Ai Coding And Product Systems hard
An AI code review bot comments on every PR. Telemetry shows 45% of its comments are dismissed as wrong, and engineers have started resolving bot threads without reading them, including the valid ones. What is the right next step?
- A Upgrade to a larger long-context model that reads the whole repository per PR, accepting roughly triple the cost to cut mistakes.
- B Post every finding but sort comments by model-reported confidence so the likeliest real issues appear first in the PR thread.
- C Retrieve more team rules for each diff chunk so the model has more guidance and can enforce a wider set of conventions.
- D Gate comments on a threshold calibrated with dismissal feedback, giving up recall until well under 1 in 3 flags are wrong.
Answer: D. Gate comments on a threshold calibrated with dismissal feedback, giving up recall until well under 1 in 3 flags are wrong.
Review tools **live or die on the false positive rate**. Above about 1 in 3 wrong, developers tune the bot out entirely, so trading recall for precision gives more value even though some real bugs go unflagged. Sorting by confidence (B) still posts the noise, so habituation continues. Sorting only works where low-confidence findings are hidden or collapsed by default.
One system design question a day, with spaced repetition so the ones you miss come back.
Get the appRelated questions
- Your assistant uses a classifier to route queries to a small or large model, which saves 60% of GPU spend. Analysis shows about 20% of hard queries are misrouted to the small model, which then gives confident wrong answers, and regenerate clicks on those threads have doubled. What should you do?
- Your workspace AI retrieves the top-k chunks from a shared vector index and passes them to the model with ACL metadata and a system prompt saying 'only use documents this user can access.' A user just received an answer quoting a private HR page they cannot open. What is the correct fix?
- Your support bot answers billing questions. One user has asked the same refund question three times, and retrieval confidence for that topic is low. What should the bot do next?
- Your AI review system receives a 4,000-line pull request spanning 60 files. Sending the whole diff in one prompt yields shallow, generic comments, and real bugs in files near the middle of the diff are missed. How should you structure the review?
- In your ChatGPT-style assistant, power users keep single threads running for weeks, reaching about 150K tokens. Cost per turn keeps climbing, latency is rising, and the model increasingly misses facts from early in the thread. What should you do?
- Your workspace AI filters retrieval using ACLs stored in the vector index, which is rebuilt nightly. An admin removes an employee from a confidential project page, but for the rest of the day the employee can still get AI answers drawn from it. What is the right fix?