K2.5 holds at 100% across both Vera modes.

I built a programming language you're not supposed to read, veralang.dev. Vera is designed for machines to write. No variable names. No style choices. Compiler errors are instructions for the model, not diagnostics for you. Go and point a model at it, negroniventurestudios.com/2026/02/28/a....

Vera — A language designed for machines to write

A programming language where verification is a first-class citizen. Mandatory contracts, algebraic effects, typed slot references, compiles to WebAssembly.

veralang.dev

"We need a new Turing Test," by me for Negroni Venture Studios. The real test for artificial general intelligence isn't fooling humans, it's demonstrating flexible real-world competence that our token-prediction architectures fundamentally can't deliver. negroniventurestudios.com/2025/11/30/w...

We need a new Turing Test

Every time there is some sort of perceived advance towards a general artificial intelligence, we talk about it as a breakthrough, right up until we understand what it’s doing, how it’s “thinking”. The...

negroniventurestudios.com

[Alexa] will know "almost [everything] in your life — your schedule, your smart home, your preferences, the devices you’re using, the people you’re connected [to and] the entertainment you [enjoy]…" they understand that most people don't think that's a *good* thing? techcrunch.com/2025/02/26/a...

Amazon unveils a new and improved Alexa, Alexa+ | TechCrunch

At an event in NYC, Amazon announced an upgraded Alexa experienced powered by generative AI tech.

techcrunch.com