Thursday, September 11, 2025

Start of archive·May 16

Top story

One Useful Thing (Ethan Mollick)

On Working with Wizards

As AI becomes more powerful and opaque, business leaders face a critical challenge: how do you verify the quality and accuracy of AI output when the technology operates like a 'black box wizard'? This verification bottleneck could become the limiting factor in AI adoption.

AI operates on a 'jagged frontier' where capabilities are unpredictable

The briefs

OpenAI's restructuring to grant equity in its PBC to the nonprofit signals a major shift in how AI's most valuable company is governance-structured—and hints at how it plans to reconcile for-profit scale with nonprofit mission alignment as valuations spiral.

OpenAI reaffirms nonprofit leadership structure

Vercel expands model routing to include Meituan's agentic model, reducing friction for developers building with non-OpenAI/Anthropic providers and signaling market fragmentation beyond the Big 3.

Meituan LongCat-Flash Chat model integrated into Vercel AI Gateway

Amazon researchers outline the critical open scientific questions blocking agentic AI progress—from agent communication protocols to privacy-preserving context sharing. This reveals where the actual R&D frontier sits, not the hype.

Four open scientific problems identified: agent language/communication, privacy-preserving context sharing, agentic negotiation modeling, commonsense policy understanding