Twenty-one posts a week for fifteen dollars a month.

The problem

A solo coaching practice lives and dies on visibility, and social media visibility runs on volume and consistency. Producing three posts a day with matching images, in a recognizable voice, would take a solo operator something like twelve hours a week. What actually happens to most solo businesses is simpler: they post for three enthusiastic weeks, then go quiet for two months.

What we built

A pipeline that runs the whole show. It generates the posts inside the brand's themes and voice, generates an image that matches each post, and schedules everything through Buffer. Built and orchestrated with Claude Code; images by Google's Nano Banana after experiments with several approaches, including locally run models.

One human step survives, on purpose: a review pass over the queue before it uploads. The machine has never needed a rescue, but content that publishes under a person's name gets a person's eyes first.

The part worth noticing

Let's be precise about what this system can and cannot do, because nobody selling AI content ever is. It will not write Kahlil Gibran. The ceiling on generated introspection is real, and if someone tells you otherwise, hold your wallet tighter.

But it also doesn't produce slop, and that's the engineering. AI content has a floor and a ceiling; the ceiling gets the demos, the floor is what actually kills small business social media. Off-brand filler, inconsistent voice, three silent weeks in November. This system is built for the floor: every post on-theme, on-voice, on schedule, every week, for the cost of a sandwich a month.

The results

Five months of twenty-one posts a week, roughly fifteen dollars a month in tokens, and a review pass as the only labor. The content holds the brand's voice well enough that tuning attention has dropped to nearly nothing.

And the honest part: this is a small account, around a hundred followers and growing. The system solves production, not popularity. Growth still depends on the things growth always depended on. What the machine guarantees is that the presence is always there, always on-brand, and costs almost nothing to keep alive, which is precisely the part solo operators fail at.

What didn't go smoothly

Images were the hard part. Several approaches went in the bin, including locally run models, before Google's Nano Banana with a strict stylistic system and a set of agents that align each image with its post, the brand, and the themes. Words and pictures come from different vendors because each is the better tool for its half; we resell neither.

The content also drifted early. The first weeks wandered off topic often enough to need real tuning of themes and guardrails. That work paid off; drift is now rare. Plan for a tuning period in any content system, because the demo never mentions one.

The pattern you can borrow

If your business needs to be visible but content is the first thing you drop when work gets busy, this is the shape of the fix: a generation pipeline tuned to your voice, a human skim before anything ships, and honest expectations about ceiling and floor. It's the same automation discipline we apply to intake and invoices, pointed at your marketing.