New protocols for AI

I really like this provo­ca­tion in Tina He’s 2026 open inquiries:

Where does value pool when every­thing interoperates? Who decides what gets included in the default spec? What becomes the DNS for agents — the uni­versal direc­tory of capabilities, reputations, revocations? And critically, what’s the TLS-equivalent for intent, the cryp­to­graphic proof that an agent’s request actu­ally rep­re­sents a human’s will?

I say “provo­ca­tion” with some precision: I am both daz­zled and dis­mayed by Tina’s ques­tions. By her whole post, actu­ally. I sup­pose I always want to see a little “why” mixed in with the urgent “what” and “how”. The broad “why” of the AI industry (and affil­i­ated enthusiasts) in early 2026 seems only to be “because we can; because it is upon us”. That’s not an uncommon response to technology, but nat­u­rally I’m inter­ested in going a bit deeper.

Of course, one approach would be to answer the ques­tions Tina is asking.

It’s a few years in the past now, but my expe­ri­ence spec­i­fying and building the pro­tocol I called Spring ’83 remains very instructive. After laboring over this, my first spec, I wrote:

I really strongly rec­om­mend this exer­cise to anyone who feels dis­sat­is­fied with their options on the internet today. Give it a few evenings. Imagine some­thing new; describe it as clearly as you can.

It just occurred to me — truly, just as I was pasting that text — that these days, you could give your fan­tasy spec its first spin by placing it before an AI coding agent and saying, “Implement this.” That’s pretty inter­esting!

Returning to the “why”, it would be exciting to see new pro­tocols for AI sys­tems and agents that aren’t only effi­cient but also deeply opinionated — as opinionated, say, as HTTP. Has there ever been a more per­snickety pro­tocol, actu­ally? In my view, HTTP’s demands over the years have been hugely productive; the ways in which it has NOT worked for every­thing have been … kinda great!

The AI industry is at this moment far more cen­tral­ized than the com­puter industry at the dawn of the internet. That doesn’t bode well for inter­esting pro­tocols … inter­esting topolo­gies of power and access. But I don’t think it’s going to stay this way; certainly, I hope it doesn’t; and good, weird, per­snickety pro­tocols won’t only arise in response to the industry’s transformation — they might assist it.

P.S. I still love RFC 865, the Quote of the Day pro­tocol.