I Gave Myself a Staff of 8 AI Employees To Run My Workday

The work around the work

Most of us are not buried by our actual job. We are buried by the layer around it.

Think about a normal Tuesday. There is the email that needs a two-line answer, sitting there since 9 a.m. There is the meeting you are about to walk into without having looked at who called it or why. There is the follow-up you promised on Friday and have not sent. There are the notes from this morning's call that you meant to turn into next steps. There is the Teams message with your name in it that you saw, meant to reply to, and lost under twelve newer ones. None of these is hard. Each takes a few minutes. Added up across a week, they are most of the week.

And then review season arrives, and you sit down to write up what you accomplished, and you cannot remember half of it. The big things you recall. The steady stream of smaller wins that actually moved your accounts and your relationships forward has evaporated.

I wanted help with that layer specifically. Not another chat window I have to open and prompt from scratch every single time. Something that watches the stream for me, pulls out the few items that genuinely need a human, and hands them to me mostly finished. Something that also remembers what I did, so I do not have to.

So I built it. It is called the Dream Team, and it runs on my own machine. I've detailed exactly what it is, how it works and how to install it for yourself. You can find out more by watching this video or continue reading below:

Where this fits

We have all watched the demos. In the last couple of years AI learned to write, summarize, answer almost anything. It is impressive, and yet, if you ask a lot of people what they use it for day to day, the answer is often a shrug. The capability has been there for a while. The everyday habit, for many people, has lagged behind it. The hard part was never raw intelligence. It was finding the spot where AI fits into your day and earns a permanent place.

Microsoft has been closing that gap quickly, and I lean on those tools myself. Microsoft 365 Copilot put AI right inside Outlook, Word, Excel, and Teams, so it works in the document and the inbox instead of a separate tab. Copilot Chat grounds its answers in your real work content, so you are not pasting in context from scratch. Copilot Cowork takes the next step into agentic, multi-step work rather than only helping in the moment. Each of those is genuinely useful, and each has its place in a normal week.

So why build another one? Because I wanted to take that same agentic direction and push it to the edge for my own workflow. I wanted a standing team of role-specific coworkers that watches my whole day at once, starts on its own instead of waiting to be asked, keeps a quiet record of what I got done, and runs on my own machine with me holding the approval switch on anything that reaches another person. The Dream Team is not a replacement for the tools above. It is what happened when one person kept pulling that thread as far as it would go.

There was one more thing I cared about. A lot of AI still assumes a generic user doing generic tasks, and a salesperson's day, a recruiter's day, and a professor's day are completely different shapes. I wanted something I could mold to a specific role rather than the lowest common task, so the Dream Team is built to be reshaped. Add the coworkers that fit your work, remove the ones that do not, and set how far each one can go on its own. That combination, a real tie-in to your everyday workflow plus the freedom to shape it to your role and keep control, is what turns AI from something impressive into something you lean on. A team that already knows where the work is.

What it actually is

The Dream Team is a small command center that runs locally, paired with eight digital coworkers. Each coworker has one job. You do not manage eight things. You talk to one of them, Major, the chief of staff, and Major hands work to the rest.

Here is the roster:

- Major, the chief of staff. You talk to Major. Major routes everything else.

- Riley, on the inbox. Triages mail and drafts replies.

- Mina, on meetings. Preps you before, and turns notes into action items after.

- Reese, on research. Answers questions with real sources instead of guesses.

- Tilly, on scheduling. Watches for conflicts and the meeting that has no breathing room around it.

- Dash, on the dashboard. Keeps your numbers, approvals, and status current.

- Drew, on content. Turns a request into a document, a deck, or a spreadsheet.

- Logan, on publishing and the record. Keeps a log of what you got done.

The piece that makes it usable rather than nerve-racking is the approval inbox. Anything that would touch another person waits for you. The team does the preparation. You make the call. Then it acts. Approve a reply and it actually goes out. Tell it to give someone a thumbs up and it gives them a thumbs up, not a paragraph. It only leaves you a draft when you ask for one. The team can be busy on your behalf all day, and nothing reaches another human until you say so.

The other quiet payoff is the record. As the team works, Logan keeps a running account of what you actually accomplished, written in plain language, the kind of thing you would want in front of you at review time. Give it your goals and how you are measured, and it frames your work against them. That part stays completely private and lives only on your machine.

Everything runs at a local address on your own computer. Nothing is shipped off to a server somewhere. The team prepares, you approve, and only then does anything leave.

How it bends to fit you

A fixed product would have been easier to build. It also would not have been mine, and it would not be yours. Flexibility was the point, so I leaned into it in four ways.

The roster is yours to shape. You can add your own employee by pointing the app at a set of instructions you already use, and it walks you through onboarding the new hire. You can remove any employee except Major, and their history is kept, so you can bring them back later. Start with the eight, grow your own, trim the ones you do not need.

