Video: "Hermes Agent Kanban Swarms AI SEO is INSANE!" by Julian Goldie on YouTube.
What the Kanban swarm actually is
A single Hermes agent is sequential: it picks a task, works through it, finishes, then picks the next. Swarm mode changes that. You configure multiple agent profiles — each with a defined role and scope — and launch them against a shared Kanban board. The planner agent maps strategy and creates tasks. The builder agent picks tasks up and produces content. The reviewer agent checks finished work against quality criteria. A router handles which agent gets which task. They all run simultaneously on the same job.
For an SEO workflow, that means keyword research, content briefs, and on-page build tasks can happen in parallel rather than in a long sequential chain. The whole job finishes in a fraction of the time, and the handoffs between agents are automatic.
What Julian Goldie's test showed in practice
The demonstration starts from a competitive keyword brief for a business service page. The planner agent splits the brief into tasks — semantic clusters, internal linking targets, heading structure, entity coverage. The builder agent picks these up and drafts the content. The reviewer agent applies a checklist against the output and flags issues back to the board. The whole run, from brief to reviewed draft, completes without a human in the loop.
That said: the output needs editing. Swarms write fast, but they write to the brief, not to your specific voice or any knowledge about your business that you have not explicitly put into the system prompt. In practice you are getting a solid draft quickly rather than finished content. That is still a meaningful time saving for any team producing content at volume.
Where the swarm approach earns its keep
High-volume content operations are the obvious fit. If you are producing 20 or 30 pages a month, removing the sequential wait between each stage saves real hours. It also works well for technical SEO audits, where a swarm can run across a site structure and collect data from multiple pages simultaneously rather than processing them one at a time.
The reviewer agent is worth highlighting separately. A single agent running at full speed tends to skip quality checks on its own output — the model is too close to what it just wrote to catch inconsistencies. A dedicated reviewer running on the same content after the fact catches issues a solo agent misses more often than not.
What to watch out for before scaling up
Swarms consume more tokens than single-agent runs because you are running multiple models in parallel. On a free model stack the cost is zero, but on paid models the bills multiply with the agent count. Factor that in before you decide to run 10 agents on the same job.
Worth knowing: the quality ceiling is still set by how well you define the individual agent roles. A poorly specified planner agent creates vague tasks, and a vague task produces vague content downstream. The Kanban board makes the problem visible quickly — you can see where the chain is breaking — but it does not fix a weak system prompt automatically.
Where this connects to NordSys
We set up AI SEO workflows for clients — including deciding when a swarm approach makes sense and what prompt engineering the individual roles need to produce output that is actually worth using. A Kanban swarm without careful role definitions tends to produce repetitive content at speed, which does not help with rankings. Done properly, it changes the throughput of an SEO operation significantly. See our SEO & AI Ranking service if you want a structured setup rather than a YouTube configuration.
See our SEO & AI Ranking service →