Hermes Agent gives you Bot Mode and Kanban for working with teams of AI agents.
At first, they can look like two different ways of solving the same problem:
How do I get multiple AI agents to work together?
But they solve different layers of the problem.
Use the wrong one and you can easily end up with a group chat that becomes difficult to follow or a full task board for something that could have been handled by a single bot.
The simplest mental model I’ve found is:
🤖 Bot Mode = the team
📋 Kanban = the work
And when a workflow needs both persistent specialists and structured execution, you can combine them.
In this tutorial, I’ll explain the difference between Hermes Agent Bot Mode and Kanban, when I would choose each one, and how I combined both to build an automated AI-powered YouTube research workflow.
Imagine I have three AI agents:
Sometimes I simply want to talk to one of them.
Sometimes I want one agent to hand something to another.
And sometimes I have a much larger assignment:
Research Topic
↓
┌───┼──────────────┐
↓ ↓ ↓
A B C
└───┼──────────────┘
↓
Final Synthesis
Those are very different workflows.
The first doesn’t necessarily need project management.
The second might.
That’s where the distinction between Hermes Bot Mode and Kanban becomes useful.
Bot Mode gives you a roster of persistent, named AI bots inside Hermes.
Each bot can be backed by a Hermes profile with its own:
Instead of treating every interaction as a new generic AI session, you can create specialists.
For example:
Research Bot
Editing Bot
Operations Bot
Source Checker
Audience Researcher
You can then interact with those bots directly.
More importantly, bots can participate in workflows involving delegation and communication between specialists.
That makes Bot Mode particularly useful when you want persistent AI workers rather than creating a new agent configuration for every task.
When I would use Bot Mode
Bot Mode makes sense when I need:
🔹 Persistent specialist bots
🔹 Direct conversations with individual agents
🔹 Recurring routines
🔹 Agent-to-agent handoffs
🔹 Group discussions between specialists
🔹 Different models, memories, or skills for different roles
For example, suppose I regularly collect three articles and want my Research Bot to summarize them.
I probably don’t need a project board.
I can simply give the job to that bot.
Things change when the request becomes a project.
Suppose I don’t want one researcher anymore.
Instead, I want four specialists to investigate the same topic independently:
Topic
│
┌───────────┼───────────┐
│ │ │
▼ ▼ ▼
Sources Audience Competition
│ │ │
└───────────┼───────────┘
│
▼
Synthesis
Now I have additional requirements.
Who owns each task?
Which tasks can run simultaneously?
Which task depends on another?
What happens if one fails?
When should synthesis begin?
This is where Kanban becomes useful.
Instead of keeping the entire workflow inside conversations, Hermes can represent the work as explicit tasks.
That gives the multi-agent system structure.
Kanban is useful when you need:
⚡ Parallel execution
👤 Named task owners
🔗 Dependencies between tasks
🔍 Review stages
♻️ Recovery when something fails
📊 Visibility into larger assignments
The key difference is that Bot Mode organizes your agents while Kanban organizes their work.
Here’s the rule I use.
You need persistent specialists and conversations between them.
You need structured execution across multiple tasks.
You have persistent specialists executing a repeatable, multi-step workflow.
Consider three examples.
You want one bot to summarize three documents.
Documents → Research Bot → Summary
Use Bot Mode.
You want four specialists to research a topic and a final agent to wait until every researcher finishes.
┌→ Researcher A ─┐
Topic ──┼→ Researcher B ─┼→ Final Agent
├→ Researcher C ─┤
└→ Researcher D ─┘
Use Kanban.
You want that research workflow to automatically happen every weekday.
Now we need:
That’s where I would combine Bot Mode + Kanban.
And that’s exactly what I built for the video.
For the demo, I wanted to answer a practical question:
What AI agent topic should I make my next YouTube video about?
Instead of asking one LLM for ideas, I created a small specialist research team.
The team contains several bots with different responsibilities.
The News Scout is responsible for finding the initial opportunity.
Its job is simple:
Find the latest AI agent news
and send it to the Orchestrator.
Rather than manually triggering this every day, I created a recurring cron job.
The workflow can therefore begin automatically.
The Orchestrator receives the news discovered by the News Scout.
But instead of trying to research everything itself, it turns the request into a Kanban workflow.
It creates tasks for several specialist agents.
This is the bridge between Bot Mode and Kanban.
Bot Mode
News Scout
│
│ handoff
▼
Orchestrator
│
│ creates tasks
▼
Kanban
Once the Orchestrator receives a potential story, several specialists investigate it.
