Your Hermes probably answers texts and manages your calendar. That's cool.
Mine books sales calls, revives stalled deals, and ships ad creative all while I'm at the gym.
Wait Eric how's yours doing all that?
Think of Hermes as a worker you run rather than a chat window you open. You talk to it in plain language from your desk or your phone, and it does the work in the background. It holds memory, follows the playbooks you've taught it, and runs jobs on a schedule without you. The same agent can manage other agents underneath it, which is what turns it from an assistant into an operator.
Pretty cool right?
And there's so much more you can do with it. The possibilities are endless.
So here are 12 jobs we give Hermes, and what each one does for the business:
1. Branded artifacts on demand
The first job we give Hermes is making things people can look at.
Google published a design spec file on GitHub called design.md. You point your agent at it, and every visual it makes comes out in your brand. It can make carousels, one-pagers, case studies, and ad concepts, all on brand.
You can keep separate design files for separate brands and switch between them depending on what you're building.
Here's where it earns its keep. Last week I needed my team to change how they handle blockers. The old way was to call an all-hands and talk at people. Instead I had Hermes build a branded one-pager that spelled out the new rule: when you hit a blocker, you say "I'm going to do X related to Y by Z unless told otherwise." It laid out the permissions, the go and no-go, the approvals, and a few examples.
A branded artifact sticks in a way a wall of Slack text never does. If you manage anyone, that staying power is most of the job.
2. Run it inside Slack
Keep the agent where the work already happens. For us that's Slack.
You can run Hermes on WhatsApp or Telegram for your own stuff, and that's fine. But Team work is different. Slack gives you threaded replies, so a messy back and forth stays organized instead of turning into one long scroll. You can split threads into separate windows and follow three workstreams at once.
Imagine a multiplayer video game. When the agent is talking to individuals and groups inside the channels your team already lives in, knowledge moves between people faster, and faster knowledge transfer means faster work. That's why it belongs in Slack and not in some side app only you can see.
3. Give yourself an agent you work in
Now give yourself an agent you actually work in, and run several threads at once.
Someone asked us for a media deal this week and wanted our media kit. I didn't want to build one from scratch, so I dragged in the version someone made two years ago and told Hermes to pull our current numbers off every platform, update the kit, and make it look right using our design file. It came back for one approval and that was it.
And that's just one thread. In other threads I'm running a closed-loop operating system that keeps work moving on its own, and a revenue system that keeps building pipeline for the team. Each thread is a workstream that progresses while you're doing something else. Your job shrinks down to approving instead of doing.
4. Wire it into Linear
An agent with no record of what it's working on stalls. So connect it to a project tool. We use Linear because the API is simple, which matters when you're wiring several systems into it.
Hermes plugs into Linear alongside Claude Code and Codex. As work happens, it opens Linear tickets so the project has a spine, then it routes those tickets back to me to review. Pair that with a closed loop and projects keep moving forward without me checking in every hour.
Try it yourself. Pick a tool with a clean API, have the agent write its work into tickets, and review the tickets instead of doing the work yourself.
5. Use a /goal command
Define what done looks like, then let the agent grind toward it. That's what a /goal command does.
In Codex or Claude Code you give it a definition of done and it keeps working until it gets there. The same works in Hermes, and you can upgrade the goal scaffolding so it runs even longer. I keep one screen running builds and a main screen where I keep moving. When I asked, Hermes put me at roughly 1.8x faster on the kind of work I do.
The trick is running these harnesses in tandem instead of one at a time. One definition of done, several engines pushing toward it, and you stop being the thing that has to start every task.
6. Give it memory with Obsidian
Your agent's built-in memory isn't good enough to run a business on. So bolt better memory onto it.
We use Obsidian as the memory store and upgrade how the agent reads and writes to it with QMD, an open setup from Shopify's Tobi Lutke that lives on GitHub under those three letters. Every day, the decisions we make get logged into Obsidian. Now you've got Obsidian holding memory, Linear holding the work, and a closed loop tying them together.
It starts to feel like working with a senior operator who gets a little sharper every day. You add things, cut things, and slowly decide which of those daily decisions are stable enough to automate. Without external memory, the agent wakes up cold every morning and you re-explain everything. With it, the system compounds.
