Singularity.Kiwi Newsletter · 19 October 2026
Robots can do 74% of the work. They're cheaper for 0.3% of it — that gap is the whole story.
The numbers this week kept arguing with each other, which is usually a sign the truth is somewhere in between. The IMF counted 60% of New Zealand workers as exposed to AI; Anthropic’s robot index says 74% of physical tasks are automatable but cost-competitive for almost nobody. Exposure isn’t destiny — and the pay data starting to move through the economy is beginning to prove it.
Robots Can Already Do 74% of Physical Work. They’re Cheaper Than People for 0.3% of It. Anthropic’s robot exposure index finds robots can do 74% of US physical tasks — but they’re cost-competitive for just 0.3%. That capability-versus-budget gap is where the real robots-and-jobs story will play out. https://singularity.kiwi/robots-can-do-the-work-but-not-the-budget-2026/
The IMF Counted New Zealand’s AI Exposure — 60 Percent of Workers, and It Complicates the Fear Story Six in ten NZ workers sit in high-exposure occupations, but exposure is not destiny: the newest job-ads and pay data increasingly split the difference between panic and comfort. https://singularity.kiwi/imf-nz-women-ai-exposure-2026/
A Four-Week Course, a $14-an-Hour Past, and a Direct Line Into the AI Buildout A rural Louisiana community college runs a four-week AI infrastructure course that lifts pay from $14 to $27 an hour, with a hiring pipeline that has barely failed anyone yet. https://singularity.kiwi/louisiana-meta-training-jobs-2026/
Four Gemini Agents Beat Minecraft Without Anyone Playing: 2,783 Programs, 27 Hours, 605 Deaths Four Gemini 4 Argon agents, zero human input, one Ender Dragon. The first end-to-end autonomous completion ran 27 hours, wrote 2,783 JavaScript programs, cost about $520 — and doubles as a proxy for long-horizon enterprise work. https://singularity.kiwi/gemini-4-argon-agents-beat-minecraft-2026/
DeepSeek Ships V4.1-Flash-1017: Same Price, Faster Tasks, and the Web Search Engine Opens Up A minor version bump with a major Pareto win: the 1017 snapshot beats DeepSeek’s own September model on eleven benchmarks at the same price — and opens the search engine to everyone. https://singularity.kiwi/deepseek-v41-flash-1017-update-search-api-2026/
The AI Grading the Fourth Graders: Albany Parents Want a Say Before the Chatbot Tests Their Kids Hundreds of Albany and Bethlehem students now take AI-run checks after class assignments. Parents formed chapters of a national group to slow the rollout; the district insists instruction stays human-centred. https://singularity.kiwi/albany-snorkl-parents-pushback-2026/
Kyutai’s REKI Gives Robots a Working Memory in Just 2.4 Billion Parameters A 2.4B-parameter open-weights VLA with built-in memory matches models twice its size on manipulation benchmarks — and Kyutai released everything: weights, code, data pipeline, deployment stack. https://singularity.kiwi/kyutai-reki-open-vla-robot-memory-2026/
— CJ Murden, Singularity.Kiwi
You’re getting this because you subscribed at singularity.kiwi — one email each Monday, that’s it.