Home Use Cases Work Insights About Contact
Back to Our Work

How a Global Technology Company Built Social Publishing That Backs Every Post with a Fact Sheet

Marketing1
Open the walkthrough

Six-Step AI Workflow with Human Oversight

Each step shows the workspace, what the AI completed, and what still needs human review.

Steps 1, 3 and 5 need a person. Steps 3 and 5 are waiting for approval right now. Click either to open the full walkthrough.

Estimates are directional and based on stated assumptions. All names, organizations, and identifying details have been anonymized in accordance with our confidentiality agreements.

The Transformation

How a Global Technology Company Built Social Publishing That Backs Every Post with a Fact Sheet

Before
A person used to write every post, then rewrite it three more times, once per network.
A person used to log into each network and post by hand, and the gaps between posts were the gaps in their day.
A person used to build this report by hand from four platform exports, and it arrived too late to change anything.
After
the month planned, every slot owned before anything is written
every post written once per network, claims cut where the facts stop
nothing queues wearing the brand until you clear it
every cleared post landing on its slot, on every network
every mention answered, angry threads handed to a person
Executive Summary

Keeps a global brand present on every network without the brand ever saying something it cannot back. Every post carries the fact sheet behind its claims, every reply matches the thread it joins, and the only things that wait on a person are the calendar, the brand voice, and the customer who is actually angry. Six stations, run as a loop: what August measured writes September's calendar.

What Was Broken

A person used to write every post, then rewrite it three more times, once per network
A person used to log into each network and post by hand, and the gaps between posts were the gaps in their day
The real cost
A person used to build this report by hand from four platform exports, and it arrived too late to change anything.

What We Built

Six stations and 12 subagents. Each subagent carries its own tasks and its own refusal.

A1
Slot Planning
Draft the month's calendar: topic, network and time for every slot
A3
Partner Tracking
Track each creator collaboration's deliverables against its agreement
A1
Post Writing
Draft each post from its brief, in the campaign's voice
A3
Asset Pairing
Match each post to the asset its brief calls for
A3
Voice Screening
Check every draft against the brand voice guide, line by line
A3
Disclosure Check
Check the disclosure line on every partner and paid post
A4
Schedule Running
Queue each cleared post into its calendar slot per network
A4
Delivery Watch
Confirm every post is actually live on every network it was sent to
A3
Reply Handling
Answer the common questions from the approved answer library
A4
Sentiment Watch
Score sentiment on every mention, on every network, around the clock
A4
Metrics Assembly
Pull reach, engagement and clicks from all four networks into one record per post
A1
Playbook Tuning
Map which formats, slots and topics carried the month

How It Runs

1

Calendar Planning

A person answers here

Agents draft the month's calendar across every network, attach the partner briefs and rights windows, and leave empty the slots the campaign gives nothing to say. A person owns the month. Nothing gets written until the calendar carries a signature.

2

Post Drafting

Nobody is asked anything here

Agents write each brief once per network, pair it with an asset cropped to every format, and cut any claim the fact sheet does not carry. Nobody sits here. Every draft passes through Brand Review before it wears the brand name.

3

Brand Review

A person answers here

Agents check every draft against the voice guide and the disclosure rules, queue the clean ones, and hold the rest with the exact line each one failed on. This is where a person clears the brand's own voice. Nothing queues wearing the brand name until they do.

4

Post Publishing

Nobody is asked anything here

Agents queue every cleared post into its slot, land it on every network on time, and stagger one story across platforms so it arrives as a sequence. Nobody sits here. A post the platform rejects goes back to Asset Pairing carrying the platform's own reason.

5

Community Engagement

A person answers here

Agents answer every question from the approved answer library, hide spam by the moderation rules, watch sentiment around the clock, and hand up any thread turning hostile with the whole thread attached. A person answers the angry ones. That is the reply a template must never write.

6

Performance Review

Nobody is asked anything here

Agents pull every post's numbers off all four networks into one record, mark what the platforms never returned, and turn what carried the month into proposed changes to next month's calendar. Nobody sits here. The proposals land on next month's calendar, where a person strikes or keeps them at sign-off.

Where a Person Decides

Step 1, Calendar Planning. Strike or keep the 6 empty slots, and move the 2 posts past the rights lapse. Sign the calendar so drafting can start. the month needs an owner.
Step 3, Brand Review. Read the 3 held drafts and clear or reword them. Decide whether the market claim gets a source or gets cut. the brand voice is yours.
Step 5, Community Engagement. Read the two escalated threads and answer in your own words. Tell the support desk about the ticket complaint before you reply. an angry customer gets a person.

Operating Model

This changes how work flows through the team.

