article --summary
The summary
This is a collection of short notes from a technology writer, moving through several recent pieces on AI. The opening section looks at the November Thoughtworks Technology Radar and argues that scaling agentic AI from a small team to a large organization raises new questions about shared context, token spending, review load, security, and maintaining the shared platform.
The middle sections cover three other threads. One discusses how opposition to new data centers, which crosses party lines, could raise token prices for existing providers and push more work onto edge devices, and suggests established firms may lobby for rules that block newcomers. Another reports that subscription plans, rather than per-token API pricing, appear to be where AI vendors subsidize costs. A third describes research into editing neural network weights directly, and a framework called CAFE(S) for assessing whether the context given to an agent is clear, actionable, faithful, efficient, and safe to use.
Editorial note: This is a summary of an article first published by Martin Fowler. The full piece is on their site.