GTM engineering: three email workflows you can build in an afternoon
Reply-based outbound, an SDR on deal threads, and a signal inbox that writes an enriched row after Ask — Clay as a table, not a claimed native column.
Discuss this post in AI
Send a pre-filled prompt to ChatGPT, Claude, Gemini, or Perplexity — get a summary, ask follow-ups, or compare ideas from this guide.
GTM engineering is often a pile of syncs: a sequencer, a spreadsheet, a CRM, and a person who pastes. Email is still where buyers answer. If you give a role a real mailbox, three workflows fit in an afternoon without pretending you shipped a native marketplace integration.
Agent Inbox is the mailbox API. Neuro OS is where the role runs, with Ask on every send and every CRM or table write. Clay, HubSpot, amoCRM, or a warehouse table are destinations you call over HTTP after approval — not claimed native columns inside those products.
Workflow 1: reply-based outbound
Stop blasting a new thread for every step. The agent owns an inbox, sends a first message after Ask, and waits. When the buyer replies, the role loads the thread, classifies intent (interested, not now, wrong person, out of office), and drafts the next message in the same conversation.
Implementation sketch: create sdr-west on agentinbox.space, store the inbox id on the sequence row, send only through Agent Inbox so threading holds. A human approves the first outbound and any later send that is not a trivial acknowledgment. Bounce and unsubscribe handling is classification plus a CRM patch on Ask, not a silent delete.
This workflow fails if you “save time” by auto-sending the second touch. The second touch is where tone and facts go wrong. Keep Ask.
Workflow 2: SDR on deal threads
Deals already have email. Forward those threads into a role inbox, or CC the role address when a human starts a conversation. The SDR agent does not invent new prospects here. It watches the deal thread, extracts next steps, drafts a CRM note, and drafts a bump only when the last inbound is older than the playbook threshold.
Scope the inbox to one pipeline or one segment. A global SDR mailbox becomes a junk drawer. When the buyer asks for legal or pricing exceptions, the agent forwards into Neuro OS for a human owner instead of guessing.
CRM writes stay drafted: stage, next meeting, competitor named. The reviewer sees the source message id next to the patch.
Workflow 3: signal inbox
Create an inbox whose only job is to receive forwarded mail: job-change notes, inbound partner intros, product-qualified replies, conference follow-ups. The role turns each message into a row: who, company, signal type, excerpt, suggested owner. After Ask, it writes that row to the GTM table you already use.
If that table is Clay, call Clay’s HTTP API (or a thin internal proxy) with the approved payload. You are writing a row, not claiming a native Clay email column or an official connector listing. The same pattern works for a spreadsheet, Attio, HubSpot, or a Postgres table.
Do not enrich from the email body by scraping random websites in the same turn as the send. Split the work: ingest and classify first; enrich as a second role with its own budget and Ask on writes.
Shared rules for all three
Use one inbox per workflow at the start. Subscribe to inbound rather than polling. Never put sequencer SMTP credentials in a skill. Sends and table writes default to Ask. Forwarding is how mail enters the system; threading is how it stays coherent.
An afternoon is enough to ship the skeleton: inbox, webhook or poll, draft, Ask, write. Quality comes from the playbook you put in git and the reviewers you staff. Measure accepted drafts, not messages generated.
Agent Inbox gives each role a mailbox people can reply to, with forwarding into Neuro OS when a human must see the thread. Outbound mail defaults to Ask. Run the role on Neuro OS. To scope the first inbox, get started.