Build · Content engine

The social posting agent

Every Sunday at 6 AM, an n8n agent drafts the week's "Meet the Staff" posts for Instagram, Facebook, and LinkedIn, lints its own output, and emails me one approve-or-reject decision. On approve, nine posts schedule through Buffer. My Sunday job takes two minutes.

Why it exists

At CoPilot AI I proved a consistent posting system moves booked meetings: 4 to 6 per rep per month became 12 to 16.
Running my own company, I wanted the same compounding engine without spending founder hours on captions.
So the playbook became software: same cadence discipline, same content standards, agent does the drafting, I keep the judgment.

Seven slides of a Meet the Staff comic episode — One episode of Meet the Staff: a comic series where a fictional trades business meets its AI staff. Fifteen episodes, three platforms, scheduled by the agent.

Weekly approval email with full captions per platform — The Sunday email: every caption for every platform, flags from the linter, one Approve all button. Human judgment stays in the loop by design.

Scheduled Instagram post in Buffer — The scheduled result in Buffer: carousel, caption, hashtags, all placed by the workflow.

How I built it

The pipeline. Read the episode schedule from a data table → pick the next three pending episodes → Claude adapts captions per platform → a code pass lints the output → approval email → Buffer schedules on approve.
The brand rules, encoded. No absolutes about the reader's business. No unverifiable claims. Statistics only from an approved bank, sources kept out of the copy. Follow-up promises always bounded by "until they reply or you say stop". Exactly five Instagram hashtags, because more suppresses reach.
Enforced twice. Every rule lives in the prompt and in the linter. A rule enforced in one place is a rule that drifts.
The unglamorous parts. Buffer has no upload endpoint and Google Drive's standard links break its media ingestion, so I found and live-verified the URL form that works. When Buffer dropped all three channel connections one Monday and deleted the queued week, I rebuilt the queue from the approved captions the same day.

What the build taught me: the approval gate is the product. Full automation of brand voice is a trap; an agent that drafts well and a human who approves fast beats both extremes. And platform quirks, Buffer's API shape, Instagram's hashtag penalties, Drive's URL redirects, are where automation projects actually live or die.

Why this matters for selling AI

A rep-built playbook that became a product. Sales insight in, software out.
When I tell an owner an agent can take a recurring job off their plate without taking the judgment away, I am describing my own Sunday.
That maps exactly to what an AI operating system installs inside a client's business.