Re: A Framework for Human-AI Collaboration in Standards Work

Hi Steven,

I used Claude Opus 4.7 using a context I built using several AI systems in a chain over several weeks. I usually use this context to pull guiding specs for my architecture. The format and depth of the output is a result of the curated context.

The beauty of this approach is that I can assess external things from my architecture and my point of view and thus I can steward the ideas in the generated documents. They reflect how I think and I could never put it into those words. Whenever I try and explain my POV it lands flat so I kind of have given up trying. Generating documents for those willing to read is my new approach. I have never learnt more than since I have been doing this for myself.

You see exactly what I was trying to achieve:

1) Point out that the problem really needs a much more comprehensive solution

2) Show that AI can help frame the problem in ways that make it clear what must be addressed. Not answers. Just a list of where to look as a starting point.

These kinds of documents help me frame the overall space of a problem so my mind can surface ideas for new approaches and primitives to achieve the outcome rapidly. My work focuses on model-driven system projection so I think very abstract.

One must recognize that any patterns discovered that lead to faster or "better" outcomes will be adopted now at scale. Any patterns that cannot keep up with where coding happens will be phased out. It is a matter of the structure of the dynamic we now find ourselves in.

Christoph



On Sat, Apr 18, 2026, at 5:02 PM, Steven Rowat wrote:
> On 2026-04-18 12:26 pm, Christoph [or an LLM] wrote:
>> 5. The Topics That Must Be Addressed
>> ------------------------------------
>> 
>> 
>> 5.1 An Outline for Moving Forward
>> 
>> To operationalize the principle "humans provide directional judgement, LLMs synthesize and execute" in W3C working groups, the following topics MUST be discussed and resolved:
> 1. Thank you Christoph for what looks to have been a carefully-crafted prompt.
> 
> 2. I'm assuming the output is from Claude, since it references Anthropic. But it would have been nice to have that attribution also.
> 
> 3. But that's just niggling at this fascinating juncture, and the document itself I found gripping, and convincing.
> 
> 4. And yet I'm left feeling a need to point out that section 5, which details "An Outline for Moving Forward", not only has 8 major sections, but several key questions in each, leading to, by my count, 24 different decisions that must be debated, agreed, codified, and implemented, to move towards the successful use of LLMs in the W3C decision process. And many of those have further sub-parts of different languages or protocols.
> 
> 5. Then, Michael Herman's running of the same prompt through ChatGPT, which he reports in a post in this thread, though using different abstract terms, ends up with, similarly, at least 7 major sections of changes that have to be addressed, some of which seem to overlap with the ones in Christoph's LLM.
> 
> So overall, I'd say: at least the first two LLMs seem to be in *a very rough consensus* about what has to happen. 😭
> 
> Which, to thrown down the gauntlet, I'm going to say is, okay, probably a good start on a process that may take quite a while, us being human an' all. 🙂
> 
> Steven Rowat
> 
> 
>