AI engineering, step by step.
Four core paths that take you from calling a model to running an AI product, and six career routes for the jobs they lead to. Each lesson points to a hands-on lab in a real terminal, checked on the state of the machine, not on clicks.
4 core paths · 6 career routes
Four core paths
Domain 01
interfaces, grounding, evaluation, production
Building and Deploying AI Applications
Follow an AI feature from its first request to a model, through grounding and testing, to running it for real users. The application is everything around the model, not the model alone.
12-lesson path · 13 h
-
Model Interaction Contracts
Send a request, set the system instructions, and agree on what comes back before anything depends on it.
-
Structured Generation Contracts
Make the model return data a program can read, and let it call your functions.
-
Evidence Representation
Turn documents into vectors so the right passage can be found for a question.
-
Retrieval and Freshness
Keep what the model is shown current, and know when an answer rests on stale evidence.
No lab for this lesson yet.
-
Behavioral Evaluation Gates
Write tests for behaviour that is not deterministic, and stop a release that fails them.
-
Serving and Recovery
Limits, retries, health checks and restarts: what keeps the service answering when something breaks.
- 6 more lessons in this path are on the way.
Domain 02
full stack, data, architecture, reliability, scale
Software Engineering Fundamentals
A coding assistant saves typing, not judgment. Learn the decisions that stay yours: how an application is delivered, where its data lives, how it is split up, how it is secured, and who owns it when it runs.
13-lesson path · 12 h 10 min
-
End-to-End Application Delivery
Take a change from a commit to a running container through an automated pipeline.
-
Data Lifecycle and Storage
Choose a store, shape the records, and understand what happens to the data over time.
-
System Architecture and Boundaries
Put a proxy in front, separate the services, and decide what may talk to what.
-
Secure and Resilient Systems
Give each part only the access it needs, and harden what you ship.
-
Production Scale and Service Ownership
Set reliability targets, measure them, and scale on purpose.
- 8 more lessons in this path are on the way.
Domain 03
frame, plan, delegate, verify, improve
Agent-Assisted Engineering
A coding agent is only as dependable as the setup around it: the task, the context, the tools, the feedback and the point where it stops. Build that setup and use it on real repository work.
16-lesson path · 15 h
-
Task Framing
State the goal, the limits and the finish line so an agent can start from something concrete.
-
Evidence-Based Planning
Have the agent read before it acts, and plan from what it found.
-
Agent Workbench
Give an agent the tools it needs through a small, well-defined server.
-
Instructions and Memory
Keep the rules and the state a session needs, and drop what it does not.
-
Runtime Feedback
Read logs and output so the agent corrects itself from what really happened.
-
Verification and Review
Check the result independently, and read the history to see what changed and why.
-
Isolated Delegation
Hand parts of a job to separate agents, each with limited authority.
-
Durable Improvement
Turn a lesson from one task into a rule that helps the next one.
No lab for this lesson yet.
- 8 more lessons in this path are on the way.
Domain 04
outcomes, evidence, risk, metrics, feedback
Product Judgment and Delivery
Before building, decide which outcome matters, what supports it and what could go wrong, and how you will know the change helped. Then ship it in small, testable steps and own what comes back.
8-lesson path · 9 h 10 min
-
Outcomes Before Output
Start from the result someone needs, not the feature list.
No lab for this lesson yet.
-
Workflow Discovery
Watch how the work is really done before you decide what to automate.
No lab for this lesson yet.
-
Assumptions and Risk
List what you are betting on, and test the riskiest bet first.
No lab for this lesson yet.
-
Testable Slices
Cut the work into pieces that each prove something on their own.
-
Executable Specifications
Write the expected behaviour as a check that can run.
No lab for this lesson yet.
-
Success Metrics
Pick numbers that show whether the change worked, and record them from day one.
-
Release Strategy
Release in stages, with a way back.
-
Feedback Ownership
Decide who reads the signals after release and who acts on them.
Practice that runs through every path
Whatever the path, the same four habits turn a lesson into work you can rely on.
- Direct the agentGive an assistant a clear task, the context it needs and a place to stop.
- Verify independentlyCheck the result with your own tests and your own reading, not the assistant's word.
- Make the product decisionDecide what is worth shipping, what is too risky and what can wait.
- Own the next iterationRead what comes back after release and turn it into the next change.
Six career routes
A route lists the paths to take, in order, for one kind of role. Finishing the labs shows you can do the work; it does not guarantee a job.
-
Route 01
Customer AI DeploymentForward-Deployed AI Engineer · Field AI Engineer · AI Solutions Engineer
Turn one customer workflow into a small AI system you can measure, then stay close to the rollout to see where it breaks.
Start with Software delivery and AI application basics.
Guided route: 12 specialist lessons · 14 h 25 min
Study the route -
Route 02
Developer Experience and EducationAI Developer Relations Engineer · AI Developer Advocate · Developer Experience Engineer
Make an AI capability easy to understand and run, then carry developers' friction back to the product team.
Start with Software basics, working with APIs and clear technical writing.
Guided route: 11 specialist lessons · 15 h 5 min
Study the route -
Route 03
AI Data SystemsAI Data Engineer · Machine Learning Data Engineer · Retrieval Engineer
Build the data and retrieval pipelines that give training, evaluation and live AI features evidence you can trust.
Start with Python, data structures and pipeline basics.
Guided route: 11 specialist lessons · 15 h 15 min
Study the route -
Route 04
Agent Systems EngineeringAgent Systems Engineer · Agentic AI Engineer · AI Agent Engineer
Engineer the runtime around a tool-using model so its context, memory, permissions and failures stay visible.
Start with Language-model basics and typed tool interfaces.
Guided route: 14 specialist lessons · 14 h 25 min
Study the route -
Route 05
LLM Product EngineeringApplied AI Engineer · LLM Engineer · AI Product Engineer
Turn what a model can do into product behaviour that is grounded, tested, guarded, affordable and recoverable.
Start with Software engineering and language-model basics.
Guided route: 12 specialist lessons · 14 h 45 min
Study the route -
Route 06
AI Evaluation and ReliabilityAI Evaluation Engineer · AI Reliability Engineer · Machine Learning Site Reliability Engineer
Make model and agent behaviour measurable, find failures before release, and build the controls for what still fails in production.
Start with Statistics, software testing and production systems.
Guided route: 12 specialist lessons · 12 h 30 min
Study the route
