Five-Step AI Workflow with Human Oversight
Each step shows the workspace, what the AI completed, and what still needs human review.
Steps 1 and 3 need a person. Steps 1 and 3 are waiting for approval right now. Click either to open the full walkthrough.
Estimates are directional and based on stated assumptions. All names, organizations, and identifying details have been anonymized in accordance with our confidentiality agreements.
How a Growth-Stage HR Services Company Built a Content Loop Where Lead Results Set Next Week's Topics
Turns the gap between what buyers search for and what the site answers into a steady stream of published articles that sound like the company, and never lets an unsourced number or an off-voice sentence carry the company's name. Five stations, run as a loop: what station 5 learns about which topics earn leads changes what station 1 puts in front of you next week.
What Was Broken
What We Built
Five stations and 10 subagents. Each subagent carries its own tasks and its own refusal.
How It Runs
Topic Research
A person answers hereAgents track every term this company and its competitors rank for, find the gaps worth filling, and draft a one-page brief per gap before a person picks any of them. A person picks which briefs get written this week. That choice is strategy, and it stays a person's.
Article Drafting
Nobody is asked anything hereAgents write each commissioned brief into a full draft in the company's voice, with the target terms worked in where they read naturally and a source beside every number. Nobody writes or edits here. A person reads the finished draft at the review desk, not the half-finished one.
Editorial Review
A person answers hereAgents run every on-page check, trace every claim to the source it cites, and lay the draft in front of a person with only the flagged sentences left to read. A person reads the flagged sentences and approves the draft or sends it back. Every article crosses this desk before it carries the company's name.
Publishing
Nobody is asked anything hereAgents slot each approved article into the calendar, draft the two title variants its live test will run, and push it to the site, the newsletter and LinkedIn at its slot. Nobody schedules or posts here. The approval at station 3 is the last human touch before the article is live.
Performance Tracking
Nobody is asked anything hereAgents track every live article's rankings, traffic and leads, call each title test only when the numbers can carry the call, and route what worked back into the topic queue. Nobody assembles a report here. The monthly readout arrives written, and the findings are already in next week's topic ranking when a person opens it.
Where a Person Decides
Operating Model
This changes how work flows through the team.
What Transfers, What Must Be True
Failure Modes
What breaks this pattern:
The site already answers the topic, and nobody points at the existing page. The new article and the old page compete for the same searches, and both rank worse.
No source carries the claim, so the draft fills the gap with something that sounds right. A reader checks it, and the article loses their trust.
The publish date arrives and the article goes out with no approval record. The first person to catch the error is a customer, not a reviewer.
A few dozen visits look like a result, so the test names a leader. The team ships the wrong variant everywhere and never finds out.
Directional Outcomes
What the agent counts, and the station that counts it.
These counts are the tallies from one monitored run of the agents. They are not monthly or annual totals.
What Runs Where
Every step names the subagent that does the work, the record it writes, the thing that raises a question for a person, and what it is allowed to touch. This is drawn from the source, not from a diagram somebody kept in sync by hand.
topic-researchpipeline/briefs/<topic-id>.jsoncommission-briefspipeline/**, rankings/** read-onlydraft-articlepipeline/drafts/<article-id>.mdclaim-without-sourcepipeline/**, voice/** read-onlyreview-draftpipeline/reviewed/<article-id>.jsonawait-approvalpipeline/**, sources/** read-onlypublish-articlepipeline/published/<article-id>.jsonapproval-record-missingpipeline/**, calendar/**track-performancereports/monthly/<period>.jsonunexplained-dropreports/**, pipeline/** read-onlyStack
Every system this agent reads or writes.