The leader track 🤝 · for C-level, managers & curious minds · 6 x 45 min · no code to write
Thinking mode: what agents actually are behind the hype, where they pay off, the risks that reach customers, and the questions that separate a real agent vendor from a demo - each session ends with questions to ask your teams.
What an agent actually is
The loop behind the hype, the official Anthropic and OpenAI definitions, what does not count - and when NOT to want one.
🧩 Leader 2 · easyThe anatomy in plain English
Tools, memory, planning, guardrails as business capabilities you can commission - and the failure each one prevents.
🎚️ Leader 3 · mediumWorkflows vs agents
The autonomy dial: five workflow patterns as cost-reliability decisions, and why both vendors say start at the boring end.
🛡️ Leader 4 · mediumRisk and governance
Prompt injection, the seven guardrail types, tool risk-rating, human-in-the-loop triggers - the risk register that actually matters.
🏢 Leader 5 · decidingAgents in your organization
The three when-to-build criteria, scanning your org for agent-shaped wins, build vs buy, team readiness, and ROI framing.
🔭 Leader 6 · decidingThe roadmap and your strategy
Multi-agent systems, MCP vs A2A, computer use, managed agents - what changes in the next 12 months, what to fund, what to wait on.
The builder track 🛠️ · for practitioners · 10 x 45 min · Python · live agent on every page
One running project, built harder every session: the Daybreak Ops Agent - a coffee brand's real operations agent on its MCP-server toolbox. The 12-task scorecard climbs 1 → 5 → 7 → 10 → 12 as the levers come on; b10 must hit 12/12 to ship. Claude API or local Ollama, your choice, throughout.
The agent loop
What qualifies as an agent, Thought-Action-Observation, the four levers, and the 12-task scorecard you will spend ten sessions climbing.
🔧 Builder 2 · easyTools + your first agent
Tool schemas, the agent-computer interface, Daybreak's MCP server as toolbox - and a working single-loop agent in raw Python.
🧱 Builder 3 · mediumThe five workflow patterns
Prompt chaining, routing, parallelization, orchestrator-workers, evaluator-optimizer - Anthropic's full pattern catalog, built on Daybreak.
🧠 Builder 4 · mediumMemory and context
Context management, compaction, scratchpads, sessions, long-term stores - keeping findings and instructions from falling out mid-task.
🗺️ Builder 5 · mediumPlanning and reasoning
ReAct, plan-then-execute, reflection, task decomposition - turning one-shot half-answers into real diagnoses and reports.
🤝 Builder 6 · hands-onMulti-agent systems
Manager vs handoffs, subagents, orchestrator-workers at agent scale - and the single-agent-first rule that saves most teams.
🛡️ Builder 7 · hands-onGuardrails
The seven guardrail types hands-on: injection defense, approval gates, budgets and stop conditions - the two dangerous tasks finally pass safely.
🎯 Builder 8 · hands-onAgent evals
Tasks, trials, graders, pass@k vs pass^k, trajectory checks, LLM-judges - the golden set formalized into a real eval harness.
🚢 Builder 9 · hardestShip and observe
Claude Agent SDK vs LangGraph vs crewAI/AutoGen, tracing, deployment, cost control - and where A2A fits the protocol map.
🏆 Builder 10 · hardestCapstone: the full ops agent
Assemble everything - toolbox, memory, planning, guardrails - run the eval harness, and do not ship until the scorecard reads 12/12.
Choose your path 🗺️
Two doors, one loop. Leaders never touch code; builders never sit through strategy they have not just wired themselves.
The knowledge map 🧠
Both tracks at a glance - hover a session to spotlight its concepts, click any node to jump in.