🤖 Agent Builder
Everything you need to design, prompt, connect and evaluate AI agents: fundamentals, agentic prompting, architecture patterns and the Model Context Protocol.
- 1
Prompting 101: The Anatomy of a Production-Grade Prompt
Anthropic's Applied AI team shows how to evolve a one-line prompt into a reliable, production-quality prompt — structure, XML tags, examples, giving the model an out, and prefills.
beginner · ~24 min
- 2
Prompting for Agents: Steering Models That Act
Agents are models using tools in a loop. This lesson covers when to build one, how to prompt it — heuristics, budgets, guardrails — and how to evaluate something that takes hundreds of steps.
intermediate · ~25 min
- 3
Building Effective Agents: Workflows, Agents and the Patterns Between
Anthropic's foundational essay distilled into a class: the five workflow patterns, what truly counts as an agent, why simplicity wins, and how to design tools your agent can actually use.
intermediate · ~20 min
- 4
The Expanding Agent Toolkit: From Scaffolding to Native Capability
How capabilities that once required heavy external scaffolding — tool use, context management, code execution, and computer use — moved into the model itself, and how they compose into agents that finish work instead of just starting it.
intermediate · ~21 min
- 5
Tool, Skill, or Subagent? Decomposing an Agent
The decision framework for knowing when agent logic belongs in a tool, a skill, or a subagent — illustrated through a live decomposition of a 400-line inventory agent.
intermediate · ~45 min
- 6
Trustworthy Agentic Workflows with a Custom DSL
How Elicit built AshPL — a Turing-incomplete, purely functional DSL — to make their AI research assistant legible, auditable, and faithfully executable.
advanced · ~30 min
- 7
MCP 201: How the Model Context Protocol Really Works
Beyond the hello-world server: why MCP exists, its client–server architecture, the three primitives and who controls them, transports, and where the protocol is heading.
advanced · ~30 min
- 8
Ship Your First Managed Agent: Agent, Environment, Session
Claude Managed Agents is the fastest path from prototype to production-ready agent. This lesson walks through the three core primitives — Agent (brain), Environment (hands), Session (the binding) — and shows how to wire them into a working incident-response agent.
intermediate · ~37 min
- 9
Get to Production Faster with Claude Managed Agents
Why building a production-ready agent is harder than it looks, and how Claude Managed Agents handles the infrastructure layer so you can focus on what actually makes your agent valuable.
intermediate · ~29 min
- 10
Build a Production-Ready Agent with Claude Managed Agents
A hands-on walkthrough of Claude Managed Agents: defining agents and environments, creating sessions, streaming events, using outcomes for iterative self-verification, and monitoring live runs in the developer console.
intermediate · ~27 min
- 11
Agents That Remember: Memory Stores and Dreaming in Claude Managed Agents
Sessions are isolated by default — agents forget everything when they close. This lesson shows how to wire persistent memory onto your agents and use Dreaming to consolidate and improve what they remember over time.
intermediate · ~29 min
- 12
Memory and Dreaming: Building Self-Improving Agents
Design production memory systems for multi-agent architectures using filesystem-based memory stores, optimistic concurrency, and the dreaming feedback loop.
advanced · ~25 min
- 13
Teaching Agents to Learn from Your Team
How to encode your team's expertise into AI agents using Skills, CLAUDE.md, and the Dreaming feedback loop — so every agent runs with your team's collective judgment, not just the model's defaults.
intermediate · ~28 min
- 14
Giving Agents Their Own Computers
How Cursor gave cloud agents onboarding, dev environments, and the ability to self-report problems — and what the 'agent experience' means for teams shipping parallel agents at scale.
intermediate · ~15 min
- 15
Agent Harness Engineering: Chasing Friction
AirOps's hard-won lessons from shipping Claude agents to non-technical enterprise users: intentional scoping, specialized tools over primitive exploration, and sub-agents for context isolation.
intermediate · ~27 min
- 16
Evals for Taste: How to Measure and Hill-Climb Your Agent
Without evals you're flying blind — reactive to complaints, unable to verify improvements. This lesson shows how to build code and model graders, run QA loops, and turn subjective quality into something you can act on.
advanced · ~39 min
- 17
Evaluating and Improving Agents at Scale
How Replit built VibeBench and the Telescope continuous improvement system to turn overnight eval runs into shipped model upgrades — without a human in the loop.
advanced · ~28 min
- 18
Self-Improving Prompts: How Metaview Automates Prompt Optimization
How Metaview built an application-review system whose prompts learn from every recruiter decision — a practical blueprint for automated prompt critique and rewriting.
advanced · ~16 min
- 19
Stop Babysitting Your Agents: From Approval Mode to Orchestration
The workflows Claude Code engineers use to stop hand-holding their AI and start orchestrating it — permission architecture, verification-first design, parallel fanout, and headless automation.
intermediate · ~37 min