7. Connect your APIs
A general chatbot guesses. An agent wired into your real tools reasons against your actual business. So connect the APIs for the tools you already use.
For us that's the call and CRM layer first: Granola, Gong, HubSpot. Then the content layer: the X API and the YouTube API. When I see something sharp on X, I send Hermes the link and ask how it helps what we're working on right now. It already knows our goals, so it tells me whether the idea is genuinely differentiated and what it'd do with it. On YouTube I skip pasting transcripts. I point it at a video and ask what we can learn. Yesterday I sent it someone explaining how they recursively improve their products, and we dropped that straight into our closed-loop skill.
Every API you connect turns the agent from a guesser into something that thinks with your data in front of it.
8. Build a content ingestor
Once the APIs are in, build a content machine on top of them.
Point it at everything you publish, your X posts, your YouTube, your short form, and have it pull in what's performing, what the market is making, and what you've actually been talking about in meetings. Run it against your Granola notes, and the topics you got most fired up about internally are usually your best content. This piece started exactly that way. I asked Hermes what was interesting about how we work, it handed back a list, and I packaged it into the video that became this.
Idea generation stops being a blank page. The machine tells you what to double down on, and making the content gets easier because you're not inventing the topic anymore.
9. Let it draft your emails
Email is where you choose how much you trust the thing.
Mode one: it drafts, you approve, you send. Nothing leaves your outbox without you. Mode two: it sends on its own and gets better over time through the closed loop. We run revenue agents at the send level for work I don't want to touch, like sponsorship replies and booking customer development calls for one of our software products. It does that today.
The same agent can revive stalled deals or run trigger-based outreach. If one of your ideal customers starts hiring for a certain role, that's a signal, and the agent can reach out on it. Start in draft mode, watch it, and move it to send once it earns the trust. The autonomy dial is yours.
10. Stack specialist agents underneath it
The biggest jump is when you stop running one agent and start running a team of them.
We build narrow specialists. One owns cold outbound infrastructure. One spins up ad creatives at scale. Each is good at a single job. Then Hermes sits on top and manages them, which means you brief one agent and it coordinates the rest. This is exactly what we build for companies under Single Brain, with a B: managed revenue agents unified by one brain that lives in Slack so the whole team can talk to it.
This is what we mean by becoming an army of one. One person operates like a department, because the manager agent is doing the coordinating. You hold the outcome. The specialists handle the execution underneath you
11. Use OAuth, not tokens
This one saves you the most money, so pay attention.
Running everything through metered API tokens gets expensive fast. Mine climbed into the thousands a month. The fix is to run on subscription access instead of per-token billing. Use OpenAI's OAuth and Claude through the CLI, and you're paying somewhere around two hundred a month instead of watching a token meter spin all day.
Same work, a fraction of the cost. For most people this one change pays for the rest of the stack.
12. Go multi-device with Tailscale
Last one, and it's the difference between an agent that runs and an agent that strands you.
Put all your machines, laptop, desktop, phone, on one private network with Tailscale. Now you can reach any of them from anywhere, securely, on a network you control. On my phone I use an app called Termius to command those machines, so if Hermes freezes I can do a gateway restart from wherever I am. This morning my desktop acted up while I was working out, and I fixed it from my phone without going home.
Without this, one frozen process at home means you're stuck until you're back at the keyboard. With it, the agent is never out of reach.
The takeaway
A lot of the people reading this will use Hermes as a fancy assistant and stop there, because the assistant version is comfortable and the revenue version asks you to change how you work. You have to give up doing the task and start designing the system that does it. Sometimes that trade is harder than it sounds.
Is it worth it? For us, absolutely yes. It's a real build. The first stretch is messy, you'll babysit it more than it saves you at the start, and you'll rip out half of what you set up.
But then the loops start compounding and one person starts operating like a team.
That's the actual prize.
If you're a business that wants AI systems built for you, check out https://www.singlebrain.com
For marketing help, go to https://www.singlegrain.com
For more like this, level up your marketing with 14,000+ marketers and founders in my Leveling Up newsletter, free: https://levelingup.beehiiv.com/subscribe
If you want to join our team, beat AI first ;)
