Re: LLMs and Agents usage in the CCG

The ideas underpinning PNLs for digital agents aren't any different any from what humans have used for millennia: PhD, DVM, MMath, CPA, LLB, etc.

Also note it's a trivial ask for digital agents to automatically and consistently include/display their personal PNLs ...this is what digital agents are good at.

Many visual presentations of the underlying did:pnl values can selected from. Here's am emoji example:

https://chatgpt.com/s/t_69dec339fed08191b816c873bbad03be

Michael Herman
Chief Digital Officer
Web 7.0 Foundation

Get Outlook for Android<https://aka.ms/AAb9ysg>
________________________________
From: Steven Rowat <steven_rowat@sunshine.net>
Sent: Tuesday, April 14, 2026 2:28:24 PM
To: Moses Ma <moses.ma@futurelabconsulting.com>; Christoph <christoph@christophdorn.com>
Cc: Credentials CG <public-credentials@w3.org>; Mahmoud Alkhraishi <mahmoud@mavennet.com>; Christian Hommrich <christian.hommrich@trailprotocol.org>; Siri Dalugoda <siri@helixar.ai>; Michael Herman (Trusted Digital Web) <mwherman@parallelspace.net>
Subject: Re: LLMs and Agents usage in the CCG


Dear CCG list,

My apologies for jumping across the subthreads here, but I perceive things across the splintering of this main thread. And maybe crossing them can lead to progress, at least in defining the different sections of the problem.

@Siri @Christoph @Christian

You appear to be onto something important in discussing the Human Delegation Provenance (HDP) and filling in its missing gap for LLM lack of persistence, with something like the did:trail method. The nuts and bolts of that are beyond me, so I won't say more about that here. I hope that will work technically.

But if it does, it's possible that it only gets us as far as the interface to the list. It's still necessary to figure out how humans will see it and interact with it. Which is where I find personally a serious problem (ie, like what happened with Morrow). And in my opinion that problem needs to be solved visibly and easily, even for non-tech-savvy readers.

@Moses @Michael

Your responses to my suggestions about what appears in the English language on the list, and how humans react to not knowing they're talking to a bot, were very interesting, and built on top of something like HDP/did:trail they might do the job.

But I have some reservations and comments.

@Moses

Your (and your LLM's) derivation of an emoji system to go with the bot's name could have a strong usefulness in many places, especially in social media. But the emojis themselves are a language, and it's not fair to assume any reader will know each emoji. Plus, email readers don't always display HTML, or anything except text. So maybe your (and your LLM's) original asterisk punctuation system to define the bot and its provenance or capabilities (*Morrow, or *Morrow{v3.2}), *Morrow[trade-exec] etc.), will be all we’ll need as readers.

@Michael

Your (and your LLM's) honing of the bot prefixes into a full provenance system, which you gave a link to, was very interesting. But in terms of its technical abilities to solve the linkage hop problems identified in the HDP/did:trail parts of the thread, it's beyond me and I won't comment. But I will comment in terms of the level of the human reader. Your suggested 'post-nominal' are laid out like a sequence of computer classes, with their values. Even the simplest of the examples involve terms like "EXEC + FIN + VERIFIED". This creates the same problem as the emojis, except the language isn't social-media level, it's code-writer level. The reader must know the terms or it's seriously off-putting. On the other hand, it would certainly be nice, optionally, as readers, to have access to such post-nominal constrictions and verifications. And so perhaps those could be added either at the very top of a post, or as a signature line; in the same way as humans often add a signature line at the very bottom.

So overall: Personally I'm still leaning toward some form of existing punctuation and words to identify the bot's contributions, in the body of the text itself, in the same way as a human name does.

In other words its seems like there needs to be a very condensed translation of the provenance trail into existing English—not all of it, but some selective disclosure, of parts that that the reader will need to know—if we're ever to be comfortable reading words generated by LLMs.

Steven Rowat