Big Idea

The argument in one line.

A multi-agent system only becomes useful when you give each agent a defined lane, a delegation map, and scheduled overnight work — four copy-paste prompts cover the entire foundation for getting real outputs by morning.

Who This Is For

Read if. Skip if.

READ IF YOU ARE…

You have a Hermes agent system set up and are not sure what to actually task it with beyond one-off prompts.
You run a solo business or small team and want research, briefs, or marketing content waiting for review each morning.
You have built agent profiles but have not written delegation rules or routing logic yet.
You want concrete copy-paste prompts instead of high-level theory about what AI agents can do.

SKIP IF…

You do not use Hermes or Claude Code as your agent platform — the prompts are Hermes-specific and will not translate directly.
You are new to AI agents and have not set up any agent profiles yet — this assumes a working system already exists.
You want agents that auto-deploy to production; the creator explicitly argues against that throughout.

TL;DR

The full version, fast.

The video presents four ready-to-use prompts for a Hermes multi-agent setup. Workflow 1 creates a scheduled daily business brief that surfaces research signals, priorities, and blockers each morning. Workflow 2 turns a single idea into a full content campaign with positioning, copy pillars, and post angles. Workflow 3 schedules an overnight local improvement build where code or creative stays local until reviewed, never touching production. Workflow 4 generates a delegation map that tells each agent which lane owns which type of request, turning a pile of chatbots into a routed team. The through-line: start with one agent, let it prove value, and always keep a human gate before anything ships.

Free for members

Chat with this breakdown — free.

Sign in and you get 23 free chat messages on us — ask for the hook, quote a framework, find the exact transcript moment, generate a markdown action plan. Bring your own key when you want unlimited.

Create a free account →

Chapters

Where the time goes.

00:00 – 00:40

01 · Intro + prop reveal

States the four workflows; points to clearmud.ai/resources for the visual prompt prop.

00:40 – 04:14

02 · Workflow 1: Daily Business Watchtower

Explains the morning brief concept. Shows Oracle agent sending a morning operator brief with research signals, priorities, blockers, and overnight build summaries. Breaks down the copy-paste prompt structure.

04:14 – 06:48

03 · Workflow 2: Content to Campaign

Live demo prompting Neo to turn a single idea into a content campaign. Shows agent generating positioning, copy pillars, and post angles. Reveals a routing gap: Mouse was not reached because that marketing lane has not been configured yet.

06:48 – 07:18

04 · PromptBrowser interlude

Quick intro to the free open-source tool that extracts spoken or on-screen prompts from YouTube videos.

07:18 – 13:46

05 · Workflow 3: Overnight Business Improvement Build

Emphasizes local-only rule: never let agents deploy to production without human review. Sends a live prompt to Neo to create a nightly marketing kit. Shows Neo picking up the Kanban task and routing to Mouse, with a cron set for overnight runs.

13:46 – 16:25

06 · Agent Atlas Plugin

Preview of an unreleased visual org-chart plugin. Shows the full team hierarchy: administrator, Neo as executive assistant, and four department heads with labeled sub-agents in each lane.

16:25 – 20:11

07 · Workflow 4: Delegation Map Builder

Copy-paste prompt for generating a delegation map. Recommends sketching an org chart by hand first and feeding it to the agent. Checks back on the overnight build: kit was created, cron is set, Kanban updated, and a missing API key was flagged as a blocker.

20:11 – 20:48

08 · Outro

Invites comment-driven video requests with personal context. Signs off as Clear Mudd, clarity matters.

Atomic Insights

Lines worth screenshotting.

