How we run Worden on AI agents
One role per agent
These are our own business agents. They cover operations and project coordination, sales, marketing, research, legal and conflicts checks, and finance and administration. Separate workers handle writing, figure verification, web lookups, long documents and code.
Work on a schedule
- Every morning: an operations digest for a person.
- Weekdays, 8 AM to 6 PM: the website inquiry inbox, checked every hour.
- 9 AM, 1 PM and 5 PM: the inbox for work handed between agents, swept.
- Daily: drafts of outreach messages for a person to review. Nothing is sent without approval.
- Fridays: a status from every agent.
- Sundays: a review of open decisions.
- Weekly and monthly: research briefs.
A person controls what goes out
Agents draft. A person sends. Email, posts, payments and anything shared outside the business require a person's approval. Agents write only internal drafts and files in our own accounts. Client material stays in that client's folder, and each agent has access only to the tools and folders its role needs.
Checks on work and failures
A run-health report runs four times a day and flags failed scheduled runs. A separate agent works through fixes on weekdays, and anything needing a decision, or any possible rule breach, goes straight to a person. Each scheduled run ends with a line stating what failed and who should fix it.
We keep a fixed set of test cases, and our rule is to rerun them whenever an agent's instructions or model change. Our written rule is that a separate verification agent, built on a different model family, rechecks the figures in a deliverable.
This website followed that process
A writing agent drafted the copy on this website. A legal agent reviewed its claims, and a different model reviewed it cold. A person approves it before it is published.
What failed in the first weeks
In the first weeks of running, a permission setting prevented the builder from making backups, so fixes piled up until a person caught it. We split long instructions into smaller skills, moved some agents to different models after testing, and filed scheduled runs into their own folders because they buried a person's own conversations. Fixes were written into rules or tests where we could. This is not set and forget.
What we offer clients
We offer the same approach: write the process down, assign repeatable steps to agents under written rules, require a person to approve what goes out, and watch closely in the first weeks. Scope and fee are agreed in writing before work starts.
Common questions
Is this a client example?
No. This describes our own operations, not a client engagement. Details are current as of November 7, 2026.
Do the agents send anything on their own?
No. Nothing is sent, posted or paid without a person's approval. Agents prepare drafts for a person to send.
Can a client get the same approach?
Yes. We offer the same approach, scoped and priced in writing before work starts. A person approves what goes out.
Start with one process
Tell us which process you want to examine.