what this post is. a build-in-public look at one engagement. a 12-person b2b saas marketing team wanted more posts without more headcount. we shipped a content-production system on Claude and Airtable. throughput went up 3x. time-per-post dropped from 6 hours to 90 minutes.
the team was shipping two long-form posts a week. they wanted six. hiring two more writers was the obvious answer, and it was slow and expensive. so they asked a different question. could ai draft the first version without flattening the brand voice they spent three years building.
pre-launch disclosure: the client is anonymized. the numbers below are from the engagement. we don't name them because the contract says not to, and the build is what matters here, not the logo.
#what did we actually build?
a brief-to-draft pipeline. the brief lives in Airtable. each row holds the working title, the target reader, the key points, and the proof to cite. Claude reads that structured brief and writes the first draft against a brand-voice constraint file. the draft lands back in Airtable in a review column. a human editor signs off before anything publishes.
- Airtable as the brief and workflow layer. one row per post, status moves left to right.
- Claude for drafting. it reads the structured brief plus a voice file with the do-and-don't rules.
- human-review gates. no draft skips the editor. the editor approves, edits, or sends it back.
we are
we built a system that drafts from a structured brief and routes every draft through a human editor before it ships.
we aren't
we did not build a content farm that publishes unedited ai slop on a schedule.
#did quality drop?
no. that was the whole point of the voice file and the review gate. the editor still reads every post and still has the final word. what changed is where their time goes. they stopped writing from a blank page. they started editing a draft that already sounds like the brand. the first-draft-to-publish edit time fell, and nothing went out without a human signing it off.
can ai increase content output without hurting brand voice?
yes, when the voice is encoded as an explicit constraint and a human edits every draft. for this team, Claude drafted from a brand-voice file in Airtable and an editor approved each post. throughput tripled and the voice held, because no draft published without human review.
#how long did it take to ship?
the working system was live in weeks, not quarters. most of the build was the brief schema and the voice file, not the model. once the brief held the right structure, Claude had what it needed. the early drafts went back to the editor often. within a few cycles the constraint file was tight enough that edits were minor. that is when time-per-post settled at 90 minutes, down from 6 hours.
the writers stopped staring at a blank page. they started editing a draft that already sounded like us.
the math is plain. six hours per post to ninety minutes is roughly a quarter of the old time. two posts a week became six. same 12 people. this is the kind of work our consultancy ships, and the marketing team is the one that felt it first.
#would this work for your team?
it works when you already know your voice and can write it down. it works when an editor is willing to own the gate. it does not work if the plan is to publish whatever the model writes. the constraint file and the review step are not overhead. they are the reason the voice held while the output tripled.
if you're a marketing lead weighing whether ai can move your throughput without losing your voice, this is the shape of the build. start with the brief, encode the voice, keep the human gate. we do this end to end in a build engagement. tell us what you're building and we'll tell you whether this fits. start here.