What to build, document, and publish to demonstrate real AI-native skills.
Module · Career Building
Lesson 11 of 12 available lessons
⬡ What you'll build
Anyone can now generate a slick-looking output. That means a portfolio of slick outputs proves nothing — a reviewer can't tell your polished result from one the AI produced while you watched. What they can't fake, and what actually convinces, is a documented account of how you worked: the decisions, the verification, and the things that went wrong. This lesson is the playbook for building that.
The previous lesson made the case that judgement and verification are what appreciate when AI is cheap. A portfolio is where you make those visible. The shift is simple but most people miss it:
A portfolio of outputs says "look what got made." A portfolio of process says "look how I think." Only the second one survives the obvious question: did the AI do this, or did you?
This mirrors the Lab's own ethos — show the work, show the system, show the failures, show what was actually learned. A project page that admits "here's the bug the AI introduced and how I caught it" is more credible than one that pretends the path was clean, because real work is never clean and reviewers know it.
You are not building a professional agency showcase. You're assembling 3–5 small artifacts that each demonstrate a real capability. Strong candidates, all achievable as a student:
| Artifact | What it proves | Realistic scope |
|---|---|---|
| Research synthesis | Verification, source discipline, judgement | One question, ~6 real sources, your synthesis |
| Automation experiment | You can make AI do something, not just answer | A script/workflow that saves you a recurring task |
| Prompt system | Repeatable, engineered AI use | A documented set of prompts for one job, with why they work |
| Study system | Applied learning design (the whole track, in fact) | Your personal AI learning workflow, written up |
| Mini-tool | End-to-end building capability | A tiny web tool that solves one real problem |
| Case-study write-up | Communication + reflection | A before/after account of a workflow you improved |
You don't need all six. Pick the ones that fit your field and that you can honestly document. A history student's research synthesis is as valid as a CS student's mini-tool — the documentation quality is what's being assessed, not the domain.
Every artifact gets a short write-up with the same five sections. This structure is what turns "a thing I made" into evidence of capability. Keep each project page to roughly 300–600 words.
Problem — what and for whom
Workflow — how you actually worked
AI's role — honestly scoped
Verification — what you checked and how
Failures & iteration — what went wrong and improved
You can generate a first pass of this write-up with AI — but only after you've done the project, and using your own notes. Prompt it like this:
Here are my raw notes from a project I built: [paste notes — the problem, the steps I took, what the AI did, what I checked, what went wrong]. Draft a project write-up with five sections: Problem, Workflow, AI's role, Verification, Failures & iteration. Use only what's in my notes — do not invent steps, results, or tools I didn't mention. Keep it under 500 words and plain.
The "do not invent" constraint matters here as much as in research: a project page with a fabricated detail collapses the moment someone asks about it.
Weak entry (output only):
"AI Study Assistant — a prompt system for summarising lectures. Built with Claude." Plus a screenshot.
A reviewer learns nothing about you from this. It could be a tutorial clone.
Strong entry (process):
Problem: I was re-reading full lecture transcripts before exams and running out of time. Workflow: Built a three-prompt system — extract key claims, generate self-test questions, flag anything I marked as "unsure." AI's role: Claude generates the questions; I answer them from memory before checking. Verification: I spot-checked 15 generated questions against the source slides — 2 misrepresented the material, so I added a "quote the slide" step to the prompt. Failures & iteration: First version summarised for me, which defeated the point (I retained nothing). I rebuilt it to quiz me instead. That single change is what made it work.
Same project. The second version proves judgement, verification, and iteration — none of which AI can supply on your behalf.
Final Project: AI-Integrated Study System
Your capstone — and a portfolio-worthy artifact in its own right. Build the system, then document it using this lesson's template.
Career Positioning in an AI World
Why process-based proof beats polished output — the strategy this portfolio executes.