I never bet on vector database tech (and I don't like RAG in general). Reasons:

- You create a new problem to solve an already tough problem (information retrieval) - You bill people for storing vectors and searching when they are already paying for information storage and search - You overcomplicate a retrieval problem that has well-defined solutions - All the BS HYPE

The LLM wiki is one good option. At WAIC, I saw a company change the filesystem to store the document index as you save documents so they never go out of sync.

Think harder boys.

A good solution is to break the system into small pieces.

You no longer think in terms of a "single system" - you think in terms of composition, components, and combination. You may find category theory useful here, too.

Before AI, these higher-level abstractions had little practical value because translating them into reality was a huge undertaking.

AI shortcuts the "building" and "translating" process, so the practical value is magnitudes more palpable.

I have seen some places keeping really old machines in their data center, but they are not running anymore. They're there either for system archaeology (figuring out how the old system worked) or because disposing of them is more expensive than keeping them.

Nice article.

Brought back a lot of memories.

In the early 2000s, when video games weren't protected against DLL injection hacks, I spent a lot of time writing DLL injectors for fun and learning.

The line I used in every experiment was:

`jmp $` (or `EB FE` in machine code)

It was the shortest infinite loop in x86 assembly, equivalent to `while 1: ;`

I would create a minimal DLL with this assembly code, inject it into games like Battlefield 1942, and observe it in OllyDbg. The injected DLL would then start on a new thread and access all the game data in memory.

My prediction for 2027 is that the world will be powered by two kinds of languages: low-level language(s) that implement the deterministic compute layer (i.e., tools) for agents, and high-level language(s) that implement agents and orchestration. Based on my observations, the strongest choice for each tier is Rust and TypeScript. You may also count Golang and Python.

In 2027, other languages will become increasingly irrelevant.

Long-time reader of simonwillison.net here. It's a nice guide, but I think it's missing a few points. Point 1: Software architecture is still a differentiator in the agent era. It is the single measure that separates absolute bu..sh..t from gold. Point 2: LLM/Agent does not suffer from the maintenance burden; the human owner does. So you must somehow "make it *feel* the same suffering" when the codebase starts to become a sh.t pile. Point 3: Agent by default does not align with your goals - they pretend to be - so you need an alignment system to ensure it would do everything to achieve *your goal* (not its own goal).

Here is a system I developed for my own projects:

1) hand-write a simple 2d TUI-based rogue-like in Rust using pretty much just the std; 2) grab opus 5.0 (it used to be opus 4.6, 4.7) and give it some vague "requests", and ask it to make this game "production-ready" and "blockbuster", but keep the 2d and TUI aspects so I can actually run it. 3) now the fun part, take a test subject, say GLM 5.3, and ask it to find code smell, architecture issues, duplication and all sort, and *simplify the code*

compare the result to my original version.

It's not a simple thing, but the concept is simple: can an LLM remove all the mud?

The winners so far are (ranked by the quality of the final result, not by token cost)

GPT 5.6 sol (extra high thinking); GLM 5.3; Grok 4.6; Qwan 3.8;

(fable could not make it to the list because it simply cannot follow the instructions)

Moved to Codex and breathing fresh air.

python, perl, LLVM, rust, Go, k8s

Seconding Python

I stopped being a Dell customer after owning 3 XPS for 13 years.

Currently using an Aussie brand named Metabox (metabox.com.au) for virtual production workload on Windows. I've been using a Macbook Pro for the last 5 years for general-purpose tasks. Also using a few $100 "mini PCs" for home automation.