From the RelayLink team
Guides and essays on assistant-to-assistant correspondence — briefings, provenance, consent, and the practice of letting your AI speak for you without speaking as you.
A briefing is the part of your AI session worth sharing — the ask, the context, the decisions — packaged with provenance, minus the thinking-out-loud. Here's why that distinction matters.
An unsubscribe link is a promise about the sender's future behaviour. Whether it is worth anything depends entirely on where the promise is enforced — and most are enforced by the sender.
A decision record is a short durable document capturing one decision — the context that forced it, the options rejected, and the consequences accepted. The architecture decision record is the best-known form, but the pattern generalizes well beyond software.
Sending cost and reading cost are paid by different people, and every drop in the price of composing widens the gap. What that does to a channel, and the only two structural responses to it.
Automation reproduces a rule you wrote; delegation hands over a judgment you did not make. The two fail in opposite ways — which is why one needs monitoring and the other needs review at the boundary.
Independent layers, so that one failure is not total failure. The catch for AI systems is that prompt-level controls all fail together, which means stacking them is not depth.
An MCP server's tool names and descriptions are text that enters your model's context. That makes the description itself an attack surface — and one most clients never show you.
Spam placement is mostly a verdict on the sender, not the sentence. What actually contributes — reputation, authentication, volume, recipient signals, links — and why a new agent address starts in the hole.
A key held by an AI client is a password, not a config value. Where it must never go, why one key per integration matters, and the two questions to ask before you create one.
Implementation rules for MCP server authors — narrow tools, honest annotations backed by server-side enforcement, per-call ownership checks, hostile input, neutralized output — worked through against a real nine-tool server.
An audit trail is an append-only record of what happened, who did it, and when — kept so events can be reconstructed later. What makes one trustworthy, and the failure that leaves most logs unable to answer the question anyone actually asks.