FORWARDED THIS? Subscribe here to get the next one.
ISSUE 13 · AI + COMPENSATION
The comp build that starts by throwing half the requirements away
TL;DR
Colette Leung took long term incentive off Google Sheets and into a system that now runs 1,200 users. The lesson she came out with has nothing to do with the tooling. Assume your requirement is stupid and too long, cut it in half, and only then let anyone build.
THE BREAKDOWN
Your requirement is probably stupid
Colette Leung says that about her own requirements cheerfully, as the person who has just spent two years replacing a long term incentive programme that used to live in a spreadsheet. She's Head of Total Rewards at Chainlink Labs.
When she joined, long term incentive ran on Google Sheets, there were four separate programmes waiting to be harmonised, and the vehicle underneath all of it was crypto. She could have gone out to market but they built in house instead, which she calls one of the crazier decisions they made, and she spent about a year as the business owner handing requirements to an external developer.
Give a developer too much context and keep piling change requests on top, and the code quality shatters. She says it's the same thing that happens when you overload a prompt. Anyone who has watched a good AI build go sideways after the fourth round of clarifications has met that failure already, just in a smaller room.
What it bought them is a real system. 1,200 users, new grants going out every month, settlements every month, live for a year. Have they been putting too much into the data foundation, probably more noise than real information would have come out of it.
The platform came from an external developer working to her requirements. The vesting simulator is the one she made herself, and her description of how is four words long: you just talk to Claude. Nothing about it is magical, she says, and she works at a company where building your own tools is the baseline expectation and not a party trick. She'd put the development share of a build at maybe 70% ten years ago and something like 10% now, and she hedges both numbers.
The simulator does one useful thing. It shows what a new joiner actually experiences on four year equal vesting with refreshes stacked on top. Years two, three and four pay above a line that was already market competitive, and then year five falls off a cliff. Change the schedule in the left hand panel and the whole picture redraws, which is how she gets leadership to see the problem rather than be told about it.
Her simulator, running on dummy data. She retypes the target and the whole board recalculates.
None of that removes her from the decision. Her exec team were impressed by the simulation and still needed her to explain the cultural and the financial impact of each option.
So the check she runs before she trusts anything she has built is to assume the requirement is stupid, assume it's too long, and cut it in half before a line of it gets built. Otherwise, in her words, you start automating stupid things.
Colette's test only works if you're open to be wrong about your own idea.
I've turned it into a one page brief you can run over your own next build by handing it over to your AI of choice.
ON THE RADAR
Being in an AI exposed job now makes you harder to hire, not likelier to be fired. The Bank of Canada published the analysis on 20 August. Setting occupations with full AI exposure against occupations with none, the gap in the job finding rate went from 2.2 percentage points across 2015 to 2019 out to 13.9 points in 2025. Over the same stretch the gap in the job separation rate moved from 0.0 to 0.1. Nobody is being let go more often. They just can't get back in. Exposure scores highest for receptionists and accountants. Read → the Bank of Canada analysis
Half of Euro area workers now use AI at work, and the median one saves three hours a week. The European Central Bank published the numbers on 26 August. Adoption ran 26% in 2024, 41% in 2025 and 52% this year. The median user saves about three hours a week, which the ECB puts at 7.7% of median working time. People using it to write or debug code save closer to eight hours, and they're 8% of the workforce. Read → the ECB numbers
A Maryland construction firm put pay up 35% to hold onto people, and AI infrastructure is why. The Federal Reserve's Beige Book, released 2 September, carries it as a retention move against data centre construction pulling workers away. The same section has recruiting agencies reporting heightened competition for AI leadership roles, and compensation for those people being driven up hard because they're difficult to find. Read → the Beige Book
PODCAST
Colette Leung: cut the requirement in half before you build. Long term incentive off a spreadsheet and into a system running 1,200 users, the vesting simulator she made herself, and the year five cliff it put on screen. Live this morning.
RECENT EDITIONS
When you're ready, here are two ways Range can help you.
1. Grab the free resources
Everything free we've made now lives in one place, the Comp AI Vault. The guides, the prompt library and the frameworks, all in the open. Browse the Comp AI Vault →
2. Join the community
Find your people. Inside Range, reward pros learn to build with AI together. Apply to join →
Reading this because a friend forwarded it? Get Range Weekly in your own inbox every Sunday. Subscribe free |
 Until next week, Giac |
 |