Scenario
Your sprint ticket is to add a complete billing module: subscription management, invoice generation, and payment processing. These touch the user service, database layer, and a third-party payment API.
1. How do you approach this with AI tools?
A Split it into subscription, invoices, and payments, and prompt for each part separately. B Before writing any code, produce a specification document: data contracts between the three components, the interface each exposes, error handling across boundaries, third-party API mapping, and acceptance criteria for each part. Validate the spec with AI for gaps, then implement module by module with explicit review at every integration boundary. C Describe the entire billing module to AI and build from what it produces. D Break the work into logical layers, sketch the interfaces between them, and implement each layer sequentially with AI, testing integration at each step.
Scenario
You have three AI sessions open: one generating tests for a feature you finished this morning, one building a new API endpoint mid-way through, and one investigating a production bug. A response arrives in the test session while you are mid-prompt in the API session.
2. What do you do?
A Finish the current prompt, then switch to review the tests. B Switch to the test output immediately. It has arrived and needs to be processed. C Complete the current atomic unit of work in the API session, update your task log (current state, what was just completed, what comes next), then move to the test review. Each session has its own context document; you never context-switch mid-execution. D Reach a natural pause point in the API session, note where you are, then review the tests.
Scenario
A developer needs to delegate the implementation of a JWT authentication module to an AI agent for a Node.js/Express API.
3. Which of the following specifications would most reliably produce a production-ready result?
A "Build a user authentication system with login and registration." B "Auth module. JWT. Register and login. TypeScript. Check existing patterns in the codebase." C "Build a JWT-based auth system in Express. Users register with email and password and log in to get a token." D "Build a JWT authentication module for our Node.js/Express API. Endpoints: POST /auth/register (name, email, password), POST /auth/login (returns access token 15min + refresh token 7d). Middleware for protected routes. Stack: Express 5, Prisma, PostgreSQL. Match error handling in /src/utils/errors.ts. No Passport.js. Full TypeScript types. Tests in existing Vitest setup. Constraints: bcrypt for passwords, no plain-text secrets."
4. During a complex AI-assisted task, the agent produces something fundamentally wrong at step 4 of 7. What do you do?
A Stop the agent, diagnose the root cause (bad specification, lost context, or wrong approach), fix it, restart from the last verified checkpoint, then document the failure pattern and evaluate whether this task type is suited to AI delegation at all. B Accept the result and work around the issue manually later in the process. C Stop the agent, diagnose whether the failure was a bad specification, lost context, or wrong approach, fix the root cause, and restart from the last verified checkpoint. D Ask the AI to redo step 4 with a correction and continue from there.