Updated 30 October 2026: To build an AI agent platform with Claude, you combine Claude Managed Agents, Anthropic's hosted service that runs agents for you, with a normal web app for sign-in, data and billing. Buildcamp's course builds Orion, a multi-user agent platform, using Claude Managed Agents, Next.js, Supabase and Vercel, with Claude Code and Codex writing the code. You do not need prior coding experience.
Who this is for
This guide is for you if you want to build a product where your users hand work to AI agents: research a folder of documents, produce a report, fill in a spreadsheet. It is also for you if you have read about Claude Managed Agents and want to know how they become a real app that other people can sign in to and use.
If you just want to understand what Managed Agents are first, start with Claude Managed Agents explained, then come back here.
What an "agent platform" actually is
An AI agent is an AI model that can take actions, not just reply. It can run commands, read and write files, search the web and use connected tools. An agent platform is the app around those agents: user accounts, the files they work on, a way to start jobs, live progress, results and a record of what it all cost.
Claude Managed Agents handles the hardest part, which is running agents safely in the cloud. Anthropic describes it as a pre-built, configurable agent harness that runs on managed infrastructure and is suited to long-running work. It is built around four ideas:
- Agent: the model, instructions and tools your agent uses.
- Environment: the secure container where the agent works.
- Session: one running piece of work, such as a single task.
- Events: the messages that flow between your app and the agent while it works.
Your job is to build everything around that.
The path, step by step
- Plan the product. Decide who your users are and what one job the agent will do for them. In the course, Orion turns your users' Google Drive files into finished documents.
- Set up your coding agents. You use Claude Code and Codex to write the app. Our guide on One AGENTS.md for Claude Code, Codex and Cursor shows how to give them shared instructions.
- Build the web app. Use Next.js for the website and Supabase for the database and sign-in. Supabase's row level security (rules in the database that decide who can see which rows) keeps each user's data private.
- Connect your users' files. The course uses Pipedream to connect Google Drive, then reads Google Docs, Sheets, PDFs and Word documents.
- Run agent "missions". Each mission is a Managed Agents session. Your app starts it, streams live status to the screen and saves the results.
- Generate real outputs. Have the agent produce Docs, Sheets and PDFs your users can open straight away.
- Track usage and cost. Record what each mission used so you can see costs per user, then add an admin dashboard to keep an eye on everything.
- Ship it. Save your work to GitHub and deploy on Vercel with your own domain name.
The full build is in Build an AI agent platform with Claude Agents.
What you will need
- A Claude Console account and API key. Managed Agents runs through the Claude API, so it needs an API key from the Claude Console rather than a Claude chat subscription.
- Access to Claude Code and Codex, or at least one of them, to write the app.
- Free accounts for GitHub, Supabase and Vercel to get started.
- A Pipedream account if you follow the Google Drive part of the course.
- Patience with a beta. Managed Agents is currently in beta, so names and details can change. Always check Anthropic's documentation for the latest.
Common mistakes
- Trying to build your own agent runner first. Managed Agents exists so you do not have to manage containers and long-running processes yourself. Start there.
- Skipping row level security. In a multi-user app, every table needs rules so users only see their own data.
- Not tracking cost from day one. Agents can run for a long time. Record usage per mission so there are no surprises.
- Giving the agent every tool. Only switch on the tools the job needs. It is safer and easier to reason about.
- Assuming it suits every kind of data. Anthropic notes that Managed Agents is not eligible for certain compliance arrangements, such as zero data retention. If you handle sensitive or regulated data, check the docs before you build.
FAQs
What are Claude Managed Agents? They are Anthropic's hosted way to run Claude as an agent. Anthropic provides the agent harness and the secure environment it works in, and your app starts sessions and listens for results.
Do I need to know how to code? No. The course is designed for people with no prior coding experience. Claude Code and Codex write the code while you plan, test and make decisions.
Is Claude Managed Agents the same as Claude Code? No. Claude Code is a coding agent that helps you build software. Managed Agents is a service for running agents inside your own product. In this course you use Claude Code to build a platform that runs Managed Agents.
Can I build a first agent before the full platform? Yes, and it is a great idea. Try Build your first Claude Managed Agent from Claude Code first.
Does a Claude Pro or Max plan cover Managed Agents? Managed Agents uses the Claude API, which needs an API key from the Claude Console. Check Anthropic's documentation and pricing for the current details.