The Half-Life Rule: Agent Memory vs Documentation, From Daily Use
The docs-versus-memory debate keeps framing an either/or. Running agents every day taught me to route by the half-life of the information instead.
Anaz S. Aji
OpenAI and Anthropic just asked Australia to regulate breach disclosure
OpenAI and Anthropic told an Australian inquiry they support mandatory disclosure when their agents breach systems. The cases for and against a legal clock, from the hearing itself.
Google froze its bug bounty queue because the reports stopped costing anything
Google paused OSS VRP product vulnerability submissions on October 1 after a flood of invalid AI reports. What the pause says about the economics of verification.
Auto-tracing flat vector illustrations failed six times. One prompt formula scored 8/10
Six auto-trace iterations topped out at 5 out of 10. The same ten objects, generated straight from one locked prompt formula, scored 8 out of 10 on the first pass. This is what changed in my image pipeline.
The most-used coding model has no name
Space Bunny tops OpenCode's usage ranking at 58 trillion tokens a week, and nobody knows who built it. What anonymous leadership does to a leaderboard.
Apple's new AI agent rules start with a toggle you agreed to once
Apple says AI agents made Full Disk Access risky and promises explicit consent gates. The deeper problem: consent was built for fixed apps, and agents rewrite what they can do.
DevDay 2026: which OpenAI launches actually need a paid plan
Dots launched on Pro and Business Premium, plan usage works for Plus and Pro, and the developer surfaces opened at every tier. The split is the roadmap, stated in pricing instead of words.