Topic

AI Agent Hub: A Complete Path from Concept to Production

An AI agent hub covering core concepts, hands-on tutorials, framework and platform selection, MCP tool integration, browser control, memory design, failure triage, and governance.

  • 20 articles
  • 6 wiki entries
  • 4 tools
Editor's guide

Learning agents is not about collecting terminology. It is a path you can actually finish: understand how an agent differs from a workflow, build a minimal version that runs, choose a framework for your scenario, then add memory, permissions, and failure handling. This hub arranges our content in that order.

Who is this hub for?

Developers building a first working agentTeam leads evaluating frameworksProduct and compliance owners setting agent boundaries

On this page

  1. 01What is an AI agent
  2. 02Build your first agent
  3. 03Choosing a framework or platform
  4. 04MCP: connecting tools and data
  5. 05Browser and computer use
  6. 06Memory, failure, and governance
  7. 07Tools you can try now

What is an AI agent

Start with the concepts, the capability boundary, and the common patterns.

Build your first agent

Start minimal: tool calls, task flow, and both no-code and self-built routes.

Choosing a framework or platform

Where LangGraph, AutoGen, CrewAI, Dify, Coze, Flowise, and n8n each fit.

MCP: connecting tools and data

How the protocol works, MCP versus function calling, useful servers, and the risks.

Browser and computer use

How tools like Browser Use drive a page, and how the main options compare.

Memory, failure, and governance

The hard part after launch: what to remember, why runs go wrong, and how far permissions go.

Tools you can try now

Agent-related products in our directory, plus related sections.

FAQ

Q: How is an AI agent different from an automated workflow?
A: A workflow follows steps a human wrote down; an agent lets the model decide which tool to call next. When the path is fixed, a workflow is cheaper and more predictable. An agent is only worth its unpredictability when the path has to be decided at run time. See our piece on agents losing control and the AI workflow wiki entry.
Q: Should a first agent start from a framework or a no-code platform?
A: Get the flow running on a platform such as Dify first and confirm the task is worth automating. Only then decide whether to rebuild it on LangGraph or AutoGen. Most teams stall on unclear task boundaries, not on framework limits.
Q: Is a multi-agent setup always better than a single agent?
A: No. Multiple agents multiply context-passing cost and debugging effort. They pay off only when roles are genuinely separate and a single agent's prompt can no longer hold the job. Our multi-agent article works through the trade-off.
Q: What risk is most often overlooked when granting an agent tool access?
A: Write permissions and memory poisoning. A faulty read tool produces a wrong answer; a faulty write tool damages real data, and poisoned long-term memory keeps affecting later tasks. See our MCP security and agent memory articles.