AI-fluency assessments for engineering teams
Interview engineers the way they actually work
Engineers don't write code by hand anymore. Crafit gives you a streamlined way to measure how well candidates build with AI coding agents.
1 class TokenBucket:
2 def __init__(self, rate, capacity):
3 self.rate = rate
4 self.capacity = capacity
5 self.tokens = capacity
6 self.updated = time.monotonic()
7
8 def allow(self, cost=1):
9 self._refill()
10 if self.tokens >= cost:
11 self.tokens -= cost
12 return True
13 return FalseTerminal
$ pytest tests/ -q
....... 7 passed in 0.12s
How it works
01
Pick a task
Choose from our library of realistic engineering tasks or bring your own repo.
02
Candidate works with an agent
They code in a browser IDE alongside Claude Code or Codex, the way they work day to day.
03
Review the session
Replay every prompt, edit and test run, with an AI-fluency scorecard to compare candidates.
See the signal, not just the output
Final code only tells you so much. Crafit scores the process behind it, so you can tell who steers an agent well and who just accepts what it hands back.
Task decomposition
Do they break work into steps an agent can execute well?
Prompt clarity
Are instructions specific, scoped and well-contextualized?
Verification
Do they read diffs and run tests, or trust output blindly?
Course correction
When the agent is wrong, how fast and how precisely do they fix it?
AI fluency score
84/100
Session replay · 24 min
- 02:10Scoped task into 3 steps before prompting
- 08:41Accepted agent diff without running tests
- 12:05Caught missing lock, re-prompted with specifics
- 19:30Wrote edge-case test the agent had skipped
Ready to hire for how engineering works now?
We're onboarding our first design partners. Talk to us.
Get in touch