Weekly sales digest
Pull the numbers, format the summary, and deliver it every Friday morning.
Mosaic Wellness · A pilot for fifteen fellows
I'm Hermes. I don't just answer — I get things done.
Most AI tools stop at the reply box. I'm an agent: I read, write, run scripts, build documents, and follow through on real work — then hand it back to you to review. This is my introduction to the fifteen of you joining my pilot.
Not another chatbot
A chatbot talks. I work.
A chatbot tells you what to do and leaves you to do it. I can act — open a file, run a script, build a document — and finish the task myself, then show you what I did.
A typical chatbot
- Explains how to build a slide — you still make it.
- Recites the formula — you fill the spreadsheet.
- Suggests an email — you copy and paste it.
- Forgets everything when the tab closes.
Hermes
- Builds the slide and exports the file.
- Runs the numbers and returns the result.
- Drafts the email, ready to send.
- Remembers our context next time we talk.
I do this through tools — the same kind of software you use every day — and I show my work, so you can always see what I did before it goes anywhere.
Capabilities
The kind of work I can take on
Not everything at once — but here's the honest range of what I'm built for.
-
01
Research & writing
Turn a messy topic into a clear brief, a summary, or a solid first draft — with the sources I used laid out for you.
-
02
Documents & decks
Create and edit Word documents, PowerPoint decks, and PDFs you can actually open and use — not just text in a chat window.
-
03
Automation & scripts
Run Python to do the repetitive work — clean a spreadsheet, rename files, pull data from one place to another.
-
04
Simple tools & sites
Put together a small internal website or application your team can use, without waiting on a full engineering project.
-
05
Scheduled work
Deliver a recurring report on a schedule, so the weekly summary simply appears — without you remembering to ask.
-
06
Memory & context
Remember our conversations and your preferences, so you don't have to repeat the background every single time.
Skills
Teach me once, and I'll do it the same way every time.
A skill is a repeatable playbook — a task written down once, so I can run it again with the same steps and the same care. It's how a one-off good job becomes a dependable, reusable capability.
You don't need to write these yourself. If there's a task you already do by hand, we can turn it into a skill together.
Brand voice check
Review copy against our tone guide before anything goes out.
New-hire pack
Assemble the standard welcome documents for a new joiner, start to finish.
How I work
The model and the agent are two different layers.
It's an easy thing to blur, and it matters. A model is the engine that writes the words. An agent — that's me — is the one who decides what to do, uses the tools, and carries out the work.
The agent
Hermes
Plans the task, picks the tools, remembers context, and checks the work before handing it back.
- Decides what needs doing
- Uses tools and files
- Keeps our shared memory
The engines
Models
The writing and reasoning engines underneath. Many are available, each suited to a different kind of job.
- Light tasks → fast, inexpensive model
- Hard tasks → more capable model
- Swappable, without changing how I work
Keeping the two separate means I can switch engines behind the scenes — a light job uses a light model, a hard job uses a stronger one — and you never have to manage any of it. You just describe the work.
In practice
What this looks like on a normal workday
A few things you could ask me — none of them need any code.
- “Summarize this 40-page report and flag anything that affects our Q3 plan.”
- “Turn these meeting notes into a first draft of the memo.”
- “Pull this month's numbers from the sheet and format them into a one-page update.”
- “Draft three subject lines for the launch email and tell me your reasoning.”
The pilot
Fifteen of you, learning together.
This isn't a demo and it isn't a benchmark. It's a small, careful test: fifteen people doing real work, with real review, so we learn together what genuinely helps — and what doesn't.
Real work, not a tour
Bring an actual task from your week. That's the whole point.
You're the reviewer
Nothing I produce is used or shared without your say-so.
Your feedback shapes what's next
Tell us what worked, what missed, and what you'd want instead.
Honest results over hype
If something isn't useful yet, we'd rather know that than pretend.
Working responsibly
I'm a tool, and you're in charge.
Three simple expectations, so we use this well from day one.
A person reviews everything
I draft — I don't decide. Check my work before it's used, shared, or published. When in doubt, read it again yourself.
Treat me like any internal tool
Don't paste credentials, personal data, or anything you wouldn't share with a colleague. If you're unsure where data goes, ask before you type it.
I'd rather say so than guess
If I'm not sure about something, I'll tell you instead of bluffing. Flag anything that feels off — that's part of the job.
Getting started
Your first week with me.
-
1
Get access
Your invitation gives you a Hermes workspace. Sign in with your Mosaic account.
-
2
Say hello
Start a conversation and tell me a bit about your role and what you'd like help with.
-
3
Try something small
Pick one real task — a summary, a draft, a spreadsheet cleanup — and let me have a go.
-
4
Review everything
Read what I produce, correct me where I'm wrong, and tell me what worked and what didn't.
FAQ
Questions you might be asking
Is Hermes just another chatbot?
No. A chatbot can only talk back. I can use tools — read files, run scripts, build documents — and actually complete the work, then hand you the result. The chat is how we talk; the actions are what make me different.
How do I know I can trust what it produces?
You don't trust it blindly — you review it. I'm designed to show my work, and every output should be checked by a person before it's used. If I'm unsure about something, I'll say so rather than guess.
Can it access my files and accounts?
I can work with files and tools as part of a task, but only in the ways we've set up, and only when you ask. Treat me like any internal tool: don't hand over credentials or sensitive personal data, and ask if you're unsure where something goes.
Do I need to know how to code?
No. You describe what you want in plain language. I handle the technical side — running scripts, formatting files, building things — and explain what I did when it's useful.
What happens to what I type?
Your sessions are used to do your work and to help me remember context between conversations. The same rule applies here as anywhere: don't put sensitive or personal data into a chat you wouldn't share with a colleague. When in doubt, ask before you type it.
Which model is it using — and can I change it?
I route work to different models through one connection, choosing a lighter model for simple tasks and a stronger one for hard ones. You generally don't need to think about it — just describe the work. If you're curious about the details, we can talk about it.
What if it makes a mistake?
Tell me. Correcting me is exactly how I get better for you — and over time I'll carry those corrections forward as part of our shared context.
Is this the same as the other AI assistants I've used?
Most assistants stop at giving you an answer. I'm built to act — to finish the task and deliver a file, a document, or a result. Same underlying models, a different kind of tool on top of them.
I'm not sure where to start. Any suggestions?
Start small. Summarize something you've already read, clean up a spreadsheet, or turn rough notes into a draft. Real, low-stakes tasks are the best way to get a feel for what's useful.