KORDUROYThursday, September 3, 2026 · Run 2 · 4:00 PM ET
Claude Goes Background as Nvidia's Chip Lead Narrows
5 stories from 1 source
TL;DR
- Claude Cowork and Claude Code can now complete tasks in the background on macOS, in beta for Pro and Max plans.
- Nvidia's AI-chip dominance is narrowing on multiple fronts, from OpenAI's own chip to Microsoft's Maia 300 to China's push for self-sufficiency.
- Shopify's CEO is betting on small, task-specific models over frontier scale for narrow workflows.
AI Tips2
Clean up a messy document with Claude in Word
- Install the Claude for Microsoft add-in and open Claude in Word's sidebar, selecting Opus 4.6.
- Upload the messy file — a Word doc, PDF, CSV, or image.
- Paste the prompt below to have Claude reorganize, clean up, and format it.
- Keep prompting in the sidebar to match a specific tone, brand style, or document format.
Clean up this document and organize it into clear sections with headings. Fix the grammar and improve clarity. Add a summary of the key points at the top, highlight important insights or repeated ideas, and structure any data into clean tables. Then rewrite everything into a polished, professional document that's ready to share.
Source: Superhuman AI
Run a prompt audit whenever a major model ships
- When a new frontier model launches, gather your saved prompts and system instructions.
- Ask the new model to review them for outdated rules, workarounds, or redundancies written for older, weaker models.
- Strip what the new model no longer needs and keep the leaner version.
Source: Superhuman AI (via AI educator Peter Yang)
AI News3
Claude can now complete computer tasks in the background
Claude Cowork and Claude Code can run tasks in the background on macOS, in beta for Pro and Max plans, limited to apps you've explicitly granted access to.
Why it mattersA real step toward set-it-and-forget-it agent workflows — worth piloting on one low-risk, permissioned task to see how much it actually offloads.
Source: Superhuman AI
Nvidia's AI-chip moat is narrowing on several fronts
Quarterly revenue has grown from roughly $7B three years ago to $96B now, and Nvidia is on track for 75% of the 2026 AI-accelerator market, but chip startups have raised $8.3B since April, OpenAI is touting its own "Jalapeño" chip, Microsoft is expected to unveil "Maia 300" this month, and Nvidia's China market share is projected to fall from 66% in 2024 to 8% in 2026 as Beijing pushes chip self-sufficiency.
Why it mattersUseful context if you're advising on AI infrastructure costs or vendor risk — pricing leverage may shift as alternatives mature, though Nvidia's CUDA software ecosystem still anchors most workloads.
Source: Superhuman AI
Shopify's CEO says small, task-specific models can beat frontier models
Tobi Lütke described an internal "self-improving flywheel" for training small models on niche tasks instead of defaulting to frontier models.
Why it mattersA useful counter to "bigger is always better" in budget conversations — a scoped, cheaper model may outperform on a narrow workflow.
Source: Superhuman AI
Summarized from Superhuman AI (Sep 3)
