What to assess in a full-stack developer
Breadth is the point of the role, but breadth without depth somewhere produces features that work in the demo and break in production. Look for one area of real depth and sound judgement everywhere else.
- End-to-end features: tracing a change through UI, API, data and back.
- The seams: validation on the server, permission checks, and errors that reach the user clearly.
- Data modelling: shaping tables and queries for the screens that use them.
- Frontend fundamentals: state, loading and error states, accessibility.
- Prioritising: knowing what to build simply now and what must be right from the start.
- Working with AI: using an assistant across layers without losing track of how they connect.
Screening questions for full-stack candidates
The screener follows your approved plan, probes vague answers, and opens a whiteboard or a code workspace when a section calls for one.
- Sketch how you would add team invitations to an app: tables, endpoints and screens.
- A user can see another team's data by changing an ID in the URL. Where did it go wrong?
- How do you show a form error that came from a database constraint?
- When would you choose a server action over a separate API route, or the other way around?
- An assistant writes the client and server halves of a feature. What do you check where they meet?
The full-stack coding round
A Next.js project suits full-stack roles because one round covers server code, client components and a live preview. The candidate works in a Cursor-like IDE with Monaco, a terminal, tests, completions, file-aware chat and an agent. If your stack splits frontend and backend, run a React (Vite) round and a Node.js round instead.
Code is tinted by authorship, and the server reruns the tests at the end. Guarded or unguarded mode is your choice, enforced by the server.
AI-fluency signals for full-stack work
| Signal | What it tells you about a full-stack developer |
|---|---|
| Time to first prompt | Whether they traced the feature through the code before asking for changes. |
| Prompt quality | Whether prompts give the assistant both sides of a change, not one layer in isolation. |
| AI-generated share | How much of the final feature came from accepted AI output, across layers. |
| Verification | Whether they checked the preview and ran tests after accepting changes. |
| Over-reliance flags | Large accepted blocks on either side of the stack that were never reviewed. |
| Recovery | What they did when client and server code from the assistant did not fit together. |
From scores to a decision
Fundamentals and AI fluency are scored separately and never averaged. Every round, from the screen to HR, sits on one record, and a named person records the outcome.
Questions
Which stack suits a full-stack coding round?
Next.js covers server and client in one project. You can also run separate React (Vite) and Node.js rounds.
Can the screen test both frontend and backend?
Yes. The plan drafted from your job description can include sections for each, with a whiteboard or a TypeScript workspace where they help.
How is a full-stack candidate scored?
Each score cites the answer or edit it rests on. A score with no evidence is dropped and shown as a gap rather than guessed.
What does it cost to assess one full-stack candidate?
An AI screen is 1 credit, or 2 with voice, and a coding round is 5 credits. Reading 40 applications against the role costs 1 credit.