Ask Who Actually Saw It. Change Nothing Without Approval.
Two agents, one Communications app. The help agent is read-only — acknowledgement rosters, the Plan, topic experts. The admin agent drafts and changes messages and campaigns only after you confirm.
The Questions Comms Teams Can't Answer Fast
Every one of these is a question somebody already has the data for — it's just spread across queues, exports, and people's memory. The agent answers them from the records the app already keeps, scoped to what the asker is allowed to see.
"Who Hasn't Acknowledged Yet?"
The policy went out Monday. Compliance wants the pending roster on Thursday. Today that means exporting the broadcast report, filtering it by hand, and cross-checking against the org chart — every single time somebody asks.
"What's Going Out This Week?"
The feed has its own queue, newsletters have theirs, campaigns have theirs. Building one weekly view means several tabs and a manual reconciliation, so back-to-back sends and audience overlaps surface too late to fix.
"Did We Already Publish Something On This?"
Newsletters, publications, and the open ideas backlog each have their own filters. The answer arrives in pieces, if at all — so good content gets re-pitched as new and the same topic goes out twice.
"Will This Even Reach Anyone?"
Reach and acknowledgement rate are known only after the send. Without a forecast, the choice between a must-read and a broadcast is a guess, and the correction costs a second send to the same audience.
"Who Actually Knows About This?"
Routing a question to the right subject-matter expert depends on somebody remembering who that is. The expertise data exists and is scored weekly — it just isn't where people are asking.
Drafting Means Leaving The Conversation
A comms lead describes the message in chat, then opens the composer to write it again from scratch. The agent absorbs the navigation while keeping the approval explicit — one prompt, one preview, one confirmation.
The Communications Agent Pair At A Glance
Communications AI
Acknowledgement lookups, the Plan, and confirmation-gated drafting and campaign changes.
Inside The Help Agent — Read-Only, Every Time
Fourteen tools, none of which can change anything. The help agent looks things up and reports; if you ask it to draft, send, or reschedule, it hands you to the admin assistant rather than doing it.
Acknowledgement Answers, Not Exports
Ask for the acknowledgement state of any broadcast and get the counts and the actual pending roster back in the conversation — the report a compliance reviewer asks for, without the export and the manual filter.
acknowledgment_summary— total, acknowledged, and pending counts for one broadcast.list_pending_acknowledgments— exactly who has not acknowledged yet.list_recent_broadcasts— formal broadcasts across any time range.show_record— a single broadcast by id or title, always cited by both.
The Calendar And The Archive, Searchable
What's scheduled, and what did we already say about this? Both answered from one place, across newsletters, publications, and the open ideas backlog.
list_upcoming_comms_items— everything scheduled on the Plan.search_comms_content— newsletters, publications, and open ideas by topic.find_topic_experts— the subject-matter experts for a topic, from the weekly expertise scoring.
A Reach Forecast Before You Commit
predicted_ack_curve estimates expected reach and acknowledgement rate for a draft before it sends — so the intensity decision is made on a number instead of a hunch, and you don't burn a second send correcting the first.
predicted_ack_curve— expected reach and ack rate for a draft, pre-send.- Audience-aware — the forecast runs against the real targeted audience, not a generic average.
Inside The Admin Agent — Every Write Gated
The admin assistant reads audiences and campaigns, and its six write tools are all classified RISKY and routed through the platform's confirmation flow — nothing changes until you confirm. It cannot publish, send, schedule or approve.
Draft A Message — It Does Not Publish
draft_communication creates a draft — a post, an announcement, a must-read or a broadcast, to the audience you name — and stops there. The confirmation step shows a preview before the draft even exists, and publishing, sending and scheduling remain human actions in the app.
- Draft only — the tool has no path to publish, send or schedule.
- Preview first — the confirmation shows what will be created.
Placements And Campaigns, On Your Say-So
Take a message off Top of Home, a store's home or a work screen early, put a message in a campaign or take it out, and choose which triggers' automated messages a campaign follows — each one asks you first.
end_placement— ends one place a sent message appears, before its end date.put_message_in_campaign/take_message_out_of_campaign— what a campaign holds.follow_trigger_in_campaign/stop_following_trigger— which automated messages join a campaign.campaign_detail— a campaign's goals, its messages and the triggers it follows (read-only).
