Sample Assessment Reports

“I usually start by writing a detailed system prompt that includes the tech stack, coding standards, and the specific output format I need. For complex tasks I break the prompt into sections with clear headers.”

“I never deploy AI-generated code without reviewing it. I check for edge cases, null handling, and whether it follows our team's patterns. I've caught subtle bugs like off-by-one errors and incorrect timezone handling.”

“If the first output isn't right, I don't just say 'try again' — I explain exactly what was wrong and why. Like 'the error handling is too generic, I need specific error types for each API endpoint because our monitoring depends on error classification.'”

“I maintain a project context file that I paste at the start of new conversations. It includes the architecture overview, key abstractions, and current sprint goals. Saves a lot of back-and-forth.”

“When I need to refactor a large file, I paste the whole file and ask for specific sections to be changed, making sure to reference line numbers and function names so the AI has clear anchors.”

“For a large feature, I break it into phases — data model first, then API, then frontend. I'll do each phase with the AI separately so it can focus on one concern at a time rather than trying to generate everything at once.”

“I keep a mental model of what the AI got right and wrong in previous attempts, and I'll reference that explicitly — 'the approach you used for the auth middleware was correct, apply that same pattern here.'”

“I use Copilot for boilerplate code and Claude for architecture decisions and code review. For frontend components, Cursor with Copilot is fastest. For complex backend logic, I prefer Claude's longer context window.”

“For database queries, I always test the generated SQL against our staging data before merging. AI tends to write correct SQL for simple cases but misses index implications on large tables.”

“AI is part of my daily workflow — I use it for code generation, debugging, writing tests, and documentation. It's integrated into my IDE and I also use the web interface for longer planning sessions.”

“I know AI struggles with math, counting, and complex multi-step reasoning. For anything involving calculations, I always verify manually or write a test.”

“I never paste production credentials or customer data into AI prompts. We have a team policy about it. For debugging data-related issues, I create synthetic datasets that mirror the structure without real PII.”

“I keep up with new models and capabilities. When Claude 3.5 Sonnet came out, I did a systematic comparison on our common tasks to see where it outperformed what we were using before.”

“Code quality has definitely improved since I started using AI for reviews. It catches things I'd miss — like inconsistent naming or missing error handling in edge cases.”

“I don't use AI to write performance reviews or sensitive communications. Those need a human touch and there are ethical concerns about bias in AI-generated evaluations.”