I did not have agents communicating in a Discord server on my 2026 bingo card. Honestly, I barely had OpenClaw on it going into this year.
But here we are.
Apparently my 2026 operating system includes a Discord server, a doc human, and me pretending that is a normal sentence. Right now, Discord has turned into mission control for a weird, useful, messy setup where humans and agents are talking to each other in the same place.
Agent to human. Human to agent. Agent to agent. Human to human. All combinations.
That sounds ridiculous when I write it out. It also happens to be true.
I have OpenClaw agents. I have Hermes agents. I use command line tools, Codex, Claude, local agents, cloud agents, and whatever else makes sense for the work in front of me. The problem was never whether the tools were useful.
The problem was communication. The tools were useful. The hard part was making the people, agents, rooms, notes, and handoffs move together without turning me into the clipboard.
How do you get all of these things to coordinate without manually carrying context between every system? That is the part I have been trying to solve. And somehow, surprisingly, Discord is the thing that started making it work.
Discord Became the Working Room
I would not have guessed this a month ago. If you asked me then whether I wanted Discord to be part of my daily operating system, I probably would have said no.
I have used Slack. I understand Slack. Slack is what a lot of teams reach for when they need a work chat layer. But for this particular workflow, Discord feels less clunky.
I know that sounds weird. I also need a minute with it. It feels weird to say. But Discord was easier to set up for what I needed. It gave me channels, rooms, agent visibility, lightweight conversation, and a place where different systems could show up without me forcing everything through one narrow tool.
This is not about Discord being perfect. It is not perfect. The routing can still be finicky. Some of that is probably the system. Some of it is probably me still getting the whole thing dialed in. When you are building a workflow like this, it is impossible to know every conversation that will happen, which agent needs to get pinged, which room it belongs in, and what the clean path should be.
You figure it out by using it. That is the uncomfortable part. You cannot spreadsheet your way into the perfect agent communication system before the agents start communicating.
I say that as someone who has absolutely tried to out-think the mess before touching the mess.
You have to start somewhere. Run the imperfect version, watch where it hurts, and make the system better one pass at a time.
The Real Problem Was Coordination
Before this started working, a lot of the coordination lived in Telegram channels. That was fine for a while. Then the work got bigger.
More agents. More problems. More snippets. More questions. More handoffs. More context that needed to move from one place to another. It started becoming a lot of copying and pasting.
That is usually the warning sign. When the human becomes the integration layer, the system has a problem. Nobody wakes up hoping to become a haunted USB adapter for their own workflow.
I do not mean that humans should be removed. That is not the point. The human still needs to provide direction, judgment, context, and clarification. But the human should not have to manually carry every little piece of context between every tool forever.
That gets old fast. It also does not scale.
The big draw for me is simple: OpenClaw agents and Hermes agents can coordinate in one place when needed. That is the thing I have been trying to solve for the last couple months.
I am sure there is an easier or better way. There probably is. But from what I could tell, nobody had really solved the exact version of the problem I was dealing with. So I started banging on it. That is usually how these things go.
The Short Version
Here is what is working so far:
- Discord is where the chats happen.
- OpenClaw and Hermes agents can show up in the same working space.
- Humans can jump in when clarification is needed.
- Agents can coordinate without every handoff going through manual copy and paste.
- Dr. Melfi keeps the notes from disappearing into chat history.
- Tony coordinates the work, refines it, and pulls me back in when needed.
- The whole thing is messy, but useful.
That last bullet matters. Messy but useful is where we are with agents right now. Not polished. Not magic. Not fake either. Useful.
Discord Is Chat. Melfi Is Memory.
At first, I was thinking about the split between Discord and Obsidian. But the good news is there really is no split.
Discord is where chats happen. Dr. Melfi is the doc human. She keeps track of notes so people do not forget things. She can capture decisions, ownership changes, completed work, cleanup, and all the little details that disappear when everything lives only in chat.
Then later, when someone needs the information, we can ask her. Or she can resurface it in daily notes when I review things.
