“Who approved that post?”
It went live with a wrong number in the headline. Nobody remembers signing it off.
ConvOps runs your content work as governed AI agent workflows. Briefs become drafts. Drafts face automated reviews, and findings loop back until they pass. A gate holds publishing until it is approved. Put it on a schedule and it runs every morning, remembering what yesterday taught it.
claude cowork · two reviews side by side · runs on a schedule
Post: onboarding checklists
A ConvOps workflow your AI agents run from brief to publish-ready: research, drafts, automated reviews, then a gate that holds publishing until it is approved.
ConvOps is the operations layer for AI agents. It holds your team's process as workflows with gates, a shared memory, scheduled and unattended runs, programs, a plan board, metrics and a full audit trail. It does not sell or host models.
key facts · october 2026
best for
not for
updated
AI drafts in seconds. Then the voice, the facts and the review depend on who was in the chat.
“Who approved that post?”
It went live with a wrong number in the headline. Nobody remembers signing it off.
“Where did that decision go?”
The team agreed on no jargon in March. Every new chat writes jargon again.
“What happens when someone leaves?”
Your editor knew the voice. Now the AI and the new hire are guessing.
A real post, run through the Blog Post workflow from Claude Cowork. Then put on a schedule.
You ask for next week's post in plain words. ConvOps creates the task and attaches the Blog Post workflow, with its reviews and its approval.
Before drafting, the AI asks shared memory: the reader, the voice rules, the rulings from past reviews. Run 40 writes with what run 1 had to be told.
askblog post voice and audience
Sources are gathered and pinned before the draft starts. Every claim in the draft points back to one of them.
Onboarding checklists people actually finish
SEO and engagement review the draft side by side. A finding sends it back to the draft until both are green. A hard red files a bug instead of shipping.
Keep the approval gate and the pipeline drafts on its own but ships only once an explicit approval is given. Remove it when the double green is enough.
Publish "Onboarding checklists people finish"?
SEO green · engagement green · 3 sources
Put the workflow on a schedule. Each morning a new task fires with the whole process attached. You stop starting it and keep governing it at the gate.
From brief to published in six steps. Reviews run side by side, and you keep the last word.
Topic, reader and angle, captured as structure, not a doc.
Voice rules and past rulings come back before a word is written.
Sources gathered first. The draft cites them.
SEO and engagement side by side. Findings send it back.
decisionKeep this gate and nothing publishes until it is approved.
needs approvalShips with its sources attached.
Labelled examples from different kinds of content work.
An industrial supplier publishes one buying guide a week. An SEO report bound to the review step must be filed before the draft reaches the approval gate.
Required report and approval gate
A consumer brand runs one task per locale from the same workflow. Each local editor approves their market's copy at the gate, and the step history shows which account approved which locale.
One workflow, per-locale approvals
A hospital group's comms team runs the digest on a monthly schedule inside a program. If a month passes with no run, the program turns silent before anyone asks where the digest went.
Monthly schedule in a program
Launch copy for a new savings account waits at two gates: product marketing approves the message, and legal approves the claims. A rejection sends the draft back to the writing step.
Two approval gates with a rejection route
The Social Post workflow writes a question hook, a story hook and a statistic hook. The editor picks one at the gate and the agent polishes it.
Social Post template
A custom report type records that every customer quote was confirmed with the customer. The case study reaches approval only after that report is filed.
Custom report type
Scripts for a training channel pass a fact-check review that loops back to the draft until it is clean. There is no approval gate: the review is the bar, and the script moves on to the production team's queue.
Review loop, no approval gate
A research step at the start of the workflow answers the piece's key question, with a source for each finding. It has no approval of its own; the writer reads it before the outline gate.
Research step, no approval
Every review makes the next draft better. Kept by the team, not by one editor.
Readers are ops leads. No jargon. Plain verbs. Recalled before every draft.
how we sound
Headlines under 60 characters. Decided once in review, applied to every draft after.
decided once
The research folder and the brand guide, so no draft starts with a search.
where it lives
Every ruling points back to the post that produced it.
traceable
A content calendar is recurring work. ConvOps runs it, shows when it stops, and counts what it costs.
