04 Engineer: Build AI systems that work reliably at scale

04 Engineer: Build AI systems that work reliably at scale

04 Engineer: Build AI systems that work reliably at scale

[Embedded HTML content]

Welcome video · 1:37

Welcome video transcript

In the third course, you put repetitive work on autopilot\. Say you built an account review for one customer: something happens, the workflow runs across your apps, and the review lands with the team\. It works, and that customer gets a lot out of it\. So you copy it for the next customer, and then the next, and before long there's a copy for every account, each with its own logic, context and cost\. This course is about turning those copies into one system, a business\-critical process that just works every time\. And you're the one who designs it\.

Here's the map again: from chatting with AI to building systems that scale\. Chat is done\. So is Delegate\. So is Automate\. Stage four, Engineer, is where you build AI systems that work reliably at scale\. That's this course\.

In this course, you'll learn to read any AI system through four questions\. What's the goal, and how small is the first version? Which steps are AI, which are deterministic, and where does a person decide? Where do the rules and the context live, and who can change them? And at scale, what does a run cost, and how do you know it's working? You'll put those four questions to real systems, built by real teams, and take what works into your own\.

This is the last stage on the map\. Right now, you are here\. Stage four\. Let's start\.

About this course

This course is about going from personal automated processes to business systems that scale\. You'll review case studies of various real systems and learn how to build your own\.

Who this course is for

You have built and run Zap workflows, and you want to start using best practices for designing for your team or company\. If you have not yet built a workflow, start with 03 Automate\.

What you'll learn

• Name the goal of a system in the business's words, and size the first version that proves it\.
• Name where in a system you need AI, deterministic steps, and human decisions\.
• Locate where a system's rules and knowledge live, and say who can change them without the builder\.
• Judge a system at scale: what a run costs, which model earns its place, and the numbers that tell you it is working\.

Every case study is read through the same four questions: Goal, Steps, Context, Scale\. The systems are real, built by real teams, and presented by the people who built them at ZapConnect 2026\. Each lesson ends with what to steal for your own build\.

Lessons

Welcome to Engineer

• Your map 1:37
• Four questions to ask of any AI system 3:26

Mercari: automating 120k support tickets a month

• How Mercari built support automation that stops instead of guessing 8:00

Zapier Support: feedback on every support conversation

• Review every conversation with one Zap 8:23
• Ask questions of your data in plain language 6:09
• Put the time saved into coaching 2:33

HiveDex: knowing when to use AI and when not to

• Why the all\-AI version failed 6:37
• Where AI decides and where a fixed step runs 6:25

Build Off: four approaches to the same problem brief

• The brief: a viral inbox full of buried orders 2:04
• Four builders show their solutions 7:38

Choosing a model: cost per task, not price per token

• Pick the right model and plan for volume 6:16

Source of truth: a doc that keeps itself current

• Building company sources of truth with a human in the loop 4:52

Finish

• Check your understanding 4 questions
• Course complete certificate

Start the course

Zapier Academy courses are free\. You sign in with your Zapier account \(a free one is fine\) so your progress is saved and your certificate has your name on it\. This is the last stage on the map; from here the role tracks continue with Admin Essentials and AI Transformation Leaders\.