An agentic brand system, for marketing.
Marketing is where most companies start with agents, because it is where the volume is. It is also where a brand drifts fastest once they arrive.
A campaign is not one asset. It is a landing page, a run of emails, a set of ad variants in every size and a week of social posts, and an agent can draft all of it before lunch.
Whether the result looks like one company depends on what the agent read first. An agentic brand system is the answer to that question, written down for marketing.
1. What changes when agents do the making
The old flow ran from a brief to a designer and a writer, then through rounds of review until the pieces agreed with each other. Most of the review was catching drift between pieces, not judging the idea.
With agents, the brief goes to an agent that makes the whole set from the same record. Review moves from each piece to the set, and from catching drift to deciding whether the campaign is right.
That only works if the record is complete enough that the agent is not guessing. Where it guesses, the drift comes back, faster and at a larger volume.
2. What the system holds for marketing
An agentic brand system is one record: a shared layer every team reads, and a layer for each department on top of it. Marketing’s layer holds three kinds of thing, and they are not interchangeable.
- Tokens are settings, with one right answer: the live tagline and the retired ones, UTM conventions, channel limits, hashtags and handles. An agent that gets a token wrong is simply wrong.
- Patterns are shapes: how this brand builds a launch email, a social post, an ad, a landing page or a subject line. Without them an agent writes something correct and generic.
- Facts are claims with something behind them: stats with sources and dates, customer quotes with permission, feature claims, competitor comparisons and launch dates. What makes them facts is the source, so an agent repeats one only with it.
What marketing agents read, and one run on it
Diagram- Shared
- Marketing
- Sales
- Support
- Product
- Creative
- taglines, live and retired
- UTM conventions
- channel limits
- hashtags and handles
- launch and nurture emails
- social posts
- ad copy
- landing and pricing pages
- subject line formulas
- stats, with sources and dates
- customer quotes, with permission
- feature claims
- competitor comparisons
- launch dates
Write the launch email
- Pricing and terms traced to a source
- Terminology matches the record
- Voice checked against voice/voice.md
- No source for the adoption stat
- Under 120 words
- Opens on the approved pricing line
- One call to action, not two
The run underneath is the point. Asked for the launch email, the agent traces the pricing to a source, matches the terminology and checks the voice, and it leaves out the adoption stat because nothing backs it.
3. The agents a marketing team runs
Not every marketing task should go to an agent, and the ones that should do not all carry the same risk. The sensitive ones are those that reach many people at once or cannot be taken back.
Marketing tasks, and who should do them
| Task | Use | Why |
|---|---|---|
| Deciding positioning or a new message | Assistant | It is a judgement about the company; an assistant can show what is already approved, and people decide what changes. |
| Writing a campaign idea | Assistant | The idea is the part that should come from a person; the assistant keeps it inside what the brand can say. |
| Making a campaign set from an approved brief | Agent, then a person | The page, the emails and the ads come from one record; a person reviews the set before it ships. |
| Sending the newsletter | Agent, then a person | It reaches every subscriber at once and cannot be recalled, so a person reads the final send. |
| Resizing a creative into every format | Agent | The rules for each size are written down, and a wrong crop is caught in the set review. |
| Checking a draft against the record | Agent | It flags claims that are not approved, figures past their date and retired wording, before a person reads it. |
| Watching what assistants say about the brand | Agent | It runs each week, compares with the last run and flags what changed. |
4. What it changes in the work
- Review is of the campaign, not the pieces. The question becomes whether the idea is right, because the pieces already agree.
- Designers set the inputs. They decide the values and the rules once, and stop taking the same request in seven sizes.
- A figure that expires stops being used. Nobody has to remember that the 2024 survey is out of date.
- A change is made once. Update the messaging in the record, and the next set every agent makes uses it.
5. Where it goes wrong
Most failures come from the agent reading the wrong thing. A guideline PDF instead of values, last year’s deck instead of the record, or a pile of everything in the hope that more context helps.
The other failure is skipping the person on the sensitive work. An agent that can send the newsletter on its own will, one day, send the wrong one.
6. Where to start
Pick one campaign type the team makes again and again, such as a product launch or a monthly offer. Write down what it needs from the record, run it with a person reading every piece, and let the agent take more of it as the checks become rules.
Start with the audit. It is free, it is one pass, and you keep what it finds whether or not anything follows it.
