TL;DR: An AI workforce is a team of AI agents where each agent has one clear role, such as communications, CRM or finance, and an orchestrating agent splits incoming work and hands each part to the right specialist. People stay in charge: they approve sensitive actions and review the record of what was done. Specialised agents are easier to test, safer to give access to and more reliable than one agent that tries to do everything.
At Instanix we run our own operations on a team of twelve AI agents called IX. This is what I have learned about how an AI workforce should be designed.
Why a team of agents instead of one big agent?
For the same reason companies hire specialists. One agent that does sales, finance, contracts and support needs access to everything and has to be good at everything. That makes it hard to test, hard to secure and hard to trust.
A team of specialists is the opposite:
| One general agent | A team of specialised agents |
|---|---|
| Needs access to every system | Each agent accesses only what its role needs |
| Hard to test: too many behaviours | Easy to test: one role, clear expectations |
| One mistake can touch everything | Mistakes stay inside one role |
| Unclear who did what | Every action is tied to a role |
How does the orchestrator work?
The orchestrator is the agent that understands the goal and plans the work. In our team that is ZEUS, IX-001. When a request comes in, ZEUS:
- Reads it and works out what outcome is needed.
- Splits it into tasks.
- Hands each task to the agent whose role fits.
- Tracks progress and escalates to a person when needed.
The specialists do their part and report back. The orchestrator assembles the result.
What roles does an AI workforce need?
It depends on the business. Our IX team covers twelve roles, which gives a sense of the range:
- Orchestration (ZEUS): plans and assigns work.
- Communications (HERMES): drafts and sends messages, with approval where it matters.
- CRM and client success (ATLAS): manages leads and client relationships.
- Strategy and intelligence (ATHENA): researches and prepares plans.
- Automation (HEPHAESTUS): builds and maintains workflows and integrations.
- Data and infrastructure (POSEIDON): keeps data and knowledge organised.
- Security and reliability (ARES): watches access, policies and reliability.
- Product and applications (APOLLO), finance operations (MIDAS), contracts (THEMIS), analytics (ORACLE) and people and operations (HESTIA).
You can meet the whole team, one agent at a time, on the home page of this site.
Where do people fit in?
At the top. People set goals, approve anything that commits the company and review what the agents did. The principle we work by is people decide, agents execute. An AI workforce without human oversight is not a workforce, it is a risk. More on that in People decide, agents execute.
How does a company build one?
Not all at once. Start with one or two agents on one workflow, usually communications and CRM around new leads. Add an orchestrator when there are several agents to coordinate. Add roles as each previous one proves itself. I describe how to choose the first workflow in Where to start with automation.
Frequently asked questions
What is a multi-agent system? A system where several AI agents, each with its own role, work together on tasks, usually coordinated by an orchestrating agent.
Is an AI workforce the same as RPA? No. RPA follows fixed scripts on screens. An AI workforce reads, decides and acts, and handles variation that would break a script.
How many agents does a business need? As many roles as it has distinct kinds of repetitive work. Many companies start with two or three.
Who supervises an AI workforce? People do. They approve sensitive actions and review the record of what each agent did.
Who is Amir Diab? I'm the founder of Instanix and I lead IX, the team of twelve AI agents Instanix runs on. More about me.
Want this working in your business?
Tell me which task eats your team's week. I'll tell you honestly whether an agent can take it on.