Social Media Briefing

Ethan Mollick argues that debates about US vs China AI competition are muddled because people mean different things by competition, and lists eight distinct framings from corporate profit to selling national tech stacks to reaching ASI first.

Mollick suggests large-scale management training for the AI era since working with AI agents resembles management, citing the WW2 Engineering, Science and Management War Training program as precedent.

Sam Altman marvels at his child combining two words for the first time, comparing his amazement to GPT-5.6 discovering new math.

Ethan Mollick argues that as working with AI agents increasingly resembles management, society may need large-scale management training for the AI era, citing the WW2 Engineering, Science, and Management War Training program that taught 1.8 million people and helped fuel the postwar boom.

Harrison Chase argues the agent industry has shifted from agent frameworks like LangChain, AI SDK, and LlamaIndex toward agent harnesses such as deepagents, Claude Agent SDK, and EVE, noting deepagents predated EVE.

Gary Marcus quotes a claim that AI now costs some companies more than the workers it was meant to replace.

Jerry Liu promotes LiteParse, a document parser for the agent loop, citing exponential growth in agent queries with file attachments and integrations with the Vercel Eve framework and Claude SDK, arguing default light parsing causes hallucinated context and poor retrieval.

Detailed prompting tips for the Fable model, advising to align claude.md with best practices, make skills less prescriptive to avoid rerouting to Opus 4.8, and include the WHY (goals, metrics, ROI) in prompts.

Simon Willison describes using Claude Fable to do a final review of software before release, where it found and fixed five release blockers at an estimated unsubsidized cost of about 149 dollars, calling the experience humbling.