You decide how far each one can go. Every employee has a trust level. Draft means it prepares and you send. Assist means it can handle reversible housekeeping on its own but waits for your nod before anything goes to another person. Autonomous means it runs its own lane from start to finish. You can dial any employee from fully supervised to fully hands-off, and you can change your mind at any time. One rule never bends: confidential content always waits for you, no matter the level.

It adapts to who is installing it. During setup it quietly checks whether you are signed in with a Microsoft account and adjusts on its own. If you work at Microsoft, it can pull in some extra depth for a couple of the employees. If you do not, you still get the entire team, running on what is in the package and what is already built into Scout.

The same team, a different job

I built mine around email, meetings, and follow-ups, because that is the shape of my work. The reason I keep saying the roster is yours is that the same machinery retargets cleanly to a very different job. The roles do not change. What they watch for, and what they produce, does.

Picture a recruiter. The stream is candidate replies, interview debriefs, and panels that never line up. The inbox agent triages candidate mail and drafts the next note. The scheduling agent wrangles five calendars into one slot. The content agent turns debrief notes into a clean summary or a draft offer. The record quietly tracks where every open role stands, so the Friday update writes itself.

Picture a founder or a small business owner. The stream is customer questions, an investor update that is overdue, and the contractor you keep forgetting to pay. The team preps you for the investor call, drafts the update from what actually happened that month, and keeps a running list of the loose threads you would otherwise drop.

Picture a professor or a researcher. The stream is student email, grant deadlines, and a field that publishes faster than anyone can read. The research agent answers with real citations instead of confident guesses. The meeting agent turns office hours and lab meetings into action items. The record captures what you produced across a term, which is exactly what nobody remembers when it is time to write it up.

None of those people need to touch code. They add the employees that fit, remove the ones that do not, point them at the workflows they already live in, and set how much each one is trusted to do on its own. Same eight roles, a different job, because the shape is yours to set.

How it was built

This is the part of the project I find most interesting.

I am not a career software engineer. I built this by working with Microsoft Scout, which is the same AI assistant the app runs on. I described what I wanted in plain language, we built a version, I ran it against my own real workday, and we fixed whatever broke. That loop ran for weeks. The changelog still carries the scars. There is the release where approving an item finally sent it for real instead of quietly saving a draft. There is the one where a background process was overriding my approvals because it was still following an old rule. There is the one where prepared documents stopped failing to open. Every one of those was a real thing that annoyed me on an ordinary afternoon, and the fix shipped a day or two later.

Under the hood it is deliberately small and boring, which is a compliment. The app is plain Python with no extra packages to install. It serves a local dashboard in your browser and keeps its data in a local database file. The intelligence does not live in that code. It lives in a set of skills, which are plain-language instruction files that tell Scout how the team thinks, how Major routes work, what each employee may do on its own, and where it has to stop and ask. A few background automations keep the team awake without you doing anything: a quick status check every few minutes, a wider sweep of your signals on a slower beat, a brief in the morning, a wrap-up in the evening.

Where to find it, and how to run it

The Dream Team is on GitHub. It is public, free, MIT licensed, and meant for demo purposes.

To install it, open Scout and paste this in a new chat:

Install The Dream Team from https://github.com/ShervinShaffie/dream-team-for-microsoft-scout. Read INSTALL-WITH-SCOUT.md in that repo and follow it exactly, including the stop conditions.

Once you've installed everything and are running the dream team, visit the Skill Shack for ideas around the types of skills your new employees can leverage.

Why this might matter to you

I am not going to pretend this one app is essential to your life. The point is bigger than the app.

The building blocks are finally within reach. An AI agent that runs on your own machine, a handful of plain-language instruction files, and a small local app are enough to assemble real help that fits the exact shape of your day, whatever that shape happens to be.

What I find exciting is that you do not have to wait for a vendor to decide your workflow is worth supporting. You can build the help you need, change it when your work changes, keep it on your own machine, and stay in control of what it does on your behalf. I built a team of eight. You can build whatever you want.

I also know the catch. Very few people have access to Microsoft Scout right now, so very few can run this today. I am sharing it anyway, because the bigger point is not this one app. It is to show how far AI has already come, and what becomes possible when it is wired into your real work and shaped to your job rather than sitting in a separate window. This is a glimpse of where things are heading. My hope is that the tools behind it reach a lot more people soon, and that when they do, you already have a sense of what you could build for yourself.

Contact Us

Questions? Reach out anytime.

Phone

+1-949-667-1624

© 2020–Present Collaboration Simplified. All rights reserved.

Collaboration Simplified® is a registered trademark of Shervin Shaffie.

email

info@cosi.pro