Where The Agent Stops
Emergency alerts, muster, and "who's safe" questions are not this agent's job — they belong to the Safety Hub assistant, and the Communications agent says so rather than guessing. Draft and scheduled broadcasts, and the "who hasn't acknowledged" roster, are visible only to broadcast authors, managers, and admins; everyone else sees only published broadcasts they authored or received.
- Safety Hub boundary — life-safety response is answered there, not here.
- Permission-scoped — the agent inherits the app's visibility rules, per user.
- Never fabricates counts — it returns tool output or says it found nothing.
Why A Pair Instead Of One Agent
Splitting help from admin is what lets the read-only assistant stay open to everyone while the tools that change things stay controlled.
Different audiences
Any employee can ask the help agent about a broadcast they received. Only admins reach the tools that draft and change things.
Different risk
Read tools need no confirmation and can answer instantly. Write tools are classified RISKY and always route through an approval step.
Different failure modes
A wrong read is a wrong answer you can correct. A wrong write reaches employees. The separation keeps the second one behind a human.
One app underneath
Both agents run against the same Communications records and audiences — the split is in permission, not in data.
Governed Like Every Other MangoApps Agent
Permission-aware by construction
Every tool runs as the asking user. The agent cannot surface a broadcast, roster, or draft that user could not open in the app.
Audit-logged
Every agent request is logged with the asking user in the AI request log, which keeps detail for 30 days, so a governance review can see who asked the agent what.
Admin opt-in
Each agent has its own switch in the Communications app's settings — the help agent's on by default, the admin agent's off until an admin turns it on — a single toggle each, per tenant.
Cited, never invented
Broadcast references always carry record id and title, and the agent returns tool output rather than estimating counts.
“We had a pretty aggressive educational campaign. We shared articles and emails, put up posters with QR codes, and hosted info sessions. And we held a naming contest. Hundreds of employees voted, which gave them a sense of ownership from day one. That early engagement went a long way.”
Jeffrey Straight
Public Relations Specialist
Santee Cooper
Customer Success
See How Real Teams Transformed Communication
Customer Case Studies
Customer Case Studies
Customer Case Studies
Video Case Studies
Customer Case Studies
Customer Case Studies
Frequently Asked Questions
They are one agent pair now, matching the merged Communications app. Rather than four agents each holding a slice of the same data, there is a read-only help agent and a confirmation-gated admin agent over the whole channel — so "who acknowledged this?" and "what's scheduled?" are answered from one place instead of three.
No. The admin agent can draft a message, take a message off a place early, and change what is in a campaign or which triggers a campaign follows — each only after a confirmation that shows what will happen. Publishing, sending and scheduling are separate human actions in the app. The help agent is read-only and cannot write at all.
No. Draft and scheduled broadcasts, and the "who hasn't acknowledged" roster, are limited to broadcast authors, managers, and admins. Everyone else sees only published broadcasts they authored or received. The agent inherits those rules per user rather than applying its own.
Those belong to Safety Hub and its assistant, which owns life-safety response end to end. Ask the Communications agent a muster question and it will tell you that rather than guessing at an answer.
Yes. The agent runs on the same translation layer as the rest of the platform, and drafts can be written in your organization's trained brand voices and translated into each recipient's language (39 languages) before an admin reviews and sends.
Let's Talk
Every team. Every employee. Every workflow. One AI platform — built for the frontline. Since 2008, trusted by 2 million+ users with an NPS of 78.
Why Choose Us?
- Frontline AI: Governed AI for every employee and workflow.
- Top Security: HITRUST, ISO & SOC 2 certified.
- Exceptional UX: Delightful on mobile and desktop.
- Proven Results: 98% customer retention rate.
Rolled out to every employee at AutoZone (125,000), PetSmart (50,000+), A.S. Watson and Raley's (20,000) — and at larger retailers we are not permitted to name.
Prefer to explore first? Ask AI about Communications AI Agent →