▶ YouTube Summaries

• https://www.youtube.com/watch?v=KO1LB7ccfdY
• Original title: Microsoft DOOMs ID with layoffs | TheStandup

A solo-recorded Standup episode where Prime, joined by veteran game programmer Casey Muratori, digs into Microsoft's latest round of layoffs and their outsized impact on id Software. Roughly 1,600 people were cut (with ~1,600 more planned through fiscal 2027), and about 10% of the cuts — 132-ish of 1,600 — landed on id, gutting most of the engine team plus designers and leads. The conversation moves from "is id doomed?" to a grounded discussion of what modern engine development actually requires and why AI does not obviously rescue it.

The layoffs

Announced via an exposé from Asha Sharma, the new head of Xbox, citing various goals and revealing management details about Xbox and its studios. id is the star of the layoff story: most of the engine team, plus designers and people running it, are gone, leaving whoever remains with what Prime calls an improbable-to-impossible task of making anything. Prime jokes about Microsoft measuring by fiscal year — imagining Satya Nadella and Asha Sharma throwing a fiscal New Year's party, singing an Auld Lang Syne about accountability margins.

Casey: no data, so no verdict

Casey declines to pass judgment because there are no inside stories. The scenarios span a wide range: engine staff waking up to an HR layoff notice, versus being asked to move onto a centralized engine division serving other properties (Minecraft, King, etc.) and choosing voluntary layoff instead. Without knowing which happened, he cannot call it good or bad — the only clearly good outcome would be the team collectively walking to form a new id-like studio elsewhere. Short of that, there is no way to spin it positively, but how bad it is slides based on facts he does not have.

What engine development actually costs

Prime's real question: can id remain id — producing technical masterpieces like the Doom series — after losing a big chunk of its team? Casey's answer reframes the problem. With good engine people, writing the code is not the bottleneck. The hard parts live in two other domains. First, figuring out what should be built at all: active research into strategies for dynamic lighting, level of detail, ray tracing, and re-architecting when hardware capabilities shift discontinuously. That work is inherently experimental — "we tried this, and this, and this" — as seen in talks by the Nanite creator, John Carmack, and Michael Abrash on Quake. Every experiment needs a person to run it.

AI does not make it free

Casey stresses that even in an AI-assisted future, the models need heavy human guidance — someone to tell them what to try constructively, build test cases, and check whether they actually did it right or broke something. Prime adds the economics: each experimental loop can cost $10–20K in compute plus two weeks of running, gathering, and testing, on top of engineers' time — AI does not make it free. Casey has seen no evidence yet of significant speed-ups in this domain; the idea that one engine coder could kick off tons of agents to run all the experiments may be a possible future, but no studio has it today. So the budget you need scales directly with how much of that experimental work you envision — that is prong one of the answer. The episode cuts off mid-discussion, pointing viewers to Spotify for the rest.