AI news for builders and product teamsUpdated Nov 17, 2026, 17:00 UTC
AI news, straight from the source.
Releases, research, and ideas for developers and product teams. Updated every hour.
Across the feeds
Newest first
Latent Space's AI news roundup for 2-3 November 2026 leads with independent replications of "weight space programming," where frontier models edit neural network weights directly, while acknowledging some claims in its earlier coverage were presented too favorably.

AWS Quick now has Smart Approvals, which uses a separate classifier model to automatically approve routine agent actions while sending consequential ones, such as external sends, access changes, deletions and payments, back to the user. It is on by default, cannot exceed existing permission controls, and each approval is single-use.

datasette-llm 0.1a10 fixes a bug where the available LLM models were not correctly restricted to those set in configuration, affecting Simon Willison's agent.datasette.io demo site.

datasette-apps 0.3a0 adds #-tag autocompletion for apps in Datasette Agent chat, new tools for managing an app's stored-query permissions, and pre-creation checks on query existence, visibility and write access.

Datasette Agent 0.5a0 introduces "the panel", which keeps an agent chat open alongside Datasette pages such as tables and query views, and with datasette-apps lets users evolve an app from that adjacent chat.

Apple has begun beta testing its Audio Intelligence features for Apple Watch, initially for developers in the iOS 27.3 update, including Live Rewind, which shows a text snippet of the last 15 seconds of speech, and Siri Recaps, which summarizes conversations.

An OSINT researcher named Colsto shows that faint aircraft noise in a video can be tied to public ADS-B flight data, using Doppler shifts and flight timing gaps to pinpoint where the footage was recorded.

Simon Willison has published the October edition of his sponsors-only monthly newsletter, with paid subscribers getting early access ahead of a later free release.

A month after President Trump's executive order rebranding government AI as "Super Intelligence," adoption is uneven: federal agencies have mostly updated websites and solicitations, but Big Tech leaders and even some administration officials still say "AI."

Google has launched its Dreambeans AI storytelling app in India, its first market outside the US, with support for 13 local languages including Hindi, Punjabi, Tamil, Telugu and Bengali. The app generates personalized illustrated stories from connected Google services without requiring prompts.

Y Combinator CEO Garry Tan said he spends roughly $100,000 to $200,000 a year on AI tokens, using agents for coding, research and running his personal life. He made the disclosure on Walter Isaacson's podcast The Builders.

Apple researchers show that a two-stage distillation recipe can turn a pretrained Transformer into a pure Mamba model that nearly matches its teacher: the distilled Pythia-1B reached 14.01 perplexity against the teacher's 13.86.

Legora CTO Jacob Lauritzen described how the legal AI startup runs agent evals without ever reading customer data, by simulating client work in environments like a 10,000-file due diligence data room. The company recently raised a $550M Series E at an $8.8B valuation, per the podcast's framing.

NVIDIA released an open-source, Apache 2.0 workload-level recovery service that watches training jobs across a cluster via logs, GPU metrics and scheduler signals, then isolates faulty nodes and restarts from checkpoints. It ships with Slurm integration, starts in a non-acting monitor mode, and logged roughly 8,550 anomaly detections over three months on one cluster.

LangChain's Managed Deep Agents, now in public beta on LangSmith Cloud, lets teams define an agent as a Python project folder of files and deploy it with one command, with LangSmith running the Deep Agents harness and production runtime.

Anthropic's revised S-1A filing says the company may be close to meeting its own Responsible Scaling Policy definition of "dramatically accelerated progress attributable to