The AI Portfolio Simulator
Not every AI project deserves to ship. You decide which ones do.
Nimbus survived its AI transformation. Now everyone has an AI idea, and the board just made you Portfolio Lead. You are Claire Nakamura, and you have twelve months, $300k of budget, three engineers, two data people, and a queue of department heads who are each certain their pitch is the one.
Greenlit runs on the portfolio triage and capital allocation practice taught in executive AI leadership programs. Most AI initiatives should be right-sized or stopped, not shipped as pitched. Saying no well, and saying "smaller" even better, is the skill this game drills.
Four pitch months, eight build months. Read the one-pager, ask the sponsor what he will tell you for free, then spend real money on probes to check the claims before you stamp anything. Fund it at the right size, defer it, or kill it. Everything you fund locks engineers and data for months, so every yes is a no to something else. At the end of the year you are graded on the calls, not the vibes.
About the Creator. Built by Gary Wong, an enterprise AI transformation leader who made this entire game with his team of AI agents.
Disclaimer: A work of fiction. Every company, character, and rollout in this game is invented. Any resemblance to real companies or actual people is entirely coincidental.
This game records the calls you make, so it can be studied and improved. No account, no name, no email, and nothing that follows you between visits. What's recorded →
Reading the bars. Each column is scaled against its own worst month, not against the other columns. A tall bar under Bleed means your worst bleed month, not that bleed is larger than an injection. Sharing one scale would make a $3k burn invisible beside a $150k injection.
Money out reads −, money in reads +, and Bleed is striped rather than merely red, so nothing here depends on telling red from green.
Recorded: the verdicts you stamp, the evidence you buy, when calls happen, and your final grade.
Not recorded: no account, no name, no email, no cookies, and nothing that follows you between visits. This game has no text box anywhere in it, so there is nothing personal it could capture even by accident.
How your row is marked on the board: each playthrough gets an identifier used once and never reused. Your browser remembers your own years; the server never learns that two of them are yours.
Where it goes and for how long: a database hosted in the United States, kept for twelve months and then reduced to anonymous totals. It is used to work out whether this game teaches anything, which is the only reason it is collected.
Greenlit is written for a professional audience and is not designed for under-18s. If you would rather contribute nothing, playing offline changes nothing about the game: every screen, every number and your own board work exactly the same with the network switched off.
Twelve months as portfolio lead. Roughly twenty minutes. Read this once and you will not be guessing.
Months 1, 4, 7 and 10 are pitch months. The other eight are build months, where the work you funded either lands or does not.
Each department head brings a one-pager: the problem, the claim, and the ask. Ask your free questions first. Sponsors are not lying, but they are selling.
Probes cost real money out of the build budget: shadow the workflow $15k, audit the data $12k, interview the worker $8k, interview the sponsor $6k. Diligence you spend is diligence you cannot build with.
Six stamps, four of them sizes: Full AI ($120k, 2 eng, 2 data, 3 months), AI-assisted ($60k), Automation ($30k), Process fix ($10k), plus Not now and Kill. Right-sizing scores better than funding.
Budget $300k builds and probes. Token $80k pays the monthly burn, and $1 of budget buys $2 of token. Engineering 3 and Data 2 lock for the whole build, so a yes today closes a door in month 7.
You may buy capacity once all year: an intern $25k or a professional $70k, either for four months. It arrives the month after you pay, so you are buying it for a wave you have not seen yet. Hiring closes at the last pitch month, when a new pair of hands would arrive with nothing left to work on.
Raise one org posture meter: data governance, AI security, AI governance. AI-assisted and Full AI stamps need more than human oversight; each scores its required meters cumulatively. You never get enough actions for everything.
Projects run on their own. Watch the projected band, rescue what is slipping, and kill the zombies — a dead project keeps eating capacity and burn until you stop it.
At month 12 your year is measured against the best year this deck allowed — a year that paid real money, out of the same budget you build with, for the evidence behind every pitch it funded or killed. Walking away is free there, as it is here. You see that benchmark, what free perfect knowledge would have been worth on top of it, and the one call that would have gained you most.
It goes on your year-end review, next to your grade. Claire signs the memos; this is what the board calls her behind the glass. Spin again if you want a different one.
The benchmarks your year was measured against, what your diligence bought, and every probe you did and did not buy. This is the working, not the verdict.
This one never reached your desk this year. No probes, no budget — just your read. Your answers here are not scored into your year.
In real life there is no replay button. Here there is.