Most businesses have met AI as a chat window. You ask it something, it answers, and then you go and do the actual work yourself. That’s useful, but it isn’t where the time goes. The time goes into the hundreds of small tasks that keep a business running: answering the same customer questions, updating records, chasing paperwork, filling shifts, writing the weekly report.
The project started from a simple question: what if AI agents could hold a role, not just answer a prompt? Not a single assistant that does everything badly, but a team of agents that each have a job, the tools to do it and a manager they report to.
A workforce, in two directions
The project works in two directions at once.
The first is AI agents as digital staff. A business can stand up agents for specific roles, connect them to the systems they already use and supervise their work the way they would a new hire.
The second is AI for the human workforce. The same platform helps plan and coordinate people: forecasting demand, building schedules, matching the right person to the right shift and spotting gaps before they turn into problems.
The two halves meet in the middle. Agents take the repetitive work, people take the judgment calls, and the platform keeps track of who (or what) is doing what.
The roles
Every agent starts with a job description, written in plain language: what it’s responsible for, which systems it can use, what it is allowed to decide on its own and when it has to ask. Some of the roles the project focuses on:
- Customer support. Answers common questions from the knowledge base, looks up orders and accounts, drafts replies for anything sensitive and hands off to a person with the full context attached, so the customer never has to repeat themselves.
- Operations and admin. Keeps records in sync between systems, processes routine requests, chases missing information and closes out the small tasks that otherwise sit in someone’s inbox for a week.
- Sales research. Researches inbound leads, summarizes who they are and what they’re likely to need, and prepares a first draft of outreach for a person to review.
- Workforce coordination. Watches the roster, flags understaffed shifts, suggests who could cover based on skills and availability, and handles the back-and-forth of confirming changes.
- Reporting. Pulls numbers from several systems, writes the weekly summary and points out what changed and why it might matter.
None of these roles is glamorous, and that’s the point. They are the jobs that eat a team’s week.
What changes for the people
The fear with any automation is that it replaces people. What we’ve seen is closer to a change in what people spend their attention on.
A support lead stops answering password reset questions and starts reviewing the tricky cases the agent escalated, and improving the knowledge base so fewer cases need escalating. A manager stops building the roster from scratch on Sunday night and starts approving a draft that already accounts for leave, skills and expected demand.
The shift is from doing the work to directing and checking it. That only works if checking is fast, which is why so much of the project’s design goes into how agents report back, ask for approval and explain what they did. The guardrails article goes into that in detail.
Why roles and not one big assistant
We tried the “one assistant that does everything” approach early on. It was impressive in a demo and frustrating in practice. It had access to too much, its instructions kept growing, and when something went wrong it was hard to tell which part of its very long brief it had misunderstood.
Roles fix most of that:
- Narrow scope means better results. An agent that only does support can have focused instructions, the right tools and examples that actually match its work.
- Permissions follow the role. The support agent can read orders but can’t issue refunds above a threshold. The reporting agent can read almost everything and change nothing.
- Accountability is clear. When something goes wrong you know which agent did it, with which tools and why.
It’s the same reason businesses have job descriptions for people.