Banger Mail

Finally a shared inbox that doesn't feel like a hacked together web app. Loved the scoped AI access, draft stayed put until I hit send which is the right default.

How scoped exactly is the agent access — can it send on its own in some scenarios, or does every outbound always wait on a human click?

Building your own send/receive infrastructure instead of layering on SES or Postmark is the choice most teams talk themselves out of because the ROI takes years to show. But it's the one that separates products that stay flexible from ones that hit ceilings the moment their upstream provider decides to change something.The "pull request for email" analogy is the tightest way to state that pattern I've heard. Review-before-send with AI drafts is exactly the missing layer, I've watched teams try to bolt this on with Slack channels, forwarded threads, and shared Google Docs, and it collapses within a month because the review context doesn't survive being ripped out of email.Genuine question, for someone running cold email operations across multiple mailboxes and warmed domains, does Banger's shared inbox model extend to that use case, or is it strictly for shared team inboxes like support@ and sales@? The reason I ask: cold email is one of the places where "one draft, three eyes on it before it goes" would materially reduce misfires, but the current tooling assumes single-operator workflows.

Shared mailboxes where agents triage and draft but a human reviews before send is exactly the setup I'd want for a support@/founder@ inbox — the 'forward the thread into Slack for a second opinion' dance is real. One thing I'd test first: does the review-before-send gate apply to every agent-drafted reply forever, or can you whitelist specific labels/intents (say, shipping-status replies) for auto-send once you trust the triage? And is an agent's scoped access set per-mailbox, or can you restrict it to certain threads/labels within a single inbox?

finally a shared inbox that feels built for mac, the scoped ai access for drafting is genuinely useful and the kanban board for threads makes support way easier to follow

finally a mac app that doesn't feel like a chrome wrapper for shared inboxes. love that ai drafts sit waiting for approval before sending, keeps things sane for support@

how does the AI scoping actually work in practice, like can two agents see different drafts of the same thread or is everything shared across the team

Scoped access + human review before send is the right trust boundary for mixing AI agents into shared inboxes. How granular does the scoping get - can one agent draft-only on support@ while another gets broader access on sales@, or is it set per-mailbox rather than per-agent?

How do the AI agents actually decide when to draft a reply vs just label something, and can I set per-agent rules or do they all run on the same defaults?

Shared inboxes are a good place for AI because the work is repetitive but the accountability is still human. The guardrail I would want is a clean separation between draft, suggested action, and actual send, especially for support or billing threads.

How do you envision the AI agents handling email threads with multiple stakeholders or complex customer support issues?

If two teammates are reviewing the same conversation at the same time, how does Banger prevent duplicate replies or conflicts?

I appreciate that you built your own email infrastructure instead of just adding another interface on top of existing services. That is a much bigger undertaking than most people realize.

The "pull request for email" framing is exactly right. I build support and sales agents for ecommerce stores, andreview-before-send is the single thing that makes owners comfortable letting an agent anywhere near their inbox. So, itis great to see it treated as a first-class feature instead of a bolt-on.The question that I think decides whether this scales: what happens after a reviewer has approved 50 routine drafts ina row and starts rubber-stamping? Do you plan per-agent or per-thread-type trust levels (auto-send the routine stuff,always hold refunds, pricing, anything with money), or sampled review once an agent earns confidence? That dialbetween safety and approval fatigue is the hardest part of this pattern in my experience.Also, respect for owning the mail infrastructure yourselves. Deliverability, bounces and abuse handling are unglamorous, but that layer is usually where products like this live or die.Congrats on the launch, Tiago.

This is very interesting. We're building an AI agent for outreach and encountered this issue. Will explore how it works. Congrats on the launch!