Adam Witanowski on interviewing AI talent now the coding test is dead
“We've done a relatively good job of measuring what a token costs, and a relatively poor job of understanding what value that token added.”
The episode
Adam Witanowski built the AI delivery capability at nib: 18 systems, a knowledge graph across 3,600 repos, and AI agents shipping production code inside a regulated health insurer, for a team of more than 300 engineers. Almost nobody in Australia has done it at that scale. Then he went back on the market. Nine interview processes, seven offers, and one company telling him: we need you more than you need us.
So he has seen the AI talent war from both sides of the desk in the same year, and he walks James through the map. What back pressure is, and why your agents write code that does not fit your business without it. Why no ROI from AI is a measurement problem, not a technology problem. What senior AI talent costs in Australia right now, with real numbers from his own offers. And how to interview an AI engineer now the coding test is dead.
What we get into
- Back pressure: the force that shapes AI-generated code to your business, and the 18 systems nib built to supply it before agents touched production.
- Why "we see no ROI from AI" is a measurement problem. Engineers historically spend about 16 percent of their time writing code, so a 200 percent uplift on a thin slice moves almost nothing. Re-engineer the whole process or stay hamstrung.
- The comp map from his nine live processes: around $220k for the role here against $500k in the US, and the $300 to $500k starting range he gives companies making their first AI hire.
- The interview rebuilt: don't ask for a function, ask to see their harness, grade the answer from vanilla ChatGPT user to dark factory, and bring the whiteboard back.
- Renters, owners and caretakers: his test for whether to build AI capability in-house, upskill, or hand it to a consultancy that keeps the learning.
- He built an AI agent to interview for jobs on his behalf. Both sides of the hiring desk are automating, and motivation and taste are what still separate candidates.
Lines worth keeping
Tap to copy, then drop it into a post.
“Software eats the world, and that makes engineers apex predators.”
“Those same roles here are paying around 220 versus 500. Which one are you going to choose?”
“We've taken a bunch of learner drivers and given them an F1 car and said: drive around the city, have fun.”
“You can't price what you don't measure. Once you can see what one of these people does to your throughput, the salary argument ends.”
Why it matters
The first serious AI hire is the one most Australian companies are about to get wrong, on price and on the test. This is the map, from a candidate holding seven offers.
Chapters
Tap any chapter to jump to that moment in the video.
- Building an AI SDLC inside a regulated insurer
- Back pressure: why solo AI coding doesn't scale
- Re-engineer whole processes, not thin slices
- Most companies can't measure the problem
- Software eats the world. Engineers are apex predators
- Build in-house or buy: the consulting blind spots
- Renters, owners and caretakers
- The dollar gap: $220k here, $500k in the US
- What to pay your first AI hire
- The interview rebuilt: show me your harness
- He built an AI agent to interview for him
- James's take: a measurement problem, not a technology problem
Headcount & Code
Get the read that goes with the show
One issue every Wednesday, 7am. The moves that matter in Australian tech hiring, for the leaders making the hire and the tech professionals making the move.