Separating your morning brief into FYI versus needs-action is what makes it actually useful instead of just more reading.
A pile of agents without a delegation map is just 30 chatbots — the map is what makes them a team.
Running an overnight build local-only and reviewing it before release is the difference between useful AI and contributing to web slop.
The routing failure you catch on camera during a live demo tells you more about your agent setup than a successful run would.
Spending your first month testing and building agent profiles before demanding output is not wasted time — it is the work.
An executive assistant agent with no external tokens tied to it is a feature, not a limitation — it keeps the master delegator in one controlled channel.
Feeding an AI agent a hand-sketched org chart is more effective than prompting it to invent your org structure from scratch.
One idea plus one content-to-campaign prompt produces positioning, copy pillars, and post angles faster than brainstorming any of them manually.
If your marketing department head agent is not routing correctly during a live test, that is the single most valuable signal the session produces.
A Kanban board as single source of truth for agent work means every task is inspectable even when the agent operates overnight without supervision.

Takeaway

Four prompts that turn scattered agents into a working system

WHAT TO LEARN

An agent that does everything does nothing useful — routing rules, scheduled briefs, and a local-only gate on deployments are what make a multi-agent setup actually reliable.

A daily brief prompt that separates for-your-information from needs-action prevents you from spending morning reading time on noise that does not require a decision.
Sending one idea through an agent and asking for a full content campaign produces positioning, copy pillars, and post angles faster than brainstorming any of them manually.
Any overnight agent build should stay local and preview-only until you have reviewed it; pushing AI-generated code or copy straight to production without a human gate is how low-quality content accumulates.
Delegation maps work better when you sketch an org chart by hand first and feed it to the agent — the visual tells the system who owns engineering, marketing, growth, and research without you having to write rules for every edge case.
Starting with one agent and letting it prove value beats building 30 profiles at once; the complexity of routing, skills, and delegation maps compounds fast and most of that scaffolding becomes overhead before it becomes useful.
When a routing call fails during a live demo — the marketing department not receiving a delegated request — that is the most useful signal the session produces, because it shows exactly which delegation rule is missing.

Glossary

Terms worth knowing.

Hermes: A multi-agent system built on Claude Code that lets you create named AI agents with defined roles, skills, and communication channels such as Telegram, Discord, and TUI chat.
Agent Atlas: An unreleased visual plugin the creator built that maps all agent profiles into an interactive org chart, showing department heads and their sub-agents with routing rules.
Delegation map: A structured document or visual that defines which agent owns which type of incoming request — engineering, marketing, growth, research — so work routes automatically without manual intervention.
Cron: A scheduled job that fires at a set time; in Hermes, agents can create and manage their own cron schedules so tasks like morning briefs or overnight builds run automatically.
Product OS: The creator's term for using Hermes to handle all front-facing revenue work while keeping internal operations on a separate agent system.
PromptBrowser: A free open-source tool from Clearmud that extracts spoken or on-screen prompts from any YouTube video, timestamping on-screen-only prompts for manual retrieval.
Kanban runtime inbox: A task board used as the single source of truth for agent assignments; the creator requires agents to log work here unless given an override phrase.

Resources

Things they pointed at.

00:19linkclearmud.ai/resources ↗

05:10toolPromptBrowser

13:48toolAgent Atlas Plugin

16:46linkHermes Bootcamp (coming soon) ↗

Quotables

Lines you could clip.

07:42

“I would not recommend allowing your agent or agents for that matter do these self-improvements without human approval. I think it is very irresponsible of us to allow our agents to deploy code to production without reviewing it ourselves.”

Contrarian stance in a landscape of let-the-AI-do-everything hype; strong take with a clear consequence→ TikTok hook↗ Tweet quote

13:36

“We don’t wanna build 30 chatbots, but we do wanna build a delegation map.”

Tight one-liner that reframes the goal of multi-agent setup; no setup needed→ IG reel cold open↗ Tweet quote

16:02

“I landed on the fact that Hermes is gonna be my front facing AI agent system that handles anything revenue generating.”

Clean strategic framing — system-of-record language that cuts through noise about which agent platform to use→ newsletter pull-quote↗ Tweet quote

19:07

“It’s so easy to get carried away to create so much fluff and not get anything done — I highly recommend just starting with one agent.”

Honest self-aware warning from someone who built 30 profiles; relatable for anyone who over-engineers before shipping→ TikTok hook↗ Tweet quote

The Script

Word for word.

Read-along

Don't just watch it. Burn it in.

