Opinions
Opinions are clearly labelled point of view, hype checks, first takes and working notes. They should never look like a tested verdict.
Check the cheap TV box, then read the AI worm story
Anthropic's 12 October 2026 research ran a 26B model across 15 cheap TV boxes, but its worm test used emulators and an open debugging port. The end-of-2026 date is a trend line.
Let Planner draft your status report. Keep the RAG call.
Microsoft's Planner Agent writes status reports from plan data, with general availability from late October 2026. A stale plan still gives a confident report, and the red, amber or green call needs a person.
Give Claude Haiku 5.5 the repeat jobs
Anthropic released Claude Haiku 5.5 on 7 October 2026 with API prices up to 90% below Haiku 4.5. Its own benchmarks still put Sonnet 5.5 well ahead on harder work, and the saving lands mainly on the API.
Sonnet 5.5 for scoped Claude work, Opus for judgement
Anthropic launched Sonnet 5.5 on 28 September 2026 as a faster, cheaper complement to Opus 5.5 for well-scoped everyday tasks. How to pin a default from that positioning without treating vendor scores as your own results.
Start with Opus 5 unless a named hard task needs Fable 5.1
A provisional rule for choosing Claude Opus 5 or Fable 5.1: start with the task, distinguish API prices from subscriptions and check the result.
Claude is adding text watermarks. Here’s what they can actually prove
Anthropic is adding statistical watermarks to text produced by future Claude models. The system can indicate likely Claude involvement, but it can’t establish authorship.
AI agent production readiness starts with a stop rule
OpenAI paused part of its frontier-model work after a security incident. The lesson for ordinary teams is simple: a successful AI agent demo proves too little unless the pilot has limits, monitoring and a stop rule.
Most AI demos skip the bit that matters
A demo is built to succeed. The part that decides whether a tool is any use, how it fails on a real task and whether you would notice, never makes the cut.
Get each opinion when it lands.
Guides, templates, builds, tests and opinions for practical AI at work, at home and for play.