Skip to content

Learn · AI Engineering

The job, not the prompts.

An opinionated track on what AI engineering actually is in 2026. Six modules, each paired with a challenge set on the platform.

Modules

Six modules.

01

The decision-shaped job

What changes when an AI pair joins every commit. A taxonomy of the new decision points and where judgment moves.

02

Tool design for tool-using agents

Naming, contracts, idempotency, error surfaces. How to make tools the model actually composes correctly.

03

Retrieval boundaries

What goes into context, what stays out, and how to draw the line. Failure modes when the boundary is wrong.

04

Evaluation that survives a quarter

Building evals that don't rot. Golden sets, drift detection, and the trap of optimizing for the eval.

05

Production failure modes

Silent regressions, plausible-but-wrong outputs, prompt-injection-as-RCE. What pages on-call at 2 a.m.

06

Replay-driven code review

How to review an AI-authored PR. What questions to ask the author. What to send back.