That is important because chat is fast, but chat is terrible as a long-term memory system. Everybody knows this. You can have the best conversation in the world, make three good decisions, assign the work, feel great about it, and then lose the whole thing under 400 new messages.
That does not work.
So Discord can stay fast. It can stay a little messy. It can be the working room. But the important stuff has to land somewhere findable.
That is where Melfi and Obsidian matter. The operating idea is not "put everything in Discord and hope." The operating idea is "let the work happen where the work happens, then make sure the system remembers what matters."
That is a big difference.
Tony Is the Orchestrator
The other piece is Tony. He is the Chief of Staff, but in this workflow he is also kind of the Director of Engineering. That sounds like a joke, but it is pretty accurate.
He communicates primarily with me. He orchestrates a lot of the work. He also steps in and does work when he needs to. Sometimes he coordinates. Sometimes he facilitates. Sometimes he refines and shapes the assignment. Sometimes he comes back to me with questions because the agents need more guidance or I did not provide enough clarification.
That is what a good handoff looks like right now. I do not need to be in every single detail. I can communicate primarily with Tony, then pop in when something needs my judgment.
That matters because I am the bottleneck more often than I want to admit. I do a lot of different things. I have a lot of context in my head. I am trying to keep an entire agent fleet on the straight and narrow while also doing real work, client work, family life, plumbing, business, and all the other things that do not care that I am excited about agent orchestration.
So yes, I am part of the problem. The master orchestrator of the agent fleet is also an obvious bottleneck sometimes. That is the honest version.
The Hard Part Is Not the Agents Doing Work
The agents are pretty dialed in. They have guardrails. They know their roles. They do excellent work.
The hard part is communicating all of the things. That is what people miss when they talk about agents. They want to talk about whether the agent can write code, write copy, summarize research, run commands, inspect a page, or generate a plan.
Fine. Those things matter. But once you have multiple agents doing real work, the bigger question becomes: who knows what is happening?
Who owns the next step? Where did the decision land? Who needs to be pulled in? What context does the next agent need? What should not be repeated? What has already been tried? What did we decide three hours ago that nobody wants to scroll back and find?
That is the actual work. The agent doing the task is only part of the system. The communication around the task is where the system either gets stronger or turns into a pile of disconnected cleverness.
Something Is Better Than Nothing
This is not clean yet. I do not want to pretend it is. The routing can be finicky. The workflow is still evolving. Some conversations go where they should. Some probably need better paths. Some things need clearer ownership. Some of this is still me figuring out the right shape as I go.
But something is better than nothing. That has been my approach. Start somewhere. Push. Use it. Watch where it breaks. Fix what you can. Keep going.
Eventually the system starts to work the way you want it to work. That is not glamorous. It is also how most useful systems actually get built.
You do not get the perfect operating model by thinking about it forever. You get it by running the imperfect version, seeing where it hurts, and making it better. That is true in software. It is true in business. It is definitely true with agents.
We Are in the Messy But Useful Stage
That is where agents are right now.
Messy but useful. That is the actual stage we are in with agents right now. Not polished. Not magic. Not fake either.
Communication is better than it was. The agents are better than they were. The guardrails are better. The handoffs are getting better. The daily notes are better. The system remembers more. I copy and paste less. The work moves with less friction than it did before.
That is progress. And progress is the point.
If someone looks at this and thinks, "that sounds like too much work," I get it. It is work. You should not try to do it all at once. You probably cannot do it all at once. But you have to start somewhere.
Do not expect magic without thought and planning. Do not expect agents to fix a workflow you refuse to understand. Do not expect the perfect system to arrive fully formed.
If you are trying to build something like this, start smaller than you think. And if you are stuck, find me on X or LinkedIn. I am happy to answer questions and help where I can, because this stuff is moving fast and nobody has the whole map yet.
But also, do not be afraid to have your agents help you figure it out. They are pretty fucking good at that if you ask them questions.
That might be the funniest part of this whole thing. The agents are not just doing the work. They are helping me build the way the work gets done.
That was definitely not on my bingo card. The bingo card is on fire at this point. But I will take it.
