"What's missing from this conversation is not concern for workers, it's accountability for capital. This fear-laden narrative is coming from the very CEOs who have received billions of dollars in funding - without the return on investment to justify the scale of those bets. Even as they forecast workforce disruptions and the end of software engineering, they're still hiring thousands of engineers. The contradiction is hard to ignore."
"Software and data stocks have plunged by many billions of dollars following the release of new tools such Anthropic's Claude Cowork and OpenAI's Codex. These systems can now write software and launch programs without users ever learning to code. They can also handle data management, review contracts and perform a wide range of industry-specific tasks. Compared to traditional software economics - expensive, per-seat licensing - this shift matters."
"There are two ways to interpret these developments. One is grounded in reality: these are engineering advances that improve productivity and reduce friction. The other narrative is far more dramatic. In that version, AI models are framed as unstoppable forces poised to replace human labor. That story isn't accurate. But there's a reason people are telling it. Training and running AI models like Codex and Claude is extraordinarily expensive."
CEOs forecast widespread job displacement while having received massive funding without commensurate returns, yet continue hiring engineers. New AI systems can write software, manage data, review contracts and perform many industry-specific tasks, eroding traditional per-seat licensing models and contributing to large stock declines. Two narratives exist: one frames AI as productivity-enhancing engineering progress, the other frames it as an uncontrollable force replacing labor. Training and operating large models requires enormous upfront investment, vast compute capacity, sustained energy use, and ongoing power and cooling costs that become persistent operational expenses.
Read at Fortune
Unable to calculate read time
Collection
[
|
...
]