Give each recurring piece a schedule: an article every Tuesday, a newsletter on the first weekday of the month, social posts every weekday. Each schedule has its own time zone and a preview of the next five runs. Group them into a program, and its verdict shows when a run is late, silent or failing.
Plan the one-off pieces on Plan, next to the scheduled ones. Spread finds the least-loaded days before each deadline and shows a preview before anything moves.
Define the numbers each article must report in Metrics, such as review rounds or findings per draft, and bind them at the review step that produces them. ConvOps also records the cost of every run by itself. Markers on the trend show each change to the workflow those metrics are bound to.
The review step tells your AI session to start both reviewers at once. ConvOps holds the step; your session runs the two agents on the same draft.
Findings send the run into a fix sub-workflow that returns to the draft step. It repeats until the reviews pass.
A schedule runs every few minutes, hourly, daily, on weekdays, weekly, monthly or on a repeat rule, in its own time zone. Each fire either creates a fresh task or wakes the same task where its workflow stopped.
Policies like "every claim carries a source" arrive with the step the agent is on. Guidance at the moment it acts.
Keep the CMS and the writing tools. ConvOps holds the process the piece goes through.
| ConvOps | AI inside the CMS | A writing tool | A chat prompt | |
|---|---|---|---|---|
| Process | Brief, research, draft, reviews and a gate, in order | One edit at a time | A document and suggestions | Whatever the prompt asks for |
| Review before publishing | A gate holds publishing until it is approved | Publish rights in the CMS | Comments in the document | None |
| Brand and legal checks | Required reports, filed before the step moves on | Style settings | Style rules and suggestions | Instructions in the prompt |
| Recurring calendar | Schedules in a program with a health verdict | Scheduled publishing | None | Started by hand |
| Record | Every step, reason and approval on the task | Revision history | Version history | The chat history |
Connect the AI tool your team already uses, install the workflows, and run one real piece of work through them.
Install the Content Studio pack: blog post, social post, newsletter, repurpose
Run one article attended: brief, research, draft, then the two reviews
Watch the fix loop send findings back to the draft until both reviews pass
Then add the schedule, and stop starting it yourself
ConvOps is the operations layer for AI agents. It holds your team's process as workflows with gates, a shared memory, scheduled and unattended runs, programs, a plan board, metrics and a full audit trail. It does not sell or host models.
In ConvOps, the content use case runs articles, social posts, newsletters and repurposed pieces as workflows your AI agents follow from brief to publish-ready. Research and drafting come first, automated reviews send findings back to the draft until it passes, and a gate holds publishing until it is approved. A schedule can start the same workflow every day or every week.
Wherever you put a gate. In ConvOps, a gate holds the step until an explicit approval is given, recorded with the account that gave it. Most teams approve the outline and the final draft, and let research and the automated reviews run without a stop. Add a gate before legal sign-off and the run waits there; remove one and the reviews become the bar.
Yes. In ConvOps, a schedule starts a task at the cadence you set, such as daily, weekdays or weekly, in its own time zone, and you can preview the next five runs before saving. Group the schedules into a program and it shows a health verdict, so a calendar that stopped firing shows up as late or silent instead of going unnoticed.
No. ConvOps holds the workflow, the gates and the record, and posts nothing itself. The agent publishes through the tools its environment provides, such as a CMS connection, or hands you the final text to post. Either way the publish step comes after the gate, so the agent reaches it only once the piece has been approved.
Bind them as reports. In ConvOps, a report type such as a brand check or a legal review can be bound to a step as required in enforce mode, and the step will not advance until that report is filed. Observe mode never blocks and records whether each step would have met it, and the simulator previews what a binding would do before you switch it on.
The ones you define first. In ConvOps you define the metrics each article must report, such as review rounds or findings per draft, and bind them at the review step, where each value is recorded with who recorded it. ConvOps also records the cost, duration, tokens and outcome of every run by itself, and agents cannot write those. Markers on the trend show every workflow change, so you can see whether an edit helped.
The schedule only starts it. In ConvOps, the run then recalls what earlier runs learned, researches, drafts, faces its reviews and stops at the gates you placed. Memory, reviews, gates and the record are what make it a governed workflow rather than a timer, and a program over the schedules tells you when the runs stop arriving.