The leader track 🤝 · for C-level, managers & curious minds · 6 x 45 min · no code, ever
Thinking mode: what agents are, how they fail, how to govern them, and how to make the build-buy-fund decisions - each session ends with questions to ask your data team.
Chatbot, robot, or colleague? The autonomy ladder
Workflows vs agents in plain English, the RPA hybrid pattern, and the agent-washing screen for vendor pitches.
🗣️ Leader 2 · easySpeaking agent: the vocabulary bridge
The 22 terms that end nodding-along meetings - each framed as the question it lets you ask your team.
🔥 Leader 3 · mediumWhen agents go wrong: the failure brief
Replit's deleted database, Klarna's reversal, runaway bills - and why failures are governance gaps, not model gaps.
🛡️ Leader 4 · mediumThe lethal trifecta and the governance stack
The one security rule every leader can retain, plus approval gates, spend caps, and audit trails that actually work.
💸 Leader 5 · decidingBuild, buy, or blend: the investment decision
Real cost ranges, the 1M-conversation crossover, and how to read a vendor whose framework is free.
📊 Leader 6 · decidingProving it: ROI and the honest scorecard
Klarna as best case AND cautionary tale, the believer and skeptic numbers, and the 5-metric scorecard to demand.
The builder track 🛠️ · for practitioners · 10 x 45 min · Claude API or local Ollama
One growing artifact: DataDesk, a data-team assistant. Every session upgrades the same code - and every capability is built by hand first, so the framework never gets to be magic.
Do you even need a framework?
Build an agent with no framework, rebuild it in three lines of LangChain 1.x, and learn the honest alternatives card.
🧱 Builder 2 · easyModels, messages, tools
init_chat_model, the Claude ↔ Ollama one-line swap, content blocks, and structured output DataDesk can validate.
🤖 Builder 3 · mediumcreate_agent, properly
Inside the 1.x harness: multi-tool agents, system prompts as constitutions, and streaming that feels alive.
🧅 Builder 4 · mediumMiddleware: the production layer
PII scrubbing, summarization, retries, Claude-to-local fallback - and the yanked-release lesson in pinning versions.
🕸️ Builder 5 · hands-onThe LangGraph mental model
State, nodes, edges, routers - the graph under everything, and when to eject from create_agent down to it.
💾 Builder 6 · hands-onState, memory, persistence
Checkpointers, thread identity, and the resurrection moment: kill the process, restart, and the conversation continues.
🚦 Builder 7 · hands-onHuman-in-the-loop and time travel
Approval gates before write-actions (approve, edit, reject) and rewinding a thread to fork a what-if.
📚 Builder 8 · mediumRAG the 1.x way
Retrieval as a tool the agent chooses to call: DataDesk reads the team wiki, cites its sources, and admits gaps.
👥 Builder 9 · hardestMulti-agent and subgraphs
Supervisor, specialists, Send fan-out - plus the cascade experiment that shows exactly why checkers exist.
🎓 Builder 10 · hardestLangSmith: trace, eval, ship
Tracing in two env vars, an eval suite with an LLM judge, shipping options - and DataDesk's graduation checklist.
Choose your path 🗺️
Two doors, one course. Leaders never see code; builders never sit through governance theory they have not felt in a demo first.
The knowledge map 🧠
Both tracks at a glance - hover a session to spotlight its concepts, click any node to jump in.