Contents7
“AI automation skills: agents, agentic workflows” now appears in many marketing job ads. Behind the phrase is a simple idea: part of marketing work consists of reading data, turning it into a list of things to do and preparing those things. An AI agent can handle a good part of that chain, provided you break it down for it.
What an agent is, and what it is not, is explained in AI agent: what it is, and what it is not. For the overall approach in a company, see Integrating AI in your business. Here, we stay concrete: four examples, then the rules.
In short
A workflow follows a path set in advance; an agent chooses its steps to reach a goal, with the tools you give it.
The best candidates are repetitive, data-based tasks whose result can be checked.
The agent prepares, a person approves. Nothing goes out to a customer or online without review.
An agent has limited rights: it reads a lot, it writes little.
Example 1: the agent that reads Search Console
From the nightly scan to the edited page
- The scan
Every night, a script reads Search Console and spots slipping pages and titles to fix.
- The agent
For each page, it prepares a suggested edit, with the reason.
- The review
A person accepts, corrects or rejects.
- Publication
Only after approval.
From the field
This is how our own website runs
On brainplusai.com, this scan has run every night since 25 August 2026, with no AI for the reading part, which costs nothing to run. Agents only come in afterwards, to prepare drafts, and every draft goes through a review before it is published.
Example 2: the workflow that sorts incoming contacts
A website form lands in the CRM. A workflow, built with an orchestrator such as n8n, takes over:
- it completes the record with the company’s public information;
- an AI model reads the message and classifies it: quote request, job application, partnership, cold pitch;
- the quote request is assigned to the right salesperson, with a two-line summary;
- the rest is filed, without disturbing anyone.
Here, the AI has only one task, classifying a text, and the rest of the path is fixed. That is often the most reliable form. To choose the orchestrator: n8n, Make or Zapier.
Example 3: the monthly report prepared by an agent
The agent gathers the numbers from GA4, Search Console and the CRM, compares them with the previous month and writes a first draft: what changed, and the possible explanations it finds in the data. The person who runs marketing reviews it, discards the wrong leads and writes the decisions. The time spent copying numbers disappears; the time spent thinking remains. See also Automating your reports: the end of Monday copy-paste.
Example 4: editing dozens of pages at once
A product range is renamed and the change touches fifty pages of the website. An agent, connected to the site’s API with read access (on WordPress, with an application password of its own), lists the passages concerned and proposes the replacements. A person approves the list, then the changes are applied. This is where coding skills make the difference: writing this small tool takes less time than the fifty pages by hand.
What you keep
What you hand to an agent
Reading and cross-checking data every day
Preparing drafts, lists, summaries
Classifying and routing requests
Applying approved changes
What you keep
The choice of priorities
The brand’s tone and what you promise
Approving everything that is published or sent
The relationship with customers
Watch out
An agent that publishes alone ends up publishing a mistake
AI models sometimes invent facts with confidence. A client learned this the hard way, when content that matched nothing real went live and had to be taken down. Our rule comes from there: review, source, sign, as we explain in Content written with AI: review, source, sign.
Frequently asked questions
Do you need to know how to code to build an AI agent?
Not for a first workflow: an orchestrator such as n8n is configured without writing code. Code becomes useful to connect a tool that has no connector, to write a small custom tool, or to make what runs in production more reliable.
What rights should an agent have?
The minimum: broad reading, narrow writing. An agent that prepares edits does not need the right to publish. Each agent has its own access, which makes it possible to know who did what and to cut one access without breaking everything.
What does an agent running every day cost?
Less than people fear when the data reading is done without AI and the model only steps in to write. The cost depends on the model chosen and the volume of text processed; we cover it in the AI at the right cost offer below.
What we can take on
- Workflows between your tools, with or without AI: n8n automation.
- The right model for each task, at a controlled cost: AI at the right cost.
- Agents and automations added project after project, in order of priority: transformation retainer.
- The SEO engine described above, applied to your website: SEO and GEO.
Sources
- WordPress, Application Passwords: Integration Guide (new tab)make.wordpress.org
Jim · founder of Brain Plus AI
Business AI
Want the same thing running at your place?
Our own expense-entry agent has run on Telegram in production since July 2026: photo of the receipt, extraction, human validation, write to the accounting system.
Also in business ai
Understand9 min
What an AI agent is, and what it is not
An AI agent chooses its own steps and tools to reach a goal. That is useful on open-ended tasks, but it is neither a relabelled chatbot nor the default answer to every automation.
Understand10 min
Agentic AI: how an agent's loop works
An AI agent does not do everything in one go: it runs in a loop, gathers context, acts, checks and starts again. Understanding that loop means knowing where to put the controls, the limits and the human review.
Practical guide8 min
AI-written content: review, source, sign
An AI model writes fast, but what goes out in the company's name commits it. Three steps make a generated text publishable: a review that checks facts and promises, a source for every number, and a person who signs.