See every word as it's spoken — crank it to 2× and still catch all of it. The same dual-channel trick behind Amazon's Kindle + Audible.

metaphorstory

00:00In this video, I give you four prompts that you can start using today to automate your business with your Hermes agent system. So without squirrel ado, let's build. Alright.

00:09So as you can see here on the screen, I had my executive assistants help me prep this visual prop as per usual. Now if you want access to this, it's completely free to access.

00:18Just go to clearmud.ai, go up top, click resources, and go to the Hermes section, and you'll find this video prop for you to copy and paste all of these prompts at your leisure.

00:30So let's show off four things that you can start doing today with your Hermes agent.

00:36Now this is all about automation. This is all about saving you time. So we're gonna start off with the brief.

00:42I know it's not the most exciting workflow, but it's honestly very important because it allows you to configure a scheduled cron and eventually tie a skill to it. We'll talk more on that in future videos where your agent or agents for that matter does research overnight and gives you a report ready to read first thing in the morning with a cup of coffee.

01:06I personally read two to three briefs every single morning. It saves me so much time from having to browse things myself, get distracted, squirrel, which I like to call it, and waste ten to fifteen minutes doing something I wasn't supposed to be doing when I was simply doing research.

01:24It is a time saver. In this prompt, we call it a daily business watchtower. Now this one more references your actual business here, so here's a copy paste prompt you can use.

01:35Hermes, create a daily business brief for my business. Check the following sources. Give me what changed, top priorities, risks or blockers, opportunities, decisions I need to make, recommended next steps.

01:49Separate for your information from needs action. Now I wanna give you an example, Right? Because it's easy to walk through this prop, but it's another thing to actually give you a real life example here.

02:01So I'm gonna show you let me actually open up in a new window here. This is Oracle.

02:09She is based off of listen. I'm a big Matrix fan here.

02:13She is based off of the Oracle and the Matrix universe. And every morning, she sends me a morning operator brief.

02:21Now this evolves on a daily basis. Anything I don't like, anything I wanna tweak, I talk to her real time saying, hey. Please adjust your cron.

02:30Uh, eventually, I'm gonna have a skill tied to it, but we're still in the early stages of configuring this morning brief to where I'm just letting her use her reasoning based off of GPT's 5.5 high reasoning to kind of just give me everything I ask for, and then I fine tune the output. So I've started doing some overnight builds, one for a micro SaaS idea, one for an open source idea, and then one for an existing product upgrade of ours.

02:57So I'm not gonna dig too much into this because this was custom built for me. But what started off with Oracle was research. She also does market research and comes up with these ideas for me.

03:08So she now coordinates with my engineering department architect to come up with these. Right?

03:14But we're not gonna dig into that on this one. We're gonna go look at the top research signals. Right?

03:19And as you can see, everything's pinpointed with a source URL if I wanna verify. Is listen. I'm not gonna read this line for line, but, you know, the purpose of my multi agent system is front facing projects, products, revenue generating ideas.

03:36It does not have to do with my internal day to day ops. When it comes to the overarching AI industry news and ClearMud related, uh, YouTube news and updates and analytics, that all happens still on my OpenCLOS setup.

03:50My Hermes agent setup, though, I just found that a lot of its strengths tie in with creating what I'm calling a product OS, which is just gonna handle all of our front facing projects and products, whether they're open source, whether they're paid, uh, and everything in between. Right?

04:05So this is just like a brief raw output of what the research looks like every night. So I would say customize this however you see fit. Right?

04:14Now next up is the content to campaign pipeline. Uh, one idea becomes a campaign. And we'll actually test this out right now just to give you a real life example.

04:25So Hermes, turn this idea into a content campaign. Okay? So what I'm gonna do is I'm going to alright.

04:32I'll actually go to my executive assistant for this one, and I'm just gonna say, Neo Neo, turn this into a idea campaign. Prompt browser.

04:46Stop wasting time watching videos and get prompts instantly.

04:56Right? So this is for one of our free open source tools that we've released called Prompt Browser. Right?

