You ran an agentic workflow, deployed it, and have the mental model for the runtime. Pick a path based on what you want to build next.
Build agents and workflows
Workflows
Orchestration patterns: retries, timeouts, signals, fan-out and fan-in.
Agents
Model-driven loops with tools, memory, and structured output.
Human-in-the-loop
Pause a workflow for approval and resume with the user’s response.
Multi-agent
Coordinate several agents on a single run via tools or handoffs.
Read the concepts
Durable execution
What survives a crash, and how journaling makes it possible.
Determinism
The rules workflow code must follow so replay produces the same result.
The improvement loop
How traces, datasets, and evals work together.
Run in production
Deployments and environments
The preview → staging → production flow.
Observability
Traces, metrics, logs, and what to watch on call.
Cost and tokens
Per-run, per-component, per-model accounting.
Improve what’s running
Setting up evals
Turn a trace into a dataset and score new runs against it.
Comparing workflow versions
Diff two runs and surface regressions before promotion.
Prompt iteration
Version prompts and replay history against the new one.