Insights
Practical analysis and guides to help you buy, build, and work with AI more effectively.
OpenAI extends prompt cache to 12 hours, and long-lived agents can cut input cost by 70%
Cache goes from half an hour to half a day On October 20, OpenAI added a new option to the Responses API: prompt_cache_options.ttl can now be set to 12h. It means the stable prefix, system instructio
Coding agents got seatbelts this month, not smarter models — what teams running them unattended need to know
In the first ten days of October, three mainstream coding agents shipped new builds within the same week. Claude Code, Codex, Cursor, none left out.Guess what the biggest change was. Not a smarter mod
Anthropic lets one agent write a program to manage a swarm, and multi-agent finally gets practical
On October 9, Anthropic added a test capability to its managed agents called dynamic workflows. In plain terms, you can now let a single agent write a program that breaks a job like reviewing hundreds
Agents start checking out on their own: WebMCP pulls agent-to-agent commerce into reality
Shopify opened checkout to browser agents In early October, Shopify let browser-based AI agents complete purchases through the WebMCP protocol: read the cart, change quantities, submit checkout, acros
OpenAI gave every Dot its own cloud computer: AI is no longer just a chat box, it has a workstation
The most visible DevDay move was handing agents a computer At DevDay in San Francisco on September 29, OpenAI launched Dots. The counterintuitive part is not that it is "smarter" but that every Dot in
Vera Rubin enters full production: NVIDIA reworks the economics of the AI factory
Huang repositioned Rubin as the foundation of the "agentic AI factory" After the GTC 2026 keynote, the Vera Rubin platform is in full production, with deliveries starting in the second half of 2026. T
Zhang Yiming Is Personally Driving It: ByteDance's Seedance-Based World Model Could Surface as Soon as October
A Bloomberg report on September 7, with the founder as the headline ByteDance is building a real-time spatial video generation model on top of its Seedance video model. What makes this unusual is Zha
30 Billion Parameters Fitted Into a Phone: On-Device Models Crossed a Line This Autumn
Get the number right: 30 billion total, 2 billion active On September 7, Huawei launched the Mate XT 2 tri-fold in Guangzhou and used the stage to unveil the Kirin 9050 Pro. The line that grabbed att
Top Two Coding Agents Are 0.4 Points Apart: The Race Is No Longer About Being Smart
The top two are 0.4 points apart. This is no longer about who is smarter. On August 21, an independent Terminal-Bench 2.1 score dropped. In the coding-agent track, the gap between first and second is
Grok 4.6 Ships Built for Long-Running Agents, Not Benchmark Glory
Five weeks later, a different kind of release On August 12, SpaceXAI (formerly xAI) shipped Grok 4.6. Just five weeks after Grok 4.5. The interesting part is not that it got smarter. It is that the mo
China's Open-Weight Wave: Qwen Drops 2.4 Trillion Parameters, DeepSeek Moves to Peak-Off-Peak Pricing
One week, two headlines Early August delivered two Chinese model stories back to back. One lifted the "largest open" bar straight to 2.4 trillion parameters. The other quietly changed its pricing, and
Gemini 3.7 Flash Is Out (2026): Near-Production Coding, and the Truth About Half-Price Tokens
Bottom line first On August 13, Google shipped Gemini 3.7 Flash. Its positioning is clear: not the smartest model, but the best value one - built specifically for agents that need to run in productio