Guides
17 posts on guides for AI agents that send and receive email.
6 August 2026 · 4 min read
Writing documentation an agent can actually answer from
Your agent answers from your help centre, which was written for people who are already reading it. What to change so retrieval finds the right passage and the answer is groundable.
6 August 2026 · 5 min read
The vocabulary of agent email, defined
SPF, DKIM, DMARC, MCP, threading, cold sends, suppression, idempotency. Every term that appears when you put an agent on an inbox, with what it means and why it matters.
5 August 2026 · 5 min read
Making an agent answer only what it can prove
The expensive failure is not a wrong tone, it is a fluent invented fact a customer believes. How to require a source for every claim, and what to do when there is not one.
4 August 2026 · 5 min read
When an email agent is the wrong answer
Agents are a poor fit for several common situations, and finding out after launch is expensive. The cases where the answer is a form, a phone call, better documentation, or nothing.
3 August 2026 · 4 min read
Subject lines an agent should write, and the ones it should leave alone
The subject is the one line that decides whether anything else gets read, and the one field agents most often break. When to keep it, when to write it, and what makes a good one.
3 August 2026 · 4 min read
What goes at the bottom of an agent's email
The signature is where disclosure, the route to a human, and your legal obligations live. What belongs there, what does not, and why an agent's footer differs from a person's.
1 August 2026 · 5 min read
Answering email in the language it arrived in
Language is the cheapest coverage win an email agent offers, and the easiest place to embarrass yourself. Detection, register, what to never translate, and when to escalate to a person who speaks it.
1 August 2026 · 5 min read
When an agent and a person are on the same thread
Cc, reply-all, and handover are where agents embarrass you in front of colleagues and customers. The rules that keep a shared thread readable, and who should be speaking when.
31 July 2026 · 4 min read
Keeping an agent inbox usable after ten thousand messages
An inbox nobody prunes becomes a list nobody can use. How agents should search, what archiving is actually for, and why unread is a signal rather than a status.
30 July 2026 · 5 min read
Deciding what reaches your agent in the first place
Whitelist or blacklist, exact-address rules that beat domain rules, and how to route one category of real mail to an agent without redirecting your whole inbox.
29 July 2026 · 5 min read
Making your product readable by the agents that will recommend it
Agents now evaluate and integrate products by reading documentation. What llms.txt is, why serving markdown beside HTML matters, and how to structure docs an agent can act on.
27 July 2026 · 5 min read
Attachments: what an agent should do with the file in the email
Invoices, contracts, and CVs arrive as attachments. How agents receive and send files over email, what to extract before the model reads, and the safety rules that keep a file from becoming an exploit.
27 July 2026 · 5 min read
Giving an email agent memory of the people it talks to
The difference between an agent that feels competent and one that feels robotic is memory. How to use contacts, recorded facts, and notes so the second conversation starts where the first ended.
26 July 2026 · 5 min read
The system prompt for an email agent, annotated
A complete, copyable system prompt for an agent with its own inbox, with the reasoning behind every section: identity, thread discipline, escalation, refusal handling, and tone.
24 July 2026 · 6 min read
Agent inbox or shared mailbox: which one should your agent use
Four ways to give an AI agent email access: shared IMAP credentials, a forwarding alias, the Gmail or Graph API, and a dedicated agent identity. What each costs in scope, threading, and oversight.
15 July 2026 · 6 min read
What an MCP email server is, and how to connect one
An MCP email server exposes send, read, thread, and wait tools over Model Context Protocol. What the tool surface looks like, how scoping and rate limits work, and how to connect Claude, GPT, and other clients.
8 July 2026 · 8 min read
How to give an AI agent its own email address
What it takes to put an AI agent on a real email address: identity, threading, waiting for replies, memory between runs, deliverability, and the oversight that keeps it from becoming a spam problem.