How many agents does a small business need?

Usually one to start with, against a complete journey. Businesses that deploy five agents on day one tend to end up with five half-working ones and no clear result.

Can agents talk to each other?

Yes, and that is where the value compounds — a sale triggering procurement, scheduling and customer onboarding. It also raises the stakes, so shared limits and audit come first.

Who is accountable when an agent gets it wrong?

The business is, always. That is why authority limits, approvals and an immutable record of what the agent used and decided are part of the build rather than an afterthought.

Do staff need training?

Mainly in approving and correcting work rather than doing it. The teams that get the most out of an AI workforce are the ones that review its output honestly in the first few weeks.

AI WORKFORCE · DEFINITION

What is an AI workforce?

An AI workforce is a set of specialist AI agents deployed as roles rather than as chat windows. Each agent has a defined job, the tools and knowledge it needs, an authority limit, a rule for when it must ask a human, and a performance measure. Managing one is closer to managing staff than to using software.

Forth Systems engineering team · Last reviewed September 2026

Roles, not a chatbot

A chatbot waits to be asked. An agent with a role has objectives and triggers: it wakes up when an enquiry arrives, when a quote goes quiet for 48 hours, when a delivery date moves, when an invoice ages past terms.

That is why the useful unit of deployment is a role — Sales, Operations, Finance, Procurement, Customer — each with a narrow remit and real accountability, rather than one general assistant with vague access to everything.

Every agent needs an employment contract

Before an agent touches a live system it should have a written constitution: role, objective, tools, knowledge, authority, financial limit, approval rules, escalation path, success metric, budget and audit requirement. If you cannot write that down, the agent is not ready to be deployed.

Managing performance

An AI workforce should be reviewed the way a team is. For each agent: work handled, average response time, human escalations, error rate, cost to operate and the outcome it influenced. This changes the question from what AI costs to what the AI workforce returned.

Where an AI workforce is weakest

Judgement calls with thin data, relationship moments, anything adversarial or contractual, and anything where being wrong is expensive and hard to reverse. Design the escalation path for those cases first, not last.

Common questions

How many agents does a small business need?: Usually one to start with, against a complete journey. Businesses that deploy five agents on day one tend to end up with five half-working ones and no clear result.
Can agents talk to each other?: Yes, and that is where the value compounds — a sale triggering procurement, scheduling and customer onboarding. It also raises the stakes, so shared limits and audit come first.
Who is accountable when an agent gets it wrong?: The business is, always. That is why authority limits, approvals and an immutable record of what the agent used and decided are part of the build rather than an afterthought.
Do staff need training?: Mainly in approving and correcting work rather than doing it. The teams that get the most out of an AI workforce are the ones that review its output honestly in the first few weeks.

NEXT STEP

Find out how autonomous your business is today, and what the realistic 90-day target looks like.

Get your Autonomy Score