Meta's $300 million package to lure OpenAI researchers isn't just a salary negotiation—it's a structural admission that the AI race is won by concentration of talent, not capital alone. Mark Zuckerberg has effectively monetised engineering the way football does strikers, and it's working. Sam Altman responded by tightening perks at OpenAI, but the math is brutal: OpenAI burns through investor money; Meta can redirect a rounding error from its annual revenue. The gap between what's fundable and what's profitable now determines who owns the next layer of AI.
But there's a darker angle buried here. The Economist's data on software developer job openings tells a story nobody wants to hear. Since February 2020, the market peaked at 230% above baseline in early 2022—then ChatGPT landed. Today, openings sit at 60% of pre-pandemic levels. The architecture is shifting: companies no longer need ten engineers doing mid-level work when five can do the work of eight using AI copilots. Junior developers and mid-career hands-on coders are being arbitraged out. Those building foundation models or AI products are golden. Generic developers are not. This isn't cyclical; it's structural.
Portugal's government AI strategy attempts to thread a different needle. Digitalising public services through centralised identity and interagency data-sharing is sound—and actually solves real friction (why submit university transcripts twice to the same state?). Adding AI to validate forms and route requests agentic-style makes sense. But the execution questions are unresolved: Will they license OpenAI's closed models or bet on Amali, the local open-source option? Will they hire internal engineering teams or contractor-dependent consultancies? The wrong answer creates lock-in or obsolescence. Meanwhile, six proposals to keep Portugal relevant—STEM expansion, overhauling computer science curricula, hosting AI multinationals, building autonomous vehicle testbeds, energy-heavy datacentres—are directionally right but require political will that budget cycles rarely sustain.