03Learning path
Design tool-using, multi-step systems with explicit state, guardrails, and human oversight.
plan
next action from goal + state
select tool
allow-listed registry only
validate args
schema check before running
execute
timeout + idempotency key
observe
append result to state
Exits
✓ respond
goal met, cite tool results
⏸ ask a human
approval required for write actions
✕ fail safely
step limit (8) or invalid tool call
01
Agents let a model decide which steps to take and which tools to call. That flexibility is powerful, but it makes systems harder to predict, test, and secure. This path teaches you to build multi-step systems that stay understandable and controllable.
You begin with tool calling and deterministic workflows, and add agent loops, memory, and planning only where the task needs them. Guardrails, least-privilege tool access, and human-in-the-loop approval are part of the design from the start rather than added at the end.
The path closes with evaluation: scenario-based tests that check both what an agent produced and the sequence of tool calls it made to get there.
Who it’s forEngineers automating multi-step tasks — triage, research, operations — who need systems that are testable and safe to run.
02
03
BuildGive a model three tools and validate every call before executing it.
MilestoneInvalid tool arguments are caught and reported back to the model instead of crashing the run.
BuildBuild a ticket triage workflow that routes requests to specialized handlers.
MilestoneYou can explain why each step in your system is a fixed workflow or an agent decision.
BuildAdd persistent state so an interrupted task can resume from its last checkpoint.
MilestoneA multi-step run survives a restart without repeating completed actions.
BuildRequire human approval before any tool call with side effects.
MilestoneNo side-effecting action runs without an approval record.
BuildWrite a scenario test suite that checks final outputs and tool-call sequences.
MilestoneYour agent's scenario suite runs in CI and flags trajectory regressions.
04
05
output · TicketTriage
{
"category": "billing",
"priority": "high",
"order_id": "48213",
"needs_human": true
}
Build a service that classifies incoming support tickets and extracts key fields into a validated JSON schema, ready for routing and reporting.
plan
select tool
validate args
execute
observe
Build an agent that completes multi-step tasks by calling a small set of well-defined tools, with explicit workflow state, step limits and predictable failure handling.
06
Outcomes describe skills and portfolio artifacts. No employment outcome is promised.
Build the foundations, projects, and habits needed to design more dependable AI applications.