# Do I need 1 AI agent or a team of agents?

Start with 1. A single agent with the right tools handles most business jobs. A team of agents earns its cost on wide research. There, many threads run side by side. Anthropic measured the price in tokens, the units AI is billed in. An agent uses about 4 times a chat's tokens. A team uses about 15 times.

By Surinder Thakur, AI Product Designer and Builder. Published 17 Nov 2026. Topic: [Building an AI tool](https://www.surinder.design/work?type=building-an-ai-tool).

## When does a team of agents win?

When the work splits into parts that do not depend on each other. Wide research is the clear case: many sources, read at the same time.

Anthropic's research system uses 1 lead agent with several helpers. It scored 90.2% higher than a single agent on Anthropic's internal test. It also used far more tokens.

## When is 1 agent the better choice?

When every step needs what came before it. An invoice, a refund or an order moves through steps in a fixed order.

Anthropic names this limit itself. A team fits badly when every agent needs the same information. It also struggles when work passes back and forth. Most business workflows look like that.

A relay is different. Agents that hand work on in order, each with 1 job, still run 1 step at a time. That is 1 line of work, not a team running side by side.

## What does the extra cost look like?

Tokens are what a model reads and writes, and you pay for each one. A chat is the starting point.

An agent runs about 4 times that, and a team of agents about 15 times. Multiply by your monthly volume before you choose the bigger design.

## How do you decide without guessing?

Build 1 agent first and test it on real cases. Add a second only where tests show it failing. The gain must pay for the extra tokens.

The Blueprint maps the job step by step before any code. That map usually shows whether the steps depend on each other.

## In 1 picture

Start with 1 agent. Add more only where the tests say so.

- A chat: 1x
- 1 agent · most jobs: About 4x
- A team · wide research: About 15x

Token figures: Anthropic, How we built our multi-agent research system.

## Sources

- [Anthropic · How we built our multi-agent research system](https://www.anthropic.com/engineering/multi-agent-research-system)

## Next questions

- [What is an orchestrator agent, in plain words?](https://www.surinder.design/work/what-is-an-orchestrator-agent): An orchestrator is the lead agent in a team of agents.
- [What does an AI agent cost to run each month?](https://www.surinder.design/work/what-does-an-ai-agent-cost-to-run): 3 parts, paid every month. The model bill, which grows with every task the agent handles.
- [AI agent, chatbot or automation: which one does my business need?](https://www.surinder.design/work/ai-agent-chatbot-or-automation): The smallest one that finishes the job. A chatbot answers questions.
- [What does a custom AI tool cost?](https://www.surinder.design/work/what-does-a-custom-ai-tool-cost): A custom AI tool or agent costs from $20,000 to build here.

Page: https://www.surinder.design/work/one-ai-agent-or-a-team-of-agents