05:01And it allows you to take any YouTube video, and assuming the person says it in voice to text, it will extract that prompt for you. If they just show it on screen, prompt browser will will show you the time stamp of when they showed it on screen so you can either screenshot the screen, open the video for yourself, uh, etcetera, etcetera.

05:20So let's see what happens, shall we? Now I already have a lot of as you saw with my, you know, agent system here.

05:27This is actually a rather complex multi agent team with a lot of moving parts. Now to be fully transparent, I've only started working on my engineering lanes and research lanes.

05:38Marketing and growth is untouched, so it's gonna be very interesting to see how Neo delegates this over to Mouse and what happens when Mouse delegates it to the necessary sub agents. And look at this. I mean, everything's already done and since campaign concept so since I haven't started working on my marketing department, right, he actually didn't even route it through Mouse.

06:03So that's good to know. It's something I need to fix on my end. Stop wasting time watching videos.

06:08Get the prompts instantly. The campaign is built around one frustration. People are sitting through eighteen minute AI tutorials just to find one useful prompt buried at minute twelve.

06:16Prompt browser turns that into a product promise. Paste the video, get prompts, start building. Core positioning.

06:22So you see? Look. Like, I don't have to go prompt it manually now.

06:26I can one up this and tell it to create pieces of copy for me, which it looks like it already did. So content pillars, the pain, tutorial fatigue.

06:37That's a great approach. The product prompt extraction, great approach.

06:41The outcome, faster building. The behavior, build a prompt library. And for those curious, you know, here's what what it looks like here.

06:51It's completely free to use if you just go to browse all. By default, right, I built it for us, but you can also go look at any other channel. These are all run by the public.

07:01I don't run any of these. Anybody using the tool is able to run it without even signing in. The sign in feature just allows you to save prompts and for for future things I plan on incorporating within this project.

07:13As you can see, it's still in beta. But, yeah, pretty nifty.

07:17Pretty nifty. Cool. Alright.

07:19So let's move on to workflow or prompt number three. It's an overnight business improvement build. This is where you get very, very, very creative.

07:28The problem, the best improvement ideas need planning, execution lanes, checks, and a morning handoff before anything touches production. Keep in mind, this workflow is meant for local only.

07:39Right? I I would not recommend allowing your agent or agents for that matter do these self improvements without human approval.

07:48I think it's very irresponsible of us to allow our agents to deploy code to production without reviewing it ourselves. You're just contributing towards this slop that is hitting the web, so please do not do that.

08:02Always be that final gate, if you will. With anything that was built overnight by itself from AI, don't just blindly release it to the web because you're just contributing towards the slop factory.

08:16You know what I mean? So but here's the prompt. Hermes or your agent for that matter, run an overnight local business improvement build.

08:23Goal. I wanna improve x, y, and z. Let's say you're building out a custom dashboard or a custom workflow.

08:30Right? Let's say you have all these agents. You say, hey.

08:34Overnight, please create one new custom workflow for us based off of all of our agent profiles. So this is how I'm gonna apply it to, uh, an example right now.

08:44So I'm gonna copy this here, and I'm gonna paste this into Neo. And we're gonna piggyback off of this last response.

08:57So he already knows he's Neo. So I'm just gonna say run an overnight local business improvement build. Goal.

09:04Use the last prompt as an example.

09:10I want to create a marketing kit every every night to either promote one of our products, one of our YouTube videos on, and I'm just gonna link our YouTube channel here so that I have copy ready to preview and edit before posting across x, LinkedIn, or any other text based social media platform.

09:56Right? I'm gonna do it strictly on text. You can one up it and say, hey.

10:02Hey. Also, create a piece of creative to go with the copy via our codex image gen two.

10:15Please force one job right now, but schedule this as the overnight improvement.

10:26Right? So, like, one would think local business improvement build.

10:31It could be anything. In my case, I wanna start seeing new pieces of marketing creative that could be built overnight without my input. And then every morning, right, I review it, I tweak it, I fine tune it, and it gets better and better and better and better.

