Raw LLM Responses

That's selection bias running the opposite direction from how you're reading it. The companies with the budget and the infra to substitute compute for headcount are exactly the AI-heavy ones, so they're where the swap would manifest first. Two takes in this thread are both off. "Just AI capex hole-plugging" misreads the cash flow (Meta and MSFT have plenty of FCF, the layoffs aren't bridging anything). "AI isn't actually replacing anyone" misreads the mechanism. What's shifting is opex composition. Each remaining engineer at hyperscalers now consumes orders of magnitude more compute than two years ago: IDE assistant tokens, agentic loops in CI, eval harnesses on every PR. Marginal productivity per engineer is rising fast enough that "cut a senior, route the budget to inference plus a junior using it" starts penciling at current model prices. That isn't AI doing the job, it's "1 engineer plus $200k of inference beats 2 engineers" on enough teams to show up in headcount. A small accounting firm in Cleveland has neither the inference budget nor the agent-tooling discipline to run that swap. Of course they aren't the bellwether.