Humanoids in Seoul, Anthropic's troubles, and when Postgres wins

Sunday afternoon, and there's a pattern emerging across the tech landscape that's worth sitting with for a moment. While governments are squaring off over AI regulation and companies are building ever more complex systems, some of the most practical work happening right now is actually the opposite - people throwing out the complexity and going back to basics.

When boring beats flashy

There's a fascinating thread running through developer conversations this week. Someone built what everyone said was the right way to handle AI - vector databases, separate services, orchestration frameworks, the works. Then they hit a real problem: when a user deleted data from their main database, the AI kept hallucinating answers based on deleted records. The fix? They threw it all away and moved everything into Postgres with a vector extension. Same machine. Same database. No sync nightmares. The result was simpler, faster, and actually more reliable than the "proper" architecture. This isn't unique - it's becoming a pattern. People are discovering that the most sophisticated solution isn't always the one with the most components.

China's humanoid moment

Meanwhile, something significant is happening in manufacturing. China's five leading humanoid robot companies - AGIBOT, Fourier, Huawei, Leju, and Unitree - are exhibiting together in Seoul next week at Automation World 2026. This is the first time these companies have gathered outside China to present to a global audience. They're bringing working robots: X2 and G2 models from AGIBOT, the G1 from Unitree. Not prototypes. Not concepts. Working machines being deployed in real manufacturing and service work. For anyone watching the global automation race, this matters. Korea is positioning itself as a hub. China is moving faster than most observers expected. And the conversation is shifting from "will we have humanoid robots?" to "who will dominate this market?"

The regulation backlash

And then there's the policy moment that's causing real friction. Anthropic was designated a supply-chain risk by the US government, effectively barring federal contractors from using Claude. The stated reason: Anthropic refused to remove safety constraints around military use. Hours later, OpenAI published a statement saying they don't think that's appropriate. What's striking isn't the designation itself - it's the speed and the apparent coordination. Gary Marcus laid it out plainly: OpenAI signed a competing deal with similar terms, but OpenAI's leadership had given more to the right political campaigns. Whether you think Anthropic's safety stance is right or wrong, the mechanism here - punishment for refusing to relax safety guardrails, then favouring a competitor with better political connections - reeks of something beyond normal competitive pressure.

These three threads tell you something about where we are: developers solving problems with pragmatism over hype, manufacturing moving into a new era at speed, and policy becoming increasingly tangled with corporate politics. If you're building something, the Postgres story matters. If you're watching manufacturing, the Seoul conference matters. If you care about how AI gets built, the Anthropic moment matters a lot.