10:48So I hope this gets your ideas starting. Let me finish this prompt here. Make sure to include it in a new cron sent to me by mouse or marketing department head.

11:08Make sure to drop a note in his agent runtime inbox to make sure he schedules the crons he's responsible for.

11:29And then your agent should already know these come prebuilt, but I wanna make sure, you know, you we included this in the prompt because not everybody, you know, took the time to let me rephrase that.

11:43Not everybody has taken the time to create rules of delegation and delegation maps for that matter nor, you know, fed these skills to your agent saying, make this our single source of truth. Make sure we're using the Kanban board for everything unless I say the magic words of do it now.

12:01Right? Keep it local preview only. Do not deploy.

12:04Yep. There we go. There we go.

12:06And I'm just gonna send it. Listen. This is the definition of building in public.

12:10I have been working with Hermes agent systems for months now. But for my first month, instead of building something, I was testing.

12:19I was experimenting. I was asking questions. I was building agent profiles, starting from scratch, rinse and repeat.

12:26Uh, now I'm at a point where I know what I want out of my system. I've built close to 30 agent profiles.

12:33Really, only five of them have their own gateways. Uh, five of them are responsible for delegation. One is the master delegator, and then he divvies out that work to everybody else who then determines who gets the work in their lanes, in their columns.

12:48This took a lot of time. I didn't just come up with this, uh, off the top of my head. This also comes from the experience of building the multi agent team with OpenClaw, which handles my internal operations, my day to day, my content creation.

13:01Um, but, you know, this particular prompt I just gave Neo to delegate to Mouse, it is all product related. Right?

13:09So I did add that little caveat with the YouTube video just because I know that I wanted to see what it would do. Uh, I'm eventually gonna remove that from Mouse's responsibilities because Mouse handles marketing for our products and services and open source projects that we release here.

13:27So so as you can see, Neo's getting to work here. Now I want to move on to the fourth and final prompt in this video, the delegation map builder.

13:38Right? We don't wanna build 30 chatbots, but we do wanna build a delegation map.

13:44Now I I'm sorry to tease you with something I'm not releasing to the public yet. The Agent Atlas will be available for public use at some point.

13:54But right now, I built it selfishly for me. Right? When I built this out, I didn't intend on it to work the way that it did, but I let Neo know about the Agent Atlas.

14:05And I said, hey. I wanna start creating rules of delegation. When a request comes in, essentially, I I always want Neo's chat to be open.

14:14Right? Unless, of course, he's configuring, uh, delegating work and handling all of that.

14:18I essentially want him to always know if a marketing related request comes in, it goes to the marketing lane. If anything engineering comes in, it goes to engineering. If anything growth, growth.

14:29If anything research, research. You see my point?

14:32Um, and I said, what do we have to do? What can I build for you to help explain this properly?

14:39Because we've created a lot of different agent profiles, and we also have to attach all these different skills to them. Right? Uh, and I was like, I've already created a visual.

14:47So go check out that visual, and then let me know what else you need from me. Well, Neil got to work. He came back to me maybe ten minutes later and said, actually, this is all I needed.

14:58You took the time to create this traditional org chart. You have your department heads. The dotted lines clearly show the sub agents.

15:05You've already labeled, so the name of these agents, I didn't give them fancy names. It's just engineering dash front end, engineering dash back end, engineering dash security, engineering dash release. I highly recommend you do the same because when an engineering related task comes in, architect is configured as that engineering routing owner.

15:25So all work comes to him first, and then he routes things accordingly to everybody in his lane, which I just labeled the raw output I wanted.

15:36Look at marketing, marketing content, marketing via video directory, marketing script writer, marketing designer.

15:42You see you see my point? So but this came from months and countless hours of testing Hermes agent, trying to figure out what team do I wanna build, what are its strengths, what are its weaknesses, and I just I landed on the fact that Hermes is gonna be my front facing AI agent system that handles anything revenue generating.

16:04Anything internal, anything day to day, I'm gonna sound like a broken record. I have so much trust and reliability, uh, on my OpenCLOS system.

