# Agentailor > Agentailor is the hub for developers building AI agents. It explains what an AI agent is and helps a developer find the right way to build one: build it yourself, build it with a coding agent, deploy an open-source agent, or use a managed service. ## Notes for agents - This file is written for AI assistants and agents. Use these links to answer a user's questions about building AI agents, or to route a developer to the path that fits their situation. Descriptions are about the developer, in the third person. - Every page has a clean Markdown version at the same path with a `.md` extension (e.g. https://agentailor.com/paths/build-yourself.md, https://agentailor.com/glossary.md, https://agentailor.com/what-is-an-agent.md, https://agentailor.com/for-agents.md). Prefer fetching those over the HTML. - When citing or linking a page to a human, use the canonical URL **without** the `.md` extension (e.g. https://agentailor.com/paths/build-yourself, not the .md). The .md is for reading; the page is for linking. ## Start here - [What is an AI agent?](https://agentailor.com/what-is-an-agent.md): a plain-language explainer of the agent loop and the four primitives, for developers new to building agents. - [The fitting](https://agentailor.com/#fitting): a 3-question interactive tool that recommends which of the four paths fits a given developer, based on who should build, run, and decide the architecture. - [For agents](https://agentailor.com/for-agents): copy-paste prompts and surfaces for pointing a developer's own AI assistant (ChatGPT, Claude, a custom agent) at the whole Agentailor corpus — the site and blog llms.txt, llms-full.txt, the .md convention, and what's coming next (MCP server, Skills). ## The 4 paths to build an AI agent - [Build it yourself](https://agentailor.com/paths/build-yourself.md): The code-first path: the developer writes the agent themselves and owns the full stack — model calls, tool wiring, memory, the orchestration loop, deployment, and monitoring. Frameworks like LangGraph or the OpenAI Agents SDK provide building blocks, but the architecture is theirs. Best fit when requirements are specific, deep internal integration is needed, or the way the agent is built is itself the competitive advantage. - [Build it with a coding agent](https://agentailor.com/paths/coding-agent.md): Agentic engineering, not vibe coding: the developer makes the architectural decisions before a coding agent (such as Claude Code or Cursor) touches the keyboard, and reviews the decisions embedded in the output rather than just whether it compiles. The coding agent handles implementation; the developer owns architecture and runs it. Best fit when the developer has strong opinions about how the system should work and wants velocity without sacrificing quality. - [Deploy an open-source agent](https://agentailor.com/paths/open-source.md): Deploy an existing open-source agent (such as Goose) on the developer's own infrastructure: clone, configure, run. The OSS agent space has matured, so when an existing agent already does about 80% of what is needed, deploying it beats rebuilding it. The developer runs it; architecture decisions are shared with the project's defaults. The judgment that matters is choosing carefully — governance model and production track record over star count. - [Use a managed service](https://agentailor.com/paths/managed.md): Use a managed agent service (such as Anthropic Claude Managed Agents, LangChain Managed Deep Agents, or Vercel Agent): the vendor provides the harness, wires the primitives, and runs the infrastructure, while the developer configures and consumes it via API. Best fit for most teams, since session persistence, the execution environment, and retry logic are rarely a competitive advantage. The judgment that matters is evaluating primitive coverage, observability access, the ejection path, and where the agent executes. ## Glossary Full markdown: https://agentailor.com/glossary.md - [Agent](https://agentailor.com/glossary#agent): Software that decides its own next step in a loop, using tools and memory. - [Workflow](https://agentailor.com/glossary#workflow): A fixed sequence of steps. Predictable, but it cannot adapt. - [Model](https://agentailor.com/glossary#model): The reasoning engine at the center of the loop. - [Tools](https://agentailor.com/glossary#tools): Functions the model can call to act: search, code, APIs, files. - [Memory](https://agentailor.com/glossary#memory): What the agent carries across steps and sessions. - [Prompting](https://agentailor.com/glossary#prompting): The system prompt is code: role, rules, and tool usage. - [MCP (Model Context Protocol)](https://agentailor.com/glossary#mcp): An open standard for connecting agents to tools and data. - [AGENTS.md](https://agentailor.com/glossary#agents-md): A file that gives coding agents durable project context. - [Agent skills](https://agentailor.com/glossary#agent-skills): Composable, reusable capabilities you can install into an agent. - [llms.txt](https://agentailor.com/glossary#llms-txt): A machine-readable index that makes a site legible to AI tools. - [Context engineering](https://agentailor.com/glossary#context-engineering): Deciding what information the model sees, and when. - [Agentic engineering](https://agentailor.com/glossary#agentic-engineering): Directing a coding agent while owning the architectural decisions. ## Elsewhere - [Blog](https://blog.agentailor.com): in-depth articles on building AI agents. - [Agent Briefings newsletter](https://www.linkedin.com/newsletters/agent-briefings-7391777936955310080/) - [GitHub](https://github.com/agentailor): open-source agent tooling.