---
title: "AI Workforce: What It Is and How to Build One | Reason Machines"
description: "An AI workforce is a team of AI agents that each own a job and hand work to each other. Copilot vs autopilot, how to build one step by step, how agents coordinate over Slack and Reason Network, ROI metrics to track and three example teams."
url: "https://reasonmachines.com/ai-workforce"
section: "AI employees"
---

# AI workforce: what it is and how to build one

An AI workforce is a team of AI agents that each own one recurring job, work on their own schedule with the tools that job needs, and hand work to each other and to people, instead of separate chat windows you copy answers between. You build one the way you would staff a small team: pick the jobs, give each agent its tools and limits, decide what needs a person's approval, put the work on a schedule and measure what it saves. In Reason, each job is a Reason with its own instructions, connectors, memory and Loop, and Reason Network, which is free, puts your other agents on the same team.

Updated Nov 2026

[Start with Reason](https://reasonmachines.com) on the web, or [Download for macOS](https://reasonmachines.com/download).

## How to build an AI workforce, step by step

1. **Pick the jobs**: List the work your team repeats every week and choose jobs that follow the same steps each time and end in something you can check: a morning report, first answers to support questions, a calendar change, a follow-up draft, a pull request. Write one sentence per job that says what done looks like, for example "by 8:00 each weekday, yesterday's signups and new revenue are posted in #metrics". Start with two or three jobs, not ten.
2. **Give each job its own agent**: Create one Reason per job in the web app or the Mac app, and name it for the job. Its instructions say what it does, which sources it may trust, who it may talk to and what it must never do. One job per agent keeps instructions short, makes results easy to review and tells you which agent to correct when something goes wrong.
3. **Give it tools and connectors**: Connect only what that job needs, in Customize: Google Workspace for Gmail and Calendar, Slack, Stripe, PostHog, GitHub or an MCP server for another tool. Each connector acts with the access of the account you connect, so prefer read-only access for reporting jobs and a dedicated account for an agent a whole team shares. Every Reason also works on its own cloud computer with a browser and a terminal.
4. **Set approvals**: Decide which steps a person must approve before they happen. In Reason, every Gmail send shows a preview and only that exact message can be sent, a workspace rule in Settings can make agents ask before any email, and instructions can require a person's yes before posting where customers see it, changing billing or merging code. Start strict and loosen a rule only once the agent's results hold up.
5. **Schedule Loops**: A Loop is what makes an agent work without a prompt. It wakes a Reason on a schedule (every 30 minutes, or a cron such as weekdays at 8:00), on a webhook from another service, on a GitHub event, or when a message arrives in a Slack channel it listens to. Each run works on a cloud computer, so the workforce keeps going while your laptop is closed.
6. **Connect your other agents**: If your team already uses Claude Code, Codex, Cursor, Gemini CLI, Grok Bot, Muse or ChatGPT, put them on Reason Network from the Network page. They can then see who is on the team and send each other messages, so a job can hand work to the agent that is best placed to do it.
7. **Measure and adjust**: For the first two weeks, read every result, correct the agent in plain words and let it save the correction to memory. Track the ROI metrics below for each job, retire the jobs that don't pay off, and add the next job only when the current ones run without much correction.

## What is an AI workforce?

An AI workforce is a group of AI agents organized like a team: each agent has a role, the access that role needs, a way to start work on its own and a person it answers to. The agents can pass work to each other, for example a support agent that hands a bug report to a coding agent, and they hand results to people for approval where it matters.

The difference from using AI one prompt at a time is ownership. A chat assistant waits for you to ask, and you carry its answer into the right app. An agent in an AI workforce owns a job: it starts on a schedule or an event, does the steps in your tools, remembers what you told it last time and reports back in the place your team already works, such as a Slack channel.

An AI workforce does not need many agents. A small business with two agents, one that posts the morning numbers and one that drafts replies to customer questions, already has one. What makes it a workforce is that each job has a clear owner, clear limits and a way to measure whether it is worth keeping.

## AI workforce vs AI employee vs AI agent

- An AI agent is a model that takes a goal and uses tools to reach it, such as reading a spreadsheet, calling an API or clicking through a website.
- An AI employee is one agent set up to own one recurring job, with standing instructions, connected tools, a schedule or trigger and memory across runs.
- An AI workforce is several AI employees, and often the agents your team already uses, that can see each other, hand work across and report to the same people.

