This week AI stopped waiting for your orders. Agents that finish tasks on their own, a voice that sounds almost human, and a free Chinese model catching up to the best in the US all point to the same shift: AI no longer just answers, it acts.
I’ve filtered out the noise for you
Every week I go through the flood of AI news and leave you only what actually matters for your business, in a few minutes.
1. Claude Duet now connects to your apps and finishes your tasks before you ask it to.
Anthropic released a preview of Claude Duet, an agent built on Claude Opus 5.5 that does not wait for instructions. With your permission it reviews your email, decides what matters, and carries out tasks on its own, including multi-step ones like buying a plane ticket and checking you in. It is the jump from AI that answers you to AI that acts for you while you are not watching.
2. Did Jack Dorsey just build the most human-sounding AI voice yet?
Block, Jack Dorsey’s company, revealed Dessa, a voice model trained only on real human conversations. Instead of running through text, it processes audio directly, so it picks up emotions like laughter or sadness in real time. Block says it can run locally on a MacBook M4 Max, with no cloud required.
3. China’s new Kimi AI rivals America’s best, and they’re giving it away for free.
Chinese company Moonshot AI released Kimi K3.1 Pro, an open-weight model with roughly 3 trillion parameters and a one-million-token context window. Moonshot claims it rivals the best US models at coding and plans to release the full weights around November 4. One striking detail: it says it deliberately reduced the model’s hacking ability before shipping it.
4. Is OpenAI coming for Jev with a new API that makes millions of decisions for cheap?
OpenAI made its Decisions API generally available, running on GPT-6 Luna with native integration into GPT-Live. It is built for small, repetitive, low-latency decisions: check a condition, choose among fixed options, or score text, images, or JSON against a rubric. The company frames it as its answer to TypeSafe’s Jev model, betting that thousands of these tiny decisions get automated for next to nothing.
5. How did a San Francisco startup called Town hit a billion dollars in just four months?
San Francisco startup Town raised $108 million in a Series B that valued it at $1 billion, just four months after its public launch. Its product runs several AI assistants that work as a team: they research your meetings, filter out the ones that add no value, and hand you a short briefing saved as a document.
Bonus: Content of the Week
This week’s pick is a video on San Francisco’s history as the world’s tech capital, framing the current AI boom as a new gold rush.