Hardcore has been published in Paged Out! zine
I’m tickled to have written a one-page article for Paged Out! zine on the hardcore protocol we developed for our “agentic” systems. I put that in quotes, because although that was the vision from the get-go, there’s something deeply limiting in using that word right now. So I’m reluctant to impose the idea of “agents” as how you should conceive of our system.
For me, the original concept was rooted in easier, simpler communication and data discovery. Every tool out there is business oriented. All of the business layers make it way way harder to get stuff running for our use cases (scientific research.) It also conflates what we do with business “use cases” and exposes us to the same types of security approaches, which is confusing at best, disruptive most of the time, and an entire non-starter at worst.
I just wanted an easier way to program intercommunicating systems. Our “agents” have no autonomy, they just give you an easy way to have them “talk” (share information, messages, and data) to each other. The protocol is specifically intended to be easy to re-implement in other languages, and our reference implementation uses HTTP (you should deploy behind HTTPS), so agents talking looks just like web traffic. Think “user agents” in HTTP, so so so much more than the modern AI agent. The blackboard just gives them a lightweight, easy-to-use shared backend they can churn through data with.
Core/hardcore’s current trust model is to keep everything public (i.e., you must run on a trusted network or understand others can post and read), but that’s obviously not the long-term vision. I envision sort of a semi-public data transit model, where metadata can be read, bytes can be downloaded, but agents (or whatever) must have a decryption key in order to make sense of it. API access can just be understood to not require urgent security, and we can put it on the publicly routed internet. The blackboard can still be put behind a VPN (while VPNs are still legal). Of course this vision is incomplete. I am a medical imaging guy. I feel the squeeze on general-purpose computing acutely and feel like we need to start trying to solve our way through it.
Another part of the vision is ephemerality. The blackboard churns. It forgets old data as new stuff arrives. Eventually the ring-buffer fills and old messages are forgotten entirely. Agents that don’t keep up get left behind. When the blackboard goes down, the memory of that conversation goes with it. I’m a huge advocate of auditing, but the auditing must have a clear scope and vision. The blackboard is supposed to be more like shared RAM than anything long-term. You need auditing? Write a auditing agent that peels off important messages and saves them to disk; it’s not the blackboard’s role in this system.
Networked computing is easier than ever from an infrastructure standpoint, and harder than ever from a regulatory and guidance standpoint. I’m very curious about transparency as a mechanism to upend the current security model a bit, or entirely. We desperately need new applications and approaches that we can use that have good enough security, or give us different guarantees other than just “secrecy” (which is of course, not real secrecy for normal folks since businesses and governments are freely purchasing and sharing your very real, very identifiable information.)
Not everything is e-commerce. Not everything is business. Not everything is spyfare. Big tech is salting the fields at this point trying to make it harder than ever to do anything at all; others (…) see an opportunity to rule over, abuse, and control people through information asymmetry. I’m not here for it.
2026-03-07 - 16:18:03