Korva is a scroll of the internet’s best ideas.

The take-home

The take-home is simple: use an agent on your own for a bit, produce an artifact, and it gets reviewed. Doing that once shows the person is AI fluent, and tool skills and other more junior abilities are becoming less relevant.

The whiteboard rounds

The whiteboard rounds go to setting up good experiments, understanding statistics, judgment, and systems design. A product team runs a lot of experiments, and each one has to reveal information instead of being fudged. With AI tools it is easy to offload thinking and judgment, and the whiteboard shows whether the person still has them.

Seniority

Hiring is biased toward more senior candidates, who grasp faster what drives the business: how the company makes revenue and keeps customers happy. Senior here is a spectrum. The target is someone still hungry who has done the job for a few years and is hitting their prime, which is put at 25 to 35.

Why test candidates with real work samples?

A 1998 review in Psychological Bulletin by Frank Schmidt and John Hunter compared selection methods across decades of studies. Work sample tests and structured interviews ranked among the best predictors of job performance, ahead of years of experience. A work sample shows what a person produces, not what they say they can produce.

Why do experiments need statistical care?

Ending an A/B test the moment results look good inflates false positives, because chance swings are easy to mistake for effects. Checking a test with a 5 percent significance threshold repeatedly can push the real error rate well above 5 percent. That threshold traces to Ronald Fisher's 1925 book Statistical Methods for Research Workers.

How does experience speed up learning a business?

Herbert Simon and William Chase's 1973 chess research found that masters recall board positions by grouping pieces into familiar patterns. Simon later estimated that masters hold tens of thousands of such patterns. Experienced people in other fields appear to build similar libraries, which lets them recognize a situation and spot what matters faster.

More from 20VC with Harry Stebbings

Tech & AIHuman data work is all edge casesRunning a human data project is a process of continually surfacing edge cases, and what gets done with each one matters a lot.
Enjoyed it?