Re: AGENTS.md

On August 5, 2026 8:06:42 AM GMT+03:00, Richard Stallman <rms@gnu=2Eorg> wr=
ote:

>Eli wrote:
>
>    Do we have any idea how effective is this in practice?  Does this
>    prevent what we want to avoid with likelihood that is close enough to
>    100% for us to consider it the "solution"?  Because if not, we'd need
>    this _and_ the hooks as Plan B=2E

It's notion (agents=2Emd)to have one file that modifies LLM prompt so that=
 proprietary companies don't need to invent many different files so that th=
eir software can run easy and that they minimize their development cost=2E

It's not effective in practice=2E

User can simply say to software to ignore the agents=2Emd file=2E

User can delete the file=2E

Too many software using LLM including Emacs packages don't observe agents=
=2Emd

Agents=2Emd is just optional and modified prompt to any prompt given to th=
e LLM in certain project=2E

It's not a static file=2E

Alternative:

It's very easy in theoretical new package to have file, dir or project mem=
ory in Emacs=2E Simple assign to specific dir some file which represents mo=
dified prompt=2E It doesn't need to be called agents=2Emd but anyhow=2E It =
can be in any format=2E Other Emacs packages can follow that to check on si=
ch fundamental package and to find modified prompts=2E When LLM works in sp=
ecific file buffer it can find memory related to the file and apply it=2E T=
hat would be Emacs way=2E We develop in Emacs=2E Why accommodate external t=
hird party Software? Why not make it better within Emacs?

Adoption of agents=2Emd serves proprietary companies and is not effective =
o=C5=BEand simply it can't prevent users to generate LLM code=2E

>That is a good question=2E  Do we actually know how much, or in what way,=
 LLMs'
>behavior will be affected by an AGENTS=2Emd file?

No, we do not know how LLMs will behave when they encounter AGENTS=2Emd=2E=
 We cannot predict it because:

=C2=B7 LLMs are pattern matchers, not intelligences
=C2=B7 They don't "understand" meaning or intent
=C2=B7 Their behavior depends on training data, model version, and user pr=
ompting
=C2=B7 Different models respond differently to the same prompt
=C2=B7 The same model responds differently at different times

We are guessing that AGENTS=2Emd will influence LLMs=2E There is no scient=
ific basis for this guess=2E We are treating statistical text generators as=
 if they were rational agents who read and comply with instructions=2E

There are millions of LLMs and much software that don't care=2E

That is not how LLMs work=2E We cannot rely on guesses when the consequenc=
e is adopting a proprietary vendor's convention so that they can have it ea=
sier and cheaper=2E

Let's not host OpenAi or Anthropic's prompt file in GNU Emacs=2E If we wan=
t policy, write it in CONTRIBUTE for humans=2E If we want enforcement, use =
technical hooks=2E AGENTS=2Emd is neither=2E

And to be more blunt that file is for those who are to use Claude which is=
 SaaS LLM running on proprietary software=2E Why accommodate such?

Users who wish to generate LLM code cannot be prevented by it=2E



Jean