Responsible YOLO — and setting up OpenRouter
Before we let a model run free, two honest counterpoints — and then the one piece of setup every project this week needs: an OpenRouter key.
01Two honest counterpoints before the fun
YOLO mode is where a lot of the excitement is right now, and we’re about to dive in. But two sobering findings are worth holding in mind — not to spoil the fun, to keep it responsible.
AI help can dent learning
Junior coders were split into an AI-assisted group and an unaided group, then quizzed on the library they’d used.
The productivity gain was small here — but understanding took a real hit.
A community draws the line
The open-source media platform published a policy for AI-assisted contributions. The heart of it: you must be able to explain your own change without the LLM. If you can’t, they’re not interested. LLM output is barred from issues and PR descriptions (translations excepted).
The golden rule: do not just let an LLM loose on the codebase with a vague vibe prompt and commit the results as-is. That’s lazy development, and it produces slop. — paraphrased from the Jellyfin AI policy
Both point the same way. It is now trivially easy to generate mountains of code and drop it on someone else to understand. That asymmetry isn’t fair — and it shifts back only if the people generating the code own it. For the rest of this course, that’s us.
Use coding agents freely and enthusiastically — but not as your sole source of code you don’t understand. Generate it, then read it, test it, and be able to explain it. Own it.
02What we’re going to build
With that spirit set, here’s the project: a personal portfolio website with an AI “digital twin” — a chatbot that answers questions about you and your career. It’s an AI that builds an app which itself contains AI, so first we need a way to make model calls from the site.
Instead of opening separate accounts, top-ups and API keys with OpenAI, Anthropic and Google, OpenRouter lets you keep one account and route to any model — free or paid. Already have an OpenAI key you like? You can use that instead; just swap the names accordingly.
03Setting up OpenRouter
Create an account
Go to openrouter.ai and sign up (Google sign-in is fine). Answer the onboarding questions and you’ll land on your dashboard.
Create an API key
Open the avatar menu → Keys → Create API key. Name it anything memorable. Optionally set a monthly spend limit and an expiry. Copy the key the moment it appears — you won’t see it again.
(Optional) Turn on free models
To use free endpoints, go to Settings → Privacy & guardrails and enable the two options that allow free endpoints to train on and publish your prompts. Free comes at the cost of privacy — enable only if you accept that.
(Optional) Add a little credit
For a stronger paid model, add credit under Credits. OpenRouter’s minimum is friendly — around $2 — and you draw down against it as you go. Entirely optional.
Your key must be copied verbatim — including the sk-or-v1- prefix
(secret key, OpenRouter, v1). One wrong character and it silently won’t work. If in doubt, delete it
and create a fresh one. Every class, someone loses an hour to a mistyped key — don’t be that person.
# keep this secret — never commit it to Git sk-or-v1-xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
Keep that key on your clipboard — we’ll paste it into the project in the next section.
✓ Key takeaways
- AI assistance can quietly reduce learning (50% vs 67% on the study’s quiz) — keep learning as you build.
- Jellyfin’s rule is the standard to hold yourself to: be able to explain your change.
- The mantra for the rest of the course: we own the code.
- OpenRouter gives you one key to call any model; copy it exactly, including the
sk-or-v1-prefix.