10 AI agents that work on my marketing every week

How to read these 10 examples

These agents run in my consulting business, on client work and on my own brand. Each one rests on the 5 components described in What is an AI agent: a personality, skills, tools, context and rules.

I've grouped them by the point in the week when they run. For each one, I give what it does, what it replaced in the way I work, and the limit I keep in mind. That last line matters most: an agent whose limit you don't know ends up costing you one review too many, or a published mistake.

Producing content

1. The monitoring agent

Every week, it cross-checks the news, weak signals, academic research and discussions on Reddit and X, over the current week and the one before. It keeps a monitoring log so it never serves me an angle I've already covered, and it returns a list of post angles, each one sourced and dated.

What it replaced: the press review done by hand, tab after tab. Its limit: it suggests and I choose. The right angle for my audience remains an editorial decision. The full method is in Automate your marketing monitoring with Claude.

2. The LinkedIn writer

It writes in my voice, calibrated on an analysis of my last 100 posts and on my editorial identity. Before writing, it asks me for raw material: something I lived through or a number. Then it suggests 5 angles and assembles the one I pick.

What it replaced: the blank page on Monday morning. Its limit: without real material, it produces nothing good, and that's by design.

A copy stripped of my data is public, ready to install on your side: create your LinkedIn agent with Claude Code.

3. The adaptation agent

From a source piece, a video or a post, it prepares the LinkedIn, TikTok and Instagram versions: the hook, the caption, the hashtags, the call to action and the export settings specific to each network.

What it replaced: copying and pasting from one network to the next, with the same hook everywhere. Its limit: it refuses to republish a video that carries another network's watermark. Native formats still have to be shot.

Checking before publishing

The next 3 agents review what the others produce. They return a verdict without changing a single file, and I make the call.

4. The brand reviewer

It runs every piece of content through 6 checks: AI writing tics, banned vocabulary, voice, numerical evidence, positioning and visuals. In July, on a client report I thought was clean, it flagged about 60 issues, including every number written out in words.

What it replaced: my end-of-day proofreading, when I can no longer see my own tics. Its limit: it checks whether the rules are followed, and the emotion of a text is still for my ear to judge.

5. The English reviewer

My site exists in English, page for page. After each adaptation, a native American English reviewer compares the page with its French source, quotes each correction, checks the facts and returns a verdict.

What it makes possible: keeping a complete English version up to date without waiting for a human reviewer at every edit. Its limit: a cultural reference still needs checking with someone from the country.

6. The video reviewer

Before any video is delivered, it measures the render and checks the film against the brief, the brand, each network's safe zones, the subtitles and the rights. It ranks its findings on 3 levels of severity, with the exact timestamp and the suggested fix.

What it replaced: watching on a loop, where an overflowing subtitle goes unnoticed. Its limit: it sees what can be measured, and the rhythm of an edit is still judged by eye. The video studio it checks is described in Brand motion design with Claude Code.

Selling and delivering

7. The sales director

After a meeting with a prospect, I give it the transcript. It asks me its questions, then works out the offer, the price, the angle and the answers to objections with me, in a closing file.

What it replaced: notes scattered after a meeting, and a proposal written days later. Its limit: the final price is my decision.

8. The online proposal agent

From the closing file, it builds an online proposal on jessem.fr, behind an access code. The prospect reads it, signs the quote and picks time slots from the page, and I get an email notification.

What it replaced: the PDF sent as an attachment, then the follow-up to find out whether it had been opened. Its limit: the official quote remains the one from my invoicing software, and the page stamps its PDF.

9. The workshop report agent

After a workshop at a client's office, it turns the transcript into a web report in the client's colors: a summary, an audio recap, interactive modules and an AI assistant that answers questions about the session's content. For my training courses, a twin agent maintains each cohort's space, with the recap, the homework, the resources and the replay.

What it replaced: the Word report, read once and then forgotten. Its limit: every recommendation is checked against the recording before it goes out, because an agent that summarizes can harden what someone said.

In the background

10. The email agent

My inbox holds about 4,000 unread messages, and each week's handful of real messages drowns among the newsletters. Twice every business day, a scheduled task launches the agent. It spots the replies I'm waiting for and the follow-ups to send (on day 3, day 7 and then day 14 for a prospect), then prepares drafts in Gmail along with a summary.

What it replaced: the fear of having left a client without an answer. Its limit: it never sends anything, and the drafts wait for my review.

The full procedure and its guardrails are detailed in Claude scheduled tasks sort my inbox twice a day.

The 10 AI agents for marketing at a glance

With my clients

The teams I work with build theirs on the same model, starting from their own folder structure.

I built a folder structure, and Claude turned it into a knowledge structure with agents at every level.

Thibault Winka

Managing director, Team Officine

At Horizon Humain, the communications officer handed meeting summaries over to an agent: Claude picks up the transcript, writes the summary in the company's brand style and sends it to the whole team.

Which one to build first

Pick a task that comes back every week and whose result you can judge in a few minutes. If the agent gets it wrong, the mistake has to be fixable without damage. Monitoring and proofreading meet these conditions; email and sales proposals come later, once your rules have proven themselves.

For each agent, write down its limit before its mission. That limit tells you which rule needs to move into code, with a hook: the method is in Controlling your AI agents with Claude Code hooks.