Role
Responsibility
Calendar Planning owner
Strike or keep the 6 empty slots, and move the 2 posts past the rights lapse. Sign the calendar so drafting can start. the month needs an owner.
Brand Review owner
Read the 3 held drafts and clear or reword them. Decide whether the market claim gets a source or gets cut. the brand voice is yours.
Community Engagement owner
Read the two escalated threads and answer in your own words. Tell the support desk about the ticket complaint before you reply. an angry customer gets a person.

What Transfers, What Must Be True

What transfers
A person is in the loop wherever the brand speaks in its own name or a customer is upset. Three of the six stations refuse to proceed on their own, Calendar Planning, Brand Review and Community Engagement, and those are the three a person carries. Everywhere else the agents run at volume and reach you only when a platform or a fact sheet stops them.
Calendar Planning stops for a person, and the month needs an owner.
Brand Review stops for a person, and the brand voice is yours.
Community Engagement stops for a person, and an angry customer gets a person.
Every subagent says what it will not do. 12 of them do.
What must be true in your environment
The agent can read the systems your records already live in. This one reads 19.
Somebody owns Calendar Planning and has time for it.
Somebody owns Brand Review and has time for it.
Somebody owns Community Engagement and has time for it.

Failure Modes

What breaks this pattern:

✗ Invented claims leave the drafting station

Without this refusal, the drafting station writes whatever makes the post read well. A product claim with no fact behind it goes out under your brand name.

✗ Lapsed partner rights slip through review

The review station trusts the approved calendar instead of checking the rights window itself. You publish a partner asset after the license ended, and the partner learns it from your feed.

✗ The agent answers the legal complaint

Without the hand-up, a complaint gets a generated reply like any other comment. That reply is now a statement your company made to someone with a grievance.

✗ Estimated numbers enter the monthly report

When the platform API returns nothing, the station fills the gap with a guess. Your team then makes budget decisions on numbers nobody measured.

Directional Outcomes

What the agent counts, and the station that counts it.

These counts are the tallies from one monitored run of the agents. They are not monthly or annual totals.

Posts drafted
Counted at Post Drafting
132
Claims invented
Counted at Post Drafting
0
Briefs handed back as unwritable
Counted at Post Drafting
2
Posts published
Counted at Post Publishing
124
Slots missed
Counted at Post Publishing
0
Held and handed back
Counted at Post Publishing
1
Posts measured
Counted at Performance Review
124
Numbers estimated
Counted at Performance Review
0
Numbers marked missing
Counted at Performance Review
2
Our measurement policy: We do not publish precise ROI without baseline methodology. Every figure above carries its basis.

What Runs Where

Every step names the subagent that does the work, the record it writes, the thing that raises a question for a person, and what it is allowed to touch. This is drawn from the source, not from a diagram somebody kept in sync by hand.

1Calendar Planning
GATE
subagentplan-calendar
writescalendar/2026-08/<slot-id>.json
raisessign-calendar
may touchcalendar/**, partners/** read-only
2Post Drafting
subagentdraft-posts
writesdrafts/<post-slug>/<network>.json
raisesbrief-unwritable
may touchdrafts/**, calendar/** read-only, assets/** read-only
3Brand Review
GATE
subagentscreen-brand
writesqueue/approved/<post-slug>.json
raisesawait-your-clearance
may touchqueue/**, drafts/** read-only, voice/** read-only
4Post Publishing
subagentpublish-posts
writespublished/<post-slug>-<network>.json
raisesplatform-rejected
may touchpublished/**, queue/approved/** read-only
5Community Engagement
GATE
subagentengage-community
writesinbox/<thread-id>.json
raisesescalate-thread
may touchinbox/**, answers/** read-only
6Performance Review
subagentmeasure-performance
writesreports/2026-08.json, playbook/proposals/<change-id>.json
raisesmetric-missing
may touchreports/**, playbook/**, published/** read-only

Stack

Every system this agent reads or writes.

System
Read at
Stations
Instagram
Post Publishing
1 of 6
LinkedIn
Post Publishing
1 of 6
X and YouTube
Post Publishing
1 of 6
and the scheduling queue
Post Publishing
1 of 6
each platform's analytics API
Performance Review
1 of 6
every network's comment and mention feed
Community Engagement
1 of 6
the approvals queue
Brand Review
1 of 6
the approved answer library
Community Engagement
1 of 6
the asset library
Post Drafting
1 of 6
the brand voice guide
Post Drafting
2 of 6
the campaign briefs
Calendar Planning
1 of 6
the content calendar
Calendar Planning
1 of 6
the disclosure rules per platform
Brand Review
1 of 6
the fact sheet library
Post Drafting
1 of 6
the monthly report
Performance Review
1 of 6
the partner agreements register
Calendar Planning
1 of 6
the posting playbook
Performance Review
1 of 6
the support desk
Community Engagement
1 of 6
through each platform's own publishing API
Post Publishing
1 of 6
Next Step

Want to see if this pattern fits your social automation?

No build commitment·Real samples, not a demo·Estimate in writing