Week 3
Agents, retrieval and the browser
By Anusheel Bhushan, co-founder of PlankAgents flow from Week 2's context/memory. Then give agents knowledge (RAG) and capabilities (scraping, browser).
What the week builds toward
WorkBot's investigation agent is an orchestrator that delegates to 2 specialists. One explores the knowledge base with agentic retrieval. The other works websites in a browser. The agent recovers from errors, logs every step as structured entries, tracks its cost and has end-to-end tests.
The 5 days
- Day 11
AI agent fundamentals
Learn the agent loop and its patterns. Then build a ReAct investigation agent from scratch, with step, cost and time limits.
You ship: A ReAct investigation agent built from scratch, which shows its reasoning as it works.
- Day 12
Multi-agent systems and frameworks
Design multi-agent architectures. Then build a research, writing and editing pipeline for customer replies on an agent framework.
You ship: A 4-agent pipeline for customer replies, with guardrails, handoffs and tracing, and an intake workflow.
- Day 13
Modern RAG: agentic retrieval
See where one-shot retrieval fails. Then build agentic retrieval over a 52-document corpus, with a virtual filesystem the agent can explore.
You ship: Agentic retrieval over a 52-document corpus, compared with top-5 retrieval on 10 questions.
- Day 14
AI scraping, browser use and desktop use
Extract structured data from the web with AI. Drive a browser from screenshots with a vision model, and learn how computer use works and where its limits are.
You ship: An AI scraper with pagination, and a browser agent that completes a task in a legacy portal.
- Day 15
The research analyst: orchestration, retrieval and the browser
Combine multi-agent orchestration, agentic retrieval and browser automation into one tested, logged research analyst.
You ship: The AI research analyst, with error recovery, logging, cost tracking and tests, filing its report on the ticket.
