AI engineering, taught through projects you can ship
Watching tutorials is not shipping software.
Stop collecting prompts. Start shipping products. Practical AI engineering for people who want something working at the end, not another saved thread.
Learn it. Build it. Ship it.
Hosted by Mara Vale, a fictional character. Why?
Choose a branch: five learning tracks
3a7e09cfix(demo):from stage demo to working app
Same model. Different Tuesday.
A demo works when everything goes right. An app works when things go wrong. Pick a failure and run both.
Fragile demo · works on stage
Rejected with message
Rejected with a message the model can act on.
500 error · request lost
Instrumented app · works on Tuesday
Rejected with message
Your tool call was not valid JSON. Send one complete JSON object.
to model: Error: Your tool call was not valid JSON. Send one complete JSON object. Fix the arguments
Both columns run the same deterministic code as the free lessons, in your browser.
4c12d8efeat(tracks):five branches, real prerequisites
Pick the branch you can start today.
Launchpad needs no code. Each track lists what it expects, so you never start three steps ahead of yourself.
Launchpad
Automate one real workflow and understand every step of it.
Needs: nothing but a repeated task
4 to 6 weeks at 2 to 3 hours a weekApp Engineer
Ship an AI feature with types, tests and a fallback, not a demo that only works on stage.
Needs: Comfortable in one language (TypeScript or Python in the examples)
4 weeks at 3 to 4 hours a weekAgent Builder
Build an agent that can act, with limits you wrote down before it ran.
Needs: App Engineer level coding, or finished Launchpad plus some practice
4 weeks at 4 hours a weekAI Systems
Make build or buy calls with evidence, and set up the checks your team will rely on.
Needs: Has shipped or managed shipping software
3 to 4 weeks at 2 hours a weekProduction
See every request, know what it cost, and catch regressions before users do.
Needs: A working AI feature, or the App Engineer project
4 weeks at 3 hours a week
Typical routes: Launchpad → App Engineer → Agent Builder → Production. Leads often go straight to AI Systems.
5e9b311docs(lessons):inside one lesson
Inside one lesson: break the call.
When a model calls a tool it does not run anything. It writes a request: a tool name and some arguments, usually as JSON. Your code reads that request and runs the function. Everything that goes wrong in between is your responsibility, not the model's.
export const refundSchema = { order_id: { type: "string", required: true, pattern: /^ord_[0-9]{4}$/ }, amount_cents: { type: "integer", required: true, min: 1, max: 50000 }, reason: { type: "enum", required: true, values: ["damaged", "late", "wrong_item", "other"] }, notify_customer: { type: "boolean", required: false, default: false },} as const; export const APPROVAL_LIMIT_CENTS = 20000; // 200.00- amount_cents must be an integer number of cents, got "6400". Example: 6400 for 64.00.
6f40ab2feat(shiftplan):seven questions, one route map
Your path, explained.
Change any answer and watch the plan move. Every recommendation says which answer caused it.
Illustrative AI demo · updates as you change answers
Agent Builder
Because you selected "An agent that takes actions with tools" and "I build apps, with some looking things up", we recommend the Agent Builder track.
first build A refund agent with limits
Branch → Build → Test → Ship · confidence High
Get the full route map7b2c5d0chore(persona):meet the host, honestly
Meet Mara. She isn't real.
“The demo is the easy part. The interesting work starts the first time it is wrong and you have to find out why.”
Mara Vale is a created character who gives SHIPSHIFT a voice. She has no career, credentials or students, and we never write first-person stories for her. The AI twin on SHIFTPLAN is a separate, labelled demo assistant that answers only from published lessons.
8d61f7afeat(builds):things you could ship
Builds with source, not screenshots.
Demo projects with real snippets, a diagram and the tests that keep them honest.
Refund agent with limits
Looks up orders, proposes refunds, and stops for a person above the limit.
Cited answer box
Answers help-centre questions from retrieved paragraphs, with a link to each one.
Trace and cost dashboard
One page showing traces, cost per run, latency per step and refusal rate.
9e03c4bfeat(contact):ways to keep going
Keep going your way.
$ git commit -m "start"