The Source Checker focuses on the evidence.
It asks questions such as:
A topic can be technically interesting and still make a terrible YouTube video.
The Audience Researcher looks at the topic from the viewer’s perspective.
It tries to determine:
Next comes competition.
If dozens of creators have already published nearly identical videos, simply repeating the same information isn’t particularly useful.
The Competition Researcher investigates existing coverage and helps identify where there may still be room for differentiation.
Finally, the Angle Editor looks for the strongest way to turn the research into an actual video.
Instead of stopping at:
“This is trending.”
I want the system to answer:
“Why should I cover this, and what should the video actually demonstrate?”
That distinction matters.
The complete workflow looks roughly like this:
CRON JOB
│
▼
🔎 News Scout
│
│ Finds AI agent news
▼
🎯 Orchestrator
│
│ Creates Kanban tasks
▼
┌─────────────────────────────┐
│ KANBAN BOARD │
│ │
│ 📚 Source Research │
│ 👥 Audience Research │
│ 📈 Competition Research │
│ 💡 Angle Research │
└──────────────┬──────────────┘
│
tasks complete
│
▼
🧠 Final Synthesis
│
▼
📄 Video Recommendation
The research tasks can execute in parallel.
The final synthesis task waits for the research to finish.
That’s exactly the kind of workflow where Kanban becomes much more useful than relying purely on conversational handoffs.
During my demo, the News Scout found a potential topic around Microsoft Agent Lightning.
But discovering the topic was only the beginning.
The specialist agents investigated it, and the final recommendation included information such as:
So instead of receiving:
“Microsoft Agent Lightning is trending. Make a video about it.”
I received something much closer to a research-backed content brief.
That’s a far more useful output.
One of the common mistakes when experimenting with multi-agent AI systems is assuming that more agents automatically means better orchestration.
It doesn’t.
Giving five AI agents access to the same conversation doesn’t necessarily create a good workflow.
The real questions are:
Who should do the work?
What should they own?
What can happen in parallel?
What depends on something else?
Who reviews the result?
What happens next?
Bot Mode solves part of this by giving you persistent specialists.
Kanban solves another part by giving their work explicit structure.
Together, they create an interesting pattern:
Persistent Agents
+
Specialized Roles
+
Agent Handoffs
+
Task Dependencies
+
Parallel Execution
+
Recurring Automation
↓
Repeatable Multi-Agent Workflow
And I think that’s a much more useful way to think about AI agent teams than simply asking:
“How many agents should I use?”
If you remember only one thing from this article, make it this:
🤖 Bot Mode gives you the team.
Your persistent specialists.
Their roles.
Their memories.
Their skills.
Their conversations.
📋 Kanban gives the team structure.
Tasks.
Owners.
Parallel execution.
Dependencies.
Reviews.
Recovery.
⚡ Bot Mode + Kanban gives you repeatable workflows.
That’s when the two features become especially interesting together.
You can have a recurring bot discover work, hand it to an orchestrator, turn that work into a structured task graph, execute research in parallel, and finally synthesize everything into an artifact.
This might actually be the most important part.
Not every AI agent task needs orchestration.
If your workflow is:
Prompt → Agent → Answer
don’t build a task graph.
If your workflow is:
Prompt → Specialist Bot → Artifact
you may still not need one.
Kanban starts becoming valuable when your workflow begins looking more like:
┌→ Agent A ─┐
Input → Plan ─┼→ Agent B ─┼→ Review → Final Output
└→ Agent C ─┘
At that point you have coordination problems rather than simply prompting problems.
And that’s where explicit workflow structure starts paying off.
The interesting part of Hermes Agent Bot Mode and Kanban isn’t choosing which feature is “better.”
They’re designed for different jobs.
Bot Mode is useful for creating persistent AI specialists and enabling conversations and handoffs between them.
Kanban is useful when those specialists need to execute a larger assignment with explicit tasks, ownership, parallelism, dependencies, and review.
And when you’re building a recurring workflow involving both?
Use them together.
For my YouTube research system:
Bot Mode → creates the persistent AI team
Cron → starts the recurring workflow
Bot handoff → passes the opportunity
Kanban → coordinates the project
Specialists → perform parallel research
Synthesis → produces the recommendation
That’s the architecture.
Not a giant group chat.
Not a Kanban board for every tiny request.
Just enough orchestration for the complexity of the work.