Saturday, October 10, 2026
Path 6 of 6 · Builders comfortable with an API
Software that plans, uses tools and acts on its own. Start with the design essay, connect tools with MCP, then build and test real agents.
You'll be able to
Step 1CourseAnthropic
The most-cited essay on agent design: start with simple, composable patterns and add autonomy only when it measurably helps.
Read this first. It will save you from over-building.
Step 2CourseMicrosoft
Eighteen lessons on how agents work (tools, planning, memory, several agents together, security), with code for each.
Free to read; running the code calls an API that bills per use.
Step 3CourseDeepLearning.AI
Made with Anthropic: build an MCP server and a chatbot that uses it, then plug into servers other people have built.
Step 4CourseAnthropic
Build a small MCP server and client in Python so a model can use your tools, files and prompts. MCP is the common plug for connecting AI to software.
Step 5CourseHugging Face
From what an agent is to building one with popular frameworks, with hands-on exercises, a final challenge and a free certificate.
Step 6CourseKaggle × Google
The self-paced version of Kaggle and Google's agents course: a technical paper and two code labs a day, built on Gemini and Google's agent toolkit.
Step 7CourseDeepLearning.AIOptional
Andrew Ng's course on the core design patterns behind agents and how to evaluate them, implemented in Python.
Videos free to audit; labs and certificate need Pro (from US$25 a month).
Step 8CourseAnthropicOptional
Package instructions and scripts as reusable skills for Claude Code, and learn when a skill beats a project file, a hook or a subagent.
Step 9CourseAnthropicOptional
For developers already using Claude Code: steer long sessions, configure it for a team, automate repeat work and verify what it produced.