16:12I am not migrating that system over. Simply not gonna do it. It's a waste of time.

16:16I'm a big advocate for if it ain't broke, don't fix it. Right? So this leads us to our next prompt.

16:23Right? Looks like Neo's still working there, but we can at least walk through the prompt. So a pile of agents creates confusion unless each role has a scope, routing rules, handoffs, approval, escalation path, output, command center, department lead, and specialist profiles with routing rules, approval gates, and starter plus expander expanded org maps.

16:41Right? So here's the prompt. Hermes, design a delegation map for my business.

16:46This is where you fill out your personal details, your personal goals, all of the different lanes you wanna focus on. And then we specify.

16:56Right? It's gonna create a command center profile. It'll be far more effective of a prompt if you take the time to just sketch one out by hand.

17:05Uh, hopefully you know, I'm planning on releasing this Atlas maybe by the end of the month.

17:14Listen. It's gonna take a lot of work to fine tune it for the public because it's not gonna intuitively know your system.

17:21I have to create an onboarding mechanism so that, you know, it doesn't create the profiles for you. You have to do that ahead of time, but you have to do it in a way where, you know, you know your lanes ahead of time.

17:33Right? So you can always just try to screenshot this screen and feed it to your agent to build your own.

17:39You're more than welcome to. But this is a custom plugin that I'm building that is gonna be modular for any Hermes agent dashboard out there. But here's the prompt.

17:48Right? This is a great starting point for you to start to build out your team, build out your delegation map, and I hope I've shown you enough sauce here to inspire you to build your own multi agent team.

18:01I would highly recommend start as simply as possible. I'm gonna reiterate when I first started building out, uh, my this Hermes agent multi agent team, you know, I started with the default profile, which I've now delegated as the administrator and admin API related work.

18:18Right? Anytime I wanna connect my system to one of my custom built dashboards or solutions, I actually use the default port, the default profile to manage all of that, to manage the entire system as a whole.

18:31And then I have Neo. He's my executive assistant. Right?

18:34When we go and check out my profiles, you can see it here. My default, he has no tokens tied to him. I can't talk to him on Telegram or Discord, Only the Hermes TUI chat.

18:45I did that for a reason. That's just how I prefer to have my system configured. Chances are this might be your default executive assistant, and that's okay.

18:54Right? I think build something that's not too complicated for you first and foremost and then expand.

19:02Because the beauty of this is you can always build more. But I I would suppose the negative here in all of this, it's like, it's so easy to get carried away to create so much fluff and not get anything done that I highly recommend just starting with one agent.

19:19But you'll eventually come to something that looks like this. Right? Alright.

19:23So before we end it, let's go check-in on that last request from Neo. So let's go see what was what was built. Scheduled and forced the first local marketing kit build.

19:32Mouse generated the first local marketing kit for prompt browser. Stop wasting time watching videos, get prompts instantly. Kit, wonderful.

19:42Re reoccurring cron created mouse overnight marketing kit, wonderful. Kanban video orchestrator, interesting.

19:50Okay. I'm getting excited to reviewing all these docs. Mouse runtime in box note dropped.

19:57Checks run blockers. Ah, we need to set a key, so it's good to know. Recommended next steps.

20:06Review KitMD. Picks this pick the strongest hook. Okay.

20:10So that's something I'm gonna go check out, right, on my own time. I wanted this video to be as short and brief as possible, so I hope you found some value in today's video. Now I'm not an AI expert.

20:21I'm building in public and sharing what actually works. If you'd like to see me make a particular video around a specific topic, drop a comment below. Let me know who you are, what you do, who your target audience is, and what your question is, and I will add a video to my queue custom tailored just for you.

20:35Thank you so much for tuning in to today's video. My name is Marcelo. This is Clear Mudd, and clarity matters.

The Hook

The bait, then the rug-pull.

Most Hermes users build the agent profiles, name them after Matrix characters, and then spend the next week prompting them one-off requests. Marcelo from Clearmud skips that phase and hands you the four prompts that actually put the system to work overnight.

Frameworks