Monday morning opens with the agent ecosystem asserting itself as a market force. A Raspberry Pi stock surge driven by OpenClaw buzz is the kind of story that sounds absurd until you remember that meme stocks move real capital — and this one has an actual use case underneath it. People genuinely want to run personal AI agents on cheap hardware. That's not a meme; that's a market signal.

Google's crackdown on OpenClaw users accessing Gemini via OAuth is more consequential than it appears. This is the first clear case of a major platform actively restricting agent access — not through pricing or rate limits, but through account-level enforcement. The 271-comment HN thread suggests this is hitting real users, not edge cases. If Google doubles down, it sets a precedent: platforms may treat agent orchestrators as unauthorized third parties, even when users explicitly authorize the connection. The agent ecosystem needs open API access to function; walled gardens are the existential threat.

The Khosla-vs-Altman frame on labor is instructive. Khosla says IT services disappear in five years; Altman says companies are "AI washing" layoffs that would've happened anyway. Both are partially right. The truth is messier: AI will eliminate certain categories of work (rote BPO tasks, tier-1 support) faster than new roles emerge to replace them, but the timeline is longer than VCs predict and shorter than incumbents hope. The Indian IT sector — employing millions — is the canary in this coal mine.

Chris Lattner's compiler review deserves close reading. When the creator of LLVM calls AI-generated code "competent but not general," he's identifying the current ceiling: AI writes code that passes tests but doesn't architect systems. That gap is where human engineers still live. For now.