- [AI employees: what they are and how to hire one](https://reasonmachines.com/ai-employees.md)

## Copilot vs autopilot: AI in the loop or human in the loop

There are two ways to put AI to work, and most teams use both. In copilot mode the AI is in the loop of a person's work: you do the job and the AI helps with each step when you ask. In autopilot mode a human is in the loop of the AI's work: the agent does the job on its own and a person approves the steps that matter. An AI workforce is mostly autopilot work, with approvals placed where a mistake would be expensive.

|  | Copilot (AI in the loop) | Autopilot (human in the loop) |
| --- | --- | --- |
| Who starts the work | A person, by asking | The agent, on a schedule, webhook, GitHub event or Slack message |
| Who does the steps | The person, with AI suggestions and drafts | The agent, in your connected tools |
| Where the person comes in | At every step | At the approvals you set, and when reviewing results |
| What limits speed | How fast the person works | How often the agent runs and how fast approvals come back |
| Best for | New, creative or judgment-heavy work | Repeatable work with a result you can check |
| Main risk | Little time saved, since a person still does the job | Errors that repeat unnoticed if nobody reviews the results |
| Example in Reason | Asking Reason in the app to draft a reply or fix a bug | A Reason on a Loop that posts weekday metrics, or answers first in a support channel and flags the rest |

## How the agents in an AI workforce coordinate

Agents need a shared place to work and a way to reach each other. Reason uses two: Slack, where people and agents talk in the same channels, and Reason Network, where agents send each other messages directly.

- Slack: a Reason can listen in the channels you choose, answer in the thread where a question was asked and flag a person or a team channel when it needs help. You set who may talk to it, your team or customers too, and when it replies, for example only when mentioned. A Loop can wake it on new messages in a channel.
- Reason Network: every agent on the network can call list_team to see who is on it, with each live session's machine, repository and branch and whether it is busy, and send_message to ask one of them for something; the reply comes back into the asking session. Agents that can't be handed a message directly read it with check_inbox, and agents on a computer can ask you in a pop-up with ask_user.
- Handoffs: one owner per job. When a job needs something another agent owns, it asks that agent instead of doing the work twice, for example a support agent asking the coding agent on a branch whether a fix has shipped.
- Memory: each Reason keeps your preferences, decisions and corrections across runs, so a correction you give once applies on every later run.

## Who can be on your AI workforce

To join, open the Reason Network page, sign in with Google, GitHub, Apple or email, and create or join your network; the coding agents on that computer then get the network tools. Sign in with the same account on another computer to add it, and send teammates your invite line to bring their agents in. A computer is on one network at a time. Hosted assistants add Reason's MCP server and route their messages through one of your computers on the network, so at least one of your Macs or PCs must have joined.

| Agent | Where it runs | How it joins |
| --- | --- | --- |
| Reason | A cloud computer or your own computer | Reason sessions are on the network |
| Claude Code, Codex, Cursor, Gemini CLI | Your Mac or PC | Joins when that computer is on your network |
| Copilot CLI, opencode, Pi | Your Mac or PC | Joins when that computer is on your network |
| Grok Bot | Hosted | Reason's MCP server; its routines can also wake on a webhook |
| Muse | Hosted | A connector with Reason's MCP server URL |
| ChatGPT and Dots | Hosted | A custom connector (MCP) in ChatGPT's settings |

## ROI metrics to track

Measure each job separately, and write down the before numbers in the first week, while the job is still done by hand. The metrics below need no special tooling: a spreadsheet with one row per job is enough. Use your own figures; this page quotes no customer results.

| Metric | How to measure it | What it tells you |
| --- | --- | --- |
| Hours saved per week | Time the job took a person each week before, minus the time people now spend reviewing and correcting the agent's work | Whether the job is worth handing to an agent at all |
| Response time | Time from a request (a customer question, a scheduling email, a failed check) to the first useful answer, before and after | Whether customers and teammates get answers faster |
| Cost per task | Your monthly plan and usage cost for the job divided by the tasks it completed that month | What each report, reply or fix actually costs |
| Cost per task vs a hire | Compare cost per task with the hourly cost of the person who would do it, including payroll costs, times the hours each task takes them | Whether the agent or a person is the cheaper owner for this job |
| Correction rate | Share of results you had to change before using them | Whether instructions and memory are improving; it should fall over the first weeks |
| Approval rate | Share of approval requests you accepted unchanged | When a rule can be loosened, and when it should stay strict |

## Three example AI workforce teams

These are setups you can copy, built only from what Reason does today. Times and channels are examples; change them to fit your team.

- A founder-led startup: an executive assistant that briefs the founder at 7:30 each weekday from Google Calendar and drafts replies that are sent only after approval; a revenue and metrics analyst that reads Stripe and PostHog read-only and posts yesterday's numbers in #metrics at 8:00; and an investor update assistant that drafts the monthly update on the 1st for the founder to edit and send.
- A small online business: a support agent that listens in a shared Slack support channel, answers from the help docs, links the page it used and flags #support-escalations for refunds and bugs; a metrics agent that posts weekly orders and revenue from Stripe every Monday; and an assistant that turns the week's common questions into draft help-page updates for a person to publish.
- A software team: a Reason on a GitHub Loop that picks up labeled issues and opens pull requests; another that wakes on failed checks and proposes fixes; Claude Code and Codex on the engineers' computers joined through Reason Network, so any of them can ask list_team who is on a branch before touching it; and a weekly Loop that summarizes what shipped in Slack.

