AI Email Workflow in Mailbase
A product guide to in-context AI drafting, the assistant's task queue, and proposal-first inbox triage.
Overview
AI in Mailbase writes into whatever you are looking at. Ask for a reply and the draft lands in the reply box; ask for campaign copy and it lands in the composer. It is ordinary editable text, and the human still presses send.
Drafting In Place
Requests with somewhere to put an answer are answered there: a reply draft in the reply box, a subject and body in the composer, a rewrite of an existing draft that leaves your own subject alone. Requests with nowhere to put an answer — an open question, research on a contact, a deliverability review — open the assistant with the context already filled in.
Every surface resolves the same model, so a workspace that connects its own provider gets one voice everywhere, and the workspace's AI policy (tone, signature, pricing rules, forbidden claims) travels with every prompt.
- Reply drafts in the reply box
- Campaign copy in the composer
- One model across every AI surface
- Workspace tone and signature applied
- No automatic send on completion
Parked Work And Proposals
Work you are not ready to do can be parked. Parked tasks show up in the assistant, where you can run one or work the whole list — it is a turn in the conversation, not a hand-off to another tool.
Inbound mail is classified as it arrives: an intent, a suggested next move, and tags proposed from the tags your workspace already created. Proposals are stored apart from applied tags and render as dashed chips beside the real ones, so a guess is never mistakable for a decision. Rejecting one stops it being proposed again.
- Parked tasks run by the assistant
- Tags proposed, never applied
- Proposals stored apart from facts
- Rejections remembered
- Workspace AI policy
Common Mistakes
- Choosing a tool before deciding who owns deliverability.
- Treating DNS authentication as a one-time checkbox instead of an operating baseline.
- Mixing product-critical transactional email with experimental marketing sends, with no clear boundary.
- Trusting headline metrics (like open rate) that privacy proxies now inflate.
Related guides
More on ai email workflow and the surrounding product workflow:
FAQ
Can the AI send an email by itself?
No. Every draft lands in an editable field and goes out through the normal send path, which a person triggers.
Can it invent new tags?
No. It may only propose tags the workspace already created, and a proposal is not applied until someone accepts it.
Which model does it use?
The workspace's own — the platform model by default, or your provider if you connect one.