The video lays out five bad omens: falling enterprise adoption/ROI, coding copilots slowing long tasks, capex concentration about to cool, investors calling a bubble, and data center/power ceilings. My stance: don’t chase hype; AI won’t save a broken process. Keep teams tiny, wire models into real workflows, and judge by shipped work and throughput. Use LLMs for vibe/scaffolding, not critical paths; read the important bits. Ship software that stands without AI; meter LLM calls and price on impact so compute won’t eat margins. For backends, modern Java with Quarkus ships fast. AI augments; it doesn’t replace.