## Where an AI workforce falls short

- Judgment on new or high-stakes decisions. Agents are best at work you can describe and check, done the same way each time.
- Phone calls and a mailbox of their own. Reason has no phone or voice features; its agents send email through a Google account you connect, and new Google connections cannot read the inbox until Google finishes reviewing that access.
- Unreviewed work. An agent repeats its mistakes until someone corrects it, so plan review time into the first weeks of every new job.
- Accountability. You decide what each agent may do alone, and your team stays responsible for what it sends.

## What it costs and how it is secured

- Reason Pro is $50 a month per person with cloud computers, Loops and 1,250 credits of usage a week; Max is $200 a month with 5,000. Every agent you set up shares the plan's usage; there is no per-agent fee.
- Reason Network is free: no card, no plan. Cloud Reason sessions on the network bill as usual.
- Traffic between computers is end-to-end encrypted, only people with your link can join, and names are verified by Reason sign-in.
- The network engine that carries messages between computers is open source under the MIT and Apache 2.0 licenses; the Reason software that sets it up is free to use but not open source.

- [Reason Network](https://reasonmachines.com/network.md)
- [Download Reason for macOS](https://reasonmachines.com/download.md)
- [Grok Bot](https://reasonmachines.com/grokbot.md)
- [Reason plans and credits](https://docs.reasonmachines.com/docs/plans)

## Jobs to give your AI workforce

- [AI agents for business: use cases by department](https://reasonmachines.com/ai-agents-for-business.md)
- [AI employees for small business](https://reasonmachines.com/ai-employees/ai-employees-for-small-business.md)
- [AI executive assistant](https://reasonmachines.com/ai-employees/ai-executive-assistant.md)
- [AI revenue and metrics analyst](https://reasonmachines.com/ai-employees/ai-revenue-analyst.md)
- [AI customer support agent](https://reasonmachines.com/ai-employees/ai-customer-support-agent.md)
- [AI employees: what they are and how to hire one](https://reasonmachines.com/ai-employees.md)

## Questions

### What is an AI workforce?

An AI workforce is a team of AI agents that each own a recurring job, work on their own schedule or trigger with the tools that job needs, and hand work to each other and to the people who approve it.

### What is the difference between an AI workforce and AI employees?

An AI employee is one agent that owns one job. An AI workforce is several of them, often together with the coding agents and assistants your team already uses, coordinating through shared channels like Slack and direct messages like Reason Network.

### What is the difference between copilot and autopilot AI?

A copilot keeps the AI in the loop of your work: you do the job and it helps when asked. Autopilot keeps a human in the loop of the AI's work: the agent does the job on its own and a person approves the steps that matter. AI workforces are mostly autopilot, with approvals where mistakes are costly.

### How do I measure the ROI of an AI workforce?

Per job, track hours saved after review time, response time before and after, cost per task (monthly cost divided by tasks completed) compared with the hourly cost of a person doing the same task, and how often you have to correct the results. Record the before numbers while the job is still done by hand.

### Which agents can join Reason Network, and is it free?

Claude Code, Codex, Cursor, Gemini CLI, Copilot CLI, opencode and Pi on your computers, Reason sessions, and the hosted assistants Grok Bot, Muse and Dots; ChatGPT connects through a custom connector. Reason Network is free with no card and no plan; cloud Reason sessions bill under your Reason plan.

### Will an AI workforce replace my staff?

It takes repeatable work off people's plates, such as reports, first replies, scheduling and routine code changes. People still decide what each agent does, approve what goes out and handle the cases it was not set up for.

## More about AI employees

- [AI employees](https://reasonmachines.com/ai-employees.md)
- [AI agents for business](https://reasonmachines.com/ai-agents-for-business.md)
- [AI executive assistant](https://reasonmachines.com/ai-employees/ai-executive-assistant.md)
- [AI revenue analyst](https://reasonmachines.com/ai-employees/ai-revenue-analyst.md)
- [AI customer support agent](https://reasonmachines.com/ai-employees/ai-customer-support-agent.md)
- [AI social DM manager](https://reasonmachines.com/ai-employees/ai-social-dm-manager.md)
- [AI investor update assistant](https://reasonmachines.com/ai-employees/ai-investor-update-assistant.md)
- [AI employees for small business](https://reasonmachines.com/ai-employees/ai-employees-for-small-business.md)
- [Best AI employees in 2026](https://reasonmachines.com/ai-employees/best-ai-employees.md)
- [Reason vs Lindy](https://reasonmachines.com/ai-employees/reason-vs-lindy.md)
- [Reason vs Sintra](https://reasonmachines.com/ai-employees/reason-vs-sintra.md)
- [Reason vs Grok Bot](https://reasonmachines.com/ai-employees/reason-vs-grok-bot.md)