GuidesAI and agents
An AI agent that prepares and publishes social media posts
The flow
- 1SourceArticle, product news, case study.
- 2AdaptationOne text per network, with its length and conventions.
- 3ApprovalA person reads, edits, approves.
- 4SchedulingPosted through the API, at the chosen time.
One tone per network
| Network | What generally works |
|---|---|
| A concrete lesson, a few short paragraphs | |
| The visual first, a short caption | |
| X | One idea, one sentence, one link |
The guide, in a few lines
Tone: direct, concrete, no superlatives.
Always: a verifiable fact or figure taken from the source.
Never: promised results, an emoji to open a sentence, topics outside the source.
Length: LinkedIn 600 to 1,200 characters, X under 280.
Safeguards
- The agent only uses the source provided, never its own memory.
- No automatic replies to comments without review.
- A log of what was published, by whom, and approved by whom.
The principle: AI agents on real data. Choosing the tool: n8n, Make or custom code.
Frequently asked questions
- Do networks allow automated posting?
- Through their official APIs, yes, with conditions: an approved app on LinkedIn, a professional account for Instagram, paid access to the X API depending on usage. A scheduling tool can act as the go-between.
- Why keep human approval?
- A published post commits the brand in public. A thirty-second review avoids the wrong tone or a false figure.
- How do we keep our style?
- With a written guide (tone, formality, banned words, length) and ten of your best posts as examples.