Innovation happens elsewhere
Things other people are building, writing and thinking about that caught our attention. Inclusion means we think something is worth looking at. It does not necessarily mean we agree with it.
CodeRabbit compared AI- and human-generated pull requests. The useful question for us is not whether AI code is simply “better” or “worse”, but whether characteristic weaknesses can be anticipated and designed around.
The Linux Foundation’s Agent2Agent project is interesting because agents from different providers need a neutral way to communicate if we want to avoid recreating provider lock-in at the agent layer.
Kiro, Microsoft’s Spec Kit, OpenSpec and Martin Fowler’s writing all approach the question of how much explicit specification AI-assisted development needs. That overlaps directly with questions we are exploring ourselves.
An essay about people consulting machine systems for guidance, and what happens when those systems begin to acquire authority in everyday life. Close to questions we care about around AI interpretation and human agency.
A project clustering language patterns in Claude-generated GitHub pull requests. Interesting as a way of looking for the linguistic fingerprints and repeated structures of AI-generated work.
Headcount models a company as an organisation of installable Claude Code agents, departments and skills. We are interested because it approaches virtual organisations from almost the opposite direction to our own work.
Apertus, GreenPT and the European Open Source LLM catalogue point at different parts of the same question: what useful European AI infrastructure can look like when openness, privacy, sovereignty and energy use matter.
Stanford HAI’s annual overview of AI research, investment, adoption and capability. Useful background when individual announcements make it difficult to see the larger direction.
Charity Majors argues that the practical case for AI-assisted development changed sharply between 2025 and 2026. Worth reading whether or not you agree with every conclusion.
A broad look at actual AI use, including the unexpectedly large role of fiction and storytelling. Useful because real use tends to diverge from the categories technology companies expect.
For this first import, the date is when the item entered Elsewhere. The pasted Slack export does not expose a reliable message date for most links. A future Slack/API collector can use the original message timestamp.