One brief in Slack · blog, carousel and LinkedIn post back in the same thread · every claim from a verified fact file
The Legacy Operating Model
The client sells impact analysis to renewable energy developers and corporate buyers. Their methodology is peer reviewed and federally funded. Their story is strong. Telling it consistently was the constraint.
The people who could write credibly about the work were the same people doing the work. A blog post, a set of LinkedIn slides, a follow-up piece after a project win, each one needed hours from someone whose calendar was already spoken for.
So content shipped when an expert found the time.
The Glue
Between the idea and the finished piece sits a sequence nobody puts on a roadmap. Research what the audience actually searches for. Draft long enough to be useful. Reformat the same argument for LinkedIn. Design ten slides. Check the voice. Check every figure.
None of it is hard. All of it is hours.
The figures were the dangerous part. This client's entire product is verified numbers. A marketing claim they could not back would contradict the thing they sell.
For a company that sells verified intelligence, an unverifiable marketing claim is a product defect.
The Shift to AI-Native
The question that reframed the build was what expert hours should buy. The answer was judgment. Which claims to make, which story leads, and the final call on what publishes.
Everything before that judgment is a sequence. Research, drafting, reformatting, rendering and reviewing follow the same steps every time. So the design put the machine on the sequence and left the person with the approval.
Nothing publishes itself. Every deliverable lands as a draft waiting for a decision.
The System
A team member writes a brief in Slack in plain English. The system reads it and runs specialist agents in sequence.
Research. What the audience searches for, the angle that carries the piece, and the customer story that proves it.
Draft. A blog in the client's house voice, twelve to fifteen hundred words, with the question-and-answer section that makes it quotable by search and AI engines.
Atomize. The same argument becomes a ten-slide carousel and a LinkedIn post, each rendered against the brand design system.
Review. A brand-voice agent scores the draft against the voice profile, alongside thirteen deterministic checks. A forbidden-phrase list catches marketing filler. A deny-list catches retired claims.
Deliver. Finished files land in the client's drive. The links post back into the same Slack thread.
One rule sits under all of it. Agents cite only figures from a per-customer fact file with explicit may-cite and may-not-cite tables. Figures that failed verification during the build were stripped and placed on the deny-list so they cannot come back.
Example In Practice
Someone pastes a five-line brief about a corporate-buyer topic into the channel.
Status updates flow into the thread as each agent finishes. Then the links arrive. A fourteen-hundred-word blog with its question section. Ten rendered slides. A LinkedIn post. And a voice review that flagged one banned word before any person had read the draft.
A full drop takes between eight and fifteen minutes. A single piece takes two to five. Model spend for a full run is a few dollars.
How We Made It Happen
Every agent was evaluated before it was trusted. Scored test briefs, dimension grades, and the thirteen rule checks ran against each agent's output during the build, and the checks kept catching real defects. One early run failed on a single banned word the reviewer would have had to spot by eye.
Training happens where the work happens. A user comments on the delivered document in plain language. The system applies the edit, and when the comment states a rule rather than a one-off fix, it files the rule so every future draft obeys it, then confirms both in the thread. No prompt engineering is asked of anyone.
The system was handed over with its documentation, its runbook and a knowledge transfer session. The client owns the code, the voice files and the fact files, and operates it in their own workspace.
The Impact
Time per content drop. From a brief in Slack to reviewed deliverables in the thread in under fifteen minutes, documented across runs.
Consistency. Every piece clears the same voice review and the same thirteen checks before a person reads it.
Trust. Every published figure traces to a verified fact table. The claims that could not be verified are structurally unable to return.
Where attention goes. The team briefs, approves and publishes instead of drafting and formatting.
What This Means
For a thin team, the constraint on content rarely sits in the ideas. It sits in the hours between the idea and the finished piece, and those hours belong to the people least able to spare them.
Ask the question that started this build. If this workflow did not exist today, how would it be designed?
The answer is usually that a machine handles the sequence and a person handles the judgment. Getting there takes a working system, evaluated before it is trusted, installed where the team already works.
Ready to redesign a workflow?
Thirty minutes, one workflow of yours, timed end to end. We come back with what it costs today and what it would look like rebuilt.
No deck.