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).
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)
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.
