← susanwang.ai
31-Day Course

Daily Agent Learning

A structured 31-day deep dive into AI agent architecture, tools, memory, and real-world deployment — one lesson per day.

📅 31 Days
⏱ ~45 min / day
🎯 Beginner → Advanced
👍 Free
All Lessons Available 31 / 31
All Lessons
Day 01
The Full Agent Stack
ReAct loop, LLM+Memory+Planning+Tools — how modern agents are architected from the ground up.
✓ Available
Day 02
Memory Architecture
4 memory tiers, RAG, MemGPT — how agents remember across context windows.
✓ Available
Day 03
Planning & Tool Use
ReWOO, Tree-of-Thought, MCP, tool schemas — how agents plan multi-step tasks.
✓ Available
Day 04
RAG Deep Dive
Chunking, hybrid search, vector stores, eval metrics — grounding agents in real knowledge.
✓ Available
Day 05
Agent Frameworks
LangGraph, AutoGen, CrewAI, LlamaIndex — four paradigms for orchestrating agents.
✓ Available
Day 06
Benchmarks & Eval
SWE-bench, OSWorld, GAIA, AgentBench — how to measure agent performance and read vendor claims.
✓ Available
Day 07
Multi-Agent Systems
Orchestration, trust hierarchies, A2A basics — when one agent isn't enough.
✓ Available
Day 08
Computer Use Agents
GUI automation, accessibility trees, vision grounding — agents that see and click.
✓ Available
Day 09
Code Agents
SWE-agent, Devin, Codex CLI, Claude Code — agents that write and debug code at repository scale.
✓ Available
Day 10
Long-Horizon Tasks
Decomposition, checkpointing, failure recovery — agents that plan and persist across hours-long tasks.
✓ Available
Day 11
Agent Safety
Prompt injection, sandboxing, Constitutional AI, red-teaming — keeping agents safe at scale.
✓ Available
Day 12
Agent Economics
Cost-per-task, token optimization, ROI frameworks — the business math of running agents.
✓ Available
Day 13
Research Frontiers
Self-improvement, meta-learning, open problems — what the field is working on right now.
✓ Available
Day 14
OpenClaw Deep-Dive
SKILL.md, hub-and-spoke, 9 CVEs, 1184 malicious skills — anatomy of an agent platform attack.
✓ Available
Day 15
A2A Protocols
Agent Cards, OAuth 2.0, inter-agent trust, 150+ partners — how agents talk to agents.
✓ Available
Day 16
Agentic Commerce
Stripe ACP, Visa TAP, agent wallets — when agents become buyers and $0.20 breaks Visa economics.
✓ Available
Day 17
Synthesis
Five strategic axes, six-layer stack, 306 production teams — pulling the first 17 days into one playbook.
✓ Available
Day 18
Reasoning Models & Test-Time Compute
o-series, R1-class, budget-forcing — a third scaling axis that buys capability from thinking time, not parameters.
✓ Available
Day 19
Agent Observability
Tracing, online evals, AgentOps — the observability layer that turns "sometimes brilliant" into a shippable product.
✓ Available
Day 20
Voice & Realtime Agents
Speech-to-speech, full-duplex, sub-second turn-taking — the architecture powering voice agents eating call centres.
✓ Available
Day 21
Agent RL & Fine-tuning
GRPO, RLVR, reward hacking — how RL is moving agents from the prompting layer to the training layer.
✓ Available
Day 22
Context Engineering
KV cache reuse, lost-in-the-middle, compaction — engineering the context window like a system, not a prompt.
✓ Available
Day 23
Open-Weight Agent Models
Qwen3, DeepSeek, Mistral — why the agentic frontier is being shaped by models you can git clone.
✓ Available
Day 24
Deep Research Agents
Orchestrator + subagent loops, citation grounding — how LLMs stop being chatbots and become analysts.
✓ Available
Day 25
Async Agents & Background Workflows
Persistence layer, agent inbox, background tasks — delegating to agents like colleagues, not chatbots.
✓ Available
Day 26
Agentic Browsers
Atlas, Comet, Dia, Claude in Chrome — when the browser becomes the agent and the web is the OS.
✓ Available
Day 27
Embodied Agents & VLA Models
RT-2, π0, GR00T N1 — dual-system VLA architecture and what happens when agents get a body.
✓ Available
Day 28
Generative UI & Agent-Native Interfaces
MCP Apps, AG-UI, A2UI — when the interface generates itself for this user, this task, this moment.
✓ Available
Day 29
Agent Sandboxing & Secure Execution
microVMs, capability-scoped tokens, runtime policy — the containment layer between your model and the OS.
✓ Available
Day 30
Agent Personalization & Persistent Memory
Temporal user graphs, memory moats, cross-session recall — how agents remember who you are across weeks.
✓ Available
Day 31
Agent Skills & the SKILL.md Standard
SKILL.md manifests, skill marketplaces, progressive disclosure — capabilities as packages, not fine-tuning.
✓ Available
Daily Agent Learning · 31-Day Course

Part of susanwang.ai · Built at botlearn.ai