Marketing Agents · five responsibilities · one operator

Automate repeatable work. Keep decisions with people.

Five agents own five operating responsibilities — part of an 18-agent workspace.

The operator coordinates the queue. Each agent prepares and checks one kind of work. People approve strategy, claims, exceptions, and publishing.


Human approval

  • ChatGPT
  • Claude
  • Perplexity
  • Gemini
  • Grok
  • DeepSeek
  • Kimi
  • Google AIO
  • Copilot

Operating contract

Responsibility first. Authority second.

Measured on our own workflow.

We used this responsibility-and-proof model to prepare and check this product update on 22 July 2026. The run stopped when a content gate failed, then resumed only after a repair.

  1. Name the responsibility. Every agent has one buyer-readable job and a boundary that another layer cannot silently cross.
  2. Provide approved context. Add the sources, positioning, claims policy, destination, and acceptance criteria before generation begins.
  3. Stop at the human gate. Strategy, new claims, exceptions, merge, deploy, and publishing remain explicit decisions.
  4. Prove the destination. The run ends only when a person can inspect the intended artifact or live behavior.

Agent 01 · measures · owns visibility analysis

Hermes Visibility

Comparable AI-visibility observations and governed opportunities.

Hermes preserves comparable observations and keeps mentions distinct from citations. AI Visibility Intelligence normalizes the evidence, ranks opportunities, and prepares the next governed handoff.

Open Hermes Visibility →

Agent 02 · produces · owns content production and quality gates

ContentOS

governed content packets with visible evidence.

ContentOS carries the prompt map, source boundary, draft, metadata, schema, checks, repair history, and human approval state together. Publishing authority stays in a separate product layer.

Open ContentOS →

Agent 03 · publishes · owns canonical publishing and distribution

Content Publisher

canonical-first publishing and bounded distribution.

Content Publisher preflights the destination, runs guarded publication, proves the live route, and keeps every approved channel adaptation connected by a visible source link.

Open Content Publisher →

Agent 04 · optimizes · owns technical website readiness and verification

Website Agentic Optimization

project-bound website changes with live proof.

The workflow can repair discovery, metadata, schema, links, layout, and performance within an approved scope. Passing CI remains human-review-required, and technical proof does not become a claim of causal attribution.

Open Website Agentic Optimization →

Agent 05 · designs · owns repeatable visual asset delivery

Visual Asset Studio

placement-specific visual assets with pixel proof.

Visual Asset Studio renders against an exact placement and verifies the final pixels before handoff. It does not invent a brand system or campaign concept when the brief is incomplete.

Open Visual Asset Studio →

13 more workspace agents run evidence gathering, client packaging, the control plane, and ops adapters. See the full platform →


One operating loop

Each handoff has a named artifact.

Observe

Visibility evidence

Hermes stores prompt-and-engine observations and keeps mentions distinct from citations.

Prepare

Governed content packet

ContentOS carries the brief, source boundary, draft, checks, and approval state.

Render

Placement-specific asset

Visual Asset Studio verifies the actual output against its destination and design pack.

Ship

Canonical publication proof

Content Publisher ships the owned asset first and records the guarded rebuild and live route.

Change

Project-bound website packet

Website Agentic prepares a scoped change and observes checks for the exact commit.

Decide

Human approval record

The operator owns exceptions and keeps a visible record of what was accepted, rejected, or deferred.


10× article throughput: 100–120/mo with one operator + 18 AI agents vs 10–12/mo from a 6-person freelance team.

Compared with a 6-person freelance/content team at 10–12 articles/month; one Marketing Operator + 18 AI agents at 100–120 articles/month.


Marketing Agents FAQ

What the agents do—and where they stop.

What are Marketing Agents?

They are five client-facing product surfaces that own repeatable responsibilities in an operated marketing loop: visibility analysis, content production, canonical publishing, website readiness, and visual delivery.

Do Marketing Agents replace people?

No. They automate repeatable work. Humans retain decision authority for strategy, claims, budgets, approvals, exceptions, publishing, and final judgment.

Who operates the agents?

A Marketing Operator coordinates the queue, evidence, approvals, and exceptions. In the operated pilot, the role is designed to transfer to a client employee from the second month.

Are the agents fully autonomous?

No. Each agent has a bounded responsibility and explicit stop points. Work that lacks evidence, authority, or proof returns to a person instead of silently shipping.

How do the agents work together?

Hermes identifies an opportunity, ContentOS prepares the governed packet, Visual Asset Studio produces placement-specific assets, Content Publisher ships the canonical work, and Website Agentic Optimization prepares and verifies site changes.

How do supporting products fit in?

Parsing Stack gathers bounded evidence. Content Package Studio packages accepted content for delivery. They support the five Marketing Agents without becoming extra operator roles. The full workspace runs 18 agents in all — marketing, support, control plane, and ops.


Choose the first responsibility. Build the operating habit.

The pilot includes the platform and the operator.

The first month establishes the queue, gates, and proof. From month two, the operator role can transfer to a client employee.