Social Media Agent
MarketingGenerates, schedules and publishes content across eight platforms, then reports what each post did.
Wrote and scheduled this week's posts for Instagram, LinkedIn and TikTok — each written for its own platform.
AI agents for digital marketing
that do the work — and show you what they did.
Most AI marketing tools draft a caption and stop there. Nommos runs seven specialist agents that publish posts, manage ad campaigns, build CRM journeys and answer customers — on your accounts, on a schedule, at the level of autonomy you set.
Agent or writing tool?
What an AI marketing agent does that a writing tool does not
Both use language models. The difference is what happens after the text is written — who publishes it, what it is based on, and who can see what happened.
A typical AI writing tool
- Drafts text or images when you ask, then hands them back to you
- Leaves you to copy the result into each channel and publish it yourself
- Usually has no view of your ad accounts, customer records or site behaviour
- Keeps no record of what was eventually published or what it led to
An AI marketing agent in Nommos
- Works on your connected accounts — publishing posts, managing campaigns, building segments and answering messages
- Reads the same customer record as your CRM, advertising and analytics
- Acts only at the autonomy level you set, and holds everything else in an approval queue
- Records every action in the audit trail and reports on results afterwards
A writing tool is the better choice when the job is the words. When the job is running the channel, compare tools by what they execute. Compare AI marketing tools side by side
The roster
Seven specialists, one customer record
Each agent owns a slice of the operation and reads from the same record as the others. Enable them one at a time — a new workspace starts with all of them off.
Advertising Agent
MarketingManages campaigns across five ad networks — creation, creative generation, monitoring and optimisation.
Flagged a campaign pacing to exhaust its budget nine days early, and proposed the adjustment.
Tracking Module Agent
AnalyticsWatches on-site behaviour, builds the funnel picture, and powers heatmaps and session replay.
Identified the checkout step losing the most people this week and linked the matching recordings.
Recommendation Agent
AnalyticsProduces personalised recommendations per customer and reports how they perform.
Ranked what to show each returning visitor based on what they have done before.
Customer Interaction Agent
EngagementHandles inbound messages across social, WhatsApp and Telegram, and escalates when a person is needed.
Answered 22 questions overnight and escalated 3, each with a recorded reason.
Chatbot Agent
EngagementAnswers website visitors from your own documents, and hands over to a person when it should.
Answered a pricing question at 02:14 from your uploaded price list.
CRM Agent
EngagementBuilds segments, broadcasts, journeys and automations from a plain-language request — and waits for approval.
Turned “win back customers who haven't bought in 60 days” into a segment, a template and a draft broadcast.
Where the agents work
Real marketing work, inside the modules your team already uses
The Social Media and Advertising agents publish and manage campaigns. The CRM and Customer Interaction agents work from contacts, journeys and the inbox. The Tracking Module Agent reads what happens on your site. Each acts inside its own module, so its work lands where you would look for it.
How an agent works
Five states, and you can see which one it is in
Automation loses trust when it fails quietly. Every agent reports where it is, and says so when it cannot run.
- EnabledYou switch it on. Nothing is on by default.
- ConfiguredYou connect an account and set what it may do. Missing setup shows as Setup Required rather than failing silently.
- ScheduledPosting times, refresh intervals and notification windows are yours to set.
- ActingIt does the work — publishing, adjusting, replying, building — inside the limits you set.
- ReportingA live activity feed, notifications, a daily brief and an audit-trail entry for every action.
The control model
Autonomy is a setting, not a philosophy
The question is never whether the AI is trustworthy in general. It is what this agent is allowed to do on your account today — and that is a control you hold.
Publish automatically, or approve first
The Social Media Agent has a publishing mode. Auto Publish sends posts at their scheduled time. Preview First holds everything as drafts in an approval queue until you release it. Same agent, same work — you choose which.
Publishing mode
Set per agent. Change it whenever you like.
An approval queue you would actually use
Generated content arrives with its platform, text, hashtags, links, media and scheduled time. Edit any of it in place — including swapping the image — then approve, reject or skip. Promotional items also show the code, the discount, the validity window and the target segment.
Nothing is created without showing you the plan
Ask the CRM Agent for a campaign and it does not go and build one. It returns what it intends to create, with a draft preview of each item and an estimate of the audience size. You choose Create All, Modify Plan or Cancel — and what it creates arrives as drafts. Sending stays a separate, deliberate action.
You describe what you want
“Create a win-back campaign for customers who haven't bought in 60 days.”
It returns a plan
A segment, a template and a broadcast — each previewable, with an estimated audience.
You approve, change or cancel
Nothing exists until you say so. What it creates is a draft, not a send.
See it work
From one instruction to an approved plan
Tell an agent what outcome you need. NOMMOS builds the steps, checks every dependency and shows you the plan before anything happens.
Describe the outcome
Start with a plain-language request instead of manually rebuilding segments, messages and workflow steps.
Review the complete plan
See the proposed audience, channels, content, timing and actions before the agent creates or sends anything.
Resolve missing setup immediately
If an account, permission or configuration is missing, NOMMOS identifies it and takes you directly to the required setup.
Approve before execution
Edit, approve or reject the plan while the audit trail records every decision and agent action.
What gets recorded
Supervision needs something to look at
Letting agents act is only reasonable if you can see what they did afterwards.
Audit trail
Every action is recorded — which matters more when the actor was an agent rather than a person.
Live activity feed
Watch an agent work in real time: initialising, running, completed.
AI Daily Brief
A short summary at the top of your dashboard: what the agents did, what moved, and how many things need you.
Notifications
An unread count and a list you can click straight through to whatever it is about.
Where we are not sure
The product tells you when it does not know
A tool that marks its own uncertainty is easier to trust than one that never does. These are real messages from the product, not marketing positions.
- “Use as guidance, not a guarantee.” — creative scoring, which is labelled experimental and Meta-only
- “This simulation does not send messages, mutate contacts, call webhooks, or create jobs.” — journey testing
- A low-sample-size warning on any heatmap built from too few sessions
- Diagnostics that state they need roughly 500 impressions before they mean anything
- Models labelled as trained or derived, so you know which is a prediction and which is a calculation
- Recommended actions that return “wait” when a customer has not opted in
Consumer trust in visibly AI-generated marketing has been falling, and most teams have no documented process for reviewing what their AI produces. The approval queue is that process, and it ships as part of the product.
Watch an agent work, then switch it off mid-demo
The fastest way to understand the control model is to see the queue fill up and then change the setting that governs it.