Hi Javier — saw the "Bring Your Own AI Agent" post and this is right in our lane. I'm building an AI triage assistant on top of Front: it reads a conversation, then drafts replies, summarizes, archives, snoozes, and proposes actions. Today we do exactly the workarounds you mentioned — the agent's writes get attributed to a human account acting as a proxy — so a native agent identity with assignment and @mention support would be a real upgrade for us.
One design point that's central to how we've built it: our agent is strictly human-in-the-loop — it proposes actions (replies, tags, archive, etc.) and a person approves them with one click; it never sends or executes on its own. Does the beta support that propose-then-human-confirms pattern (draft/suggested replies, gated actions), or is it oriented toward agents that act autonomously? That distinction is the thing I'd most want to get right.
Happy to share concrete use cases — we have a working app and real triage flows.
Thanks,
Jim