How email threading actually works
A mail thread is not something the protocol stores. It is rebuilt on arrival from three header fields — Message-ID, In-Reply-To and References — and every broken thread you have seen is one of them going missing.
From the RelayLink team
Blog
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.
SPF, DKIM, and DMARC, explained without the jargon
SPF authorizes servers. DKIM signs messages. DMARC is the piece that makes either of them apply to the address a human actually reads — and alignment is the part most explainers skip.
What is streamable HTTP (and why MCP servers use it)
An MCP server can be a local subprocess or a URL. Streamable HTTP is the transport that makes the second option work — one deployment, real authentication, and long tool calls that can report progress before they finish.
MCP tool annotations explained — readOnly, idempotent, destructive, openWorld
MCP servers can tag each tool with hints about what calling it will do — read-only, idempotent, destructive, open-world. Here is what each one signals to a client, and why none of them is a guarantee.
How to evaluate an AI communication tool
Seven questions to put to any vendor whose product sends messages on your behalf, each with the shape of a good answer and the shape of an evasion. Useful whichever tool you end up choosing.
AI communication failure modes: the six ways it goes wrong
"The AI got it wrong" covers six distinct failures with six different repairs. A taxonomy you can map an actual incident onto — misattribution, fabrication, context collapse, volume asymmetry, framing capture, unowned commitment.
The principal-agent problem, applied to AI assistants
A classic economics problem — you delegate to someone whose choices you can't fully observe — with one ingredient missing when the agent is software, and the other made considerably worse.
What is a context brief?
A context brief is a short passage — roughly 150 to 400 words — that gives a reader with none of your history enough background to answer one specific question, and nothing beyond it. How it differs from a summary, an executive summary, and documentation.
Why AI context doesn't transfer to another person
The understanding you built across a long AI session is not a document you can hand over. It is accumulated state — corrections, abandoned directions, and inferences nobody typed — indexed to you.
What is a magic link? The one-click login pattern, explained
A magic link is a one-time, tokenized URL that authenticates a click instead of a password - common for email confirmations, resets, and one-off access. How it works, and what it quietly trades away.