Module 1 · Compute and execution · Lesson 3

Durable execution engines

How Temporal, Inngest, Restate, Trigger.dev and Vercel Workflow keep multi-step agent work alive across crashes, deploys and week-long waits, through one shared model of steps, journals, replay and retries.

80 min 1 video 3 exercises 7-question quiz Answers

Setup for the exercises:Install Claude Code Python with uv Run Temporal locally

By the end of this lesson you can

Explain the durable execution model of workflows, steps, an event journal and deterministic replay
Predict what re-runs after a crash, and identify code that breaks replay
Compute retry schedules from a backoff policy and choose retry settings for model and API calls
Compare Temporal, Inngest, Restate, Trigger.dev and Vercel Workflow on execution model, language support and operations
Write a long-running payment workflow in Python that waits for an external event with a deadline

A completion statement is not one model call. The agent loads the matter, waits for the lender's redemption figure, asks the seller's agent for the commission invoice, computes the apportionments in a sandbox, drafts the statement, waits for a solicitor to approve it, and sends it to the client. That can take five days. Over those days you'll deploy several times, a function will time out, and the lender's API will return 503 for an afternoon. A plain script dies at the first of these and starts over from the top, re-spending tokens and re-sending emails. Durable execution is the family of engines that make that script survive.

Production Agent Systems covers durable sessions in eve and the idempotency patterns your tools need. This lesson opens the engine: the model they share, the differences that decide which one you pick, and the rules your code must follow.

Durable execution engines

Durable execution engines

Durable execution engines

Sign in to keep reading.

Sign in to keep reading.