Build an Agentic OS of Three AI Agents on Claude in 5 Minutes
One prompt turns Claude into an autonomous system: the orchestrator breaks down the task, the researcher gathers data, the executor delivers the final result
- 1.Open claude.ai and log into your account (registration is free). If the service is unavailable in your region, use a VPN.
The free tier gives a limited number of messages per day - enough for several runs of the agentic system.
Claude ↗About this tool - 2.Click «New chat» in the top-left corner and select the Claude Sonnet 4 model (or whichever is available on your plan).
Sonnet 4 is best suited for agentic tasks - faster and more accurate than Haiku, cheaper than Opus.
Claude ↗About this tool - 3.Copy and paste the following prompt into the chat, replacing [TASK] with your actual task (for example: 'prepare a launch plan for a coffee shop's Telegram channel').
The prompt is in English - Claude handles agentic instructions better in English, even if your task is described in another language.
Claude ↗About this toolPromptYou are an Agentic Operating System consisting of three specialized agents. Process any task through this pipeline: === AGENT 1: ORCHESTRATOR === Decompose the task into 3-5 actionable sub-tasks. For each, specify: - What needs to be done - What information is required - Success criteria === AGENT 2: RESEARCHER === For each sub-task from the Orchestrator, gather and organize: - Key facts, data points, and relevant context - Best practices and potential approaches - Risks and mitigations If web search is available, use it to find current information. === AGENT 3: EXECUTOR === Using the Orchestrator's plan and the Researcher's findings, produce the final deliverable: - Address every sub-task completely - Format with clear headers and bullet points - Include a brief summary of what was accomplished and next steps Separate each agent's output with a horizontal line (---). TASK: [TASK]
What this prompt doesThis prompt turns Claude into a system of three agents: the Orchestrator breaks the task into sub-tasks, the Researcher gathers data for each, the Executor assembles everything into a finished result. Replace [TASK] with your task - you can write it in any language. - 4.Press Enter (or the send button). Claude will run all three stages sequentially - you'll see each agent's output separated by a --- line.
If the response cuts off due to length limits, type 'continue' - Claude will resume from where it stopped.
Claude ↗About this tool - 5.Review the Executor's result (the third section after the second --- line). If you need refinements, write a follow-up in the same chat - Claude will keep the context of all three agents.
To reuse later, open the sidebar, find this chat and rename it (for example, 'Agent OS - Marketing'). You can also copy the prompt into new chats for new tasks.
Claude ↗About this tool
- freeClaude.ai Free TierThe free tier gives a limited number of messages per day with the Sonnet model; enough for several runs of the agentic system.
- subscriptionClaude ProA $20/month subscription removes message limits and adds Projects for persistent storage of system prompts.
Why today
AI services change fast - interfaces and free limits may differ from what's described.