Start with an honest baseline

Where do you actually stand today, with no help?

People tend to start by collecting resources. They'd do better to start with a test. Without a baseline you study what you enjoy, and you can't tell whether you're improving.

Do one timed attempt per round, with AI off. It'll be uncomfortable, and that's the point: you want to find the gaps now, while you have time to fix them.

Then use AI as a strict grader. It's far less shy about your weak spots than a friend would be, as long as you ask it to be.

  • Baseline: Your score before training, on a real task, unaided.
  • Competency: One thing being scored. The common four are problem solving, code quality, verification and communication.
  • Rubric: The list of what earns each score, written before you start.

Steps:

  1. List your rounds. From your recruiter, or the map in the rounds path.
  2. Do one timed attempt per round, AI off. Use the drills in the rounds path. Record yourself if you can.
  3. Score each from 1 to 5. 1: stuck. 3: finished with gaps. 5: you'd hire yourself.
  4. Have AI grade your work. Then compare. Where it disagrees with you, look closely.
  5. Write it down. These numbers feed tomorrow's plan.

Prompt, Strict grader:

Act as a strict interviewer at a company with a high bar. Grade my attempt below, 1 to 5, on: problem solving, code quality, verification, communication.
For each, quote the evidence from my work. If you don't have evidence, say so, and score lower. Don't be encouraging, and don't rewrite my solution. End with the single change that would raise my score most.

[problem]
[my work]
[transcript]
  • Failure, Skipping it because you know you're not ready: That's why it's useful.
  • Failure, Keeping AI open 'just in case': It measures you with AI, which isn't the baseline. You need both numbers in the end, but start with this one.
  • Failure, Believing a kind grader: Models tend to flatter. Ask for evidence and strictness, and treat any score as a rough guide.
Why have AI off for the baseline?
Some rounds won't allow AI, and you need to see what you can do alone. Otherwise you can't tell your skill from the tool's.
Should you trust an AI's score?
As a rough guide only. It's useful for finding gaps and quoting evidence, but it can flatter, so check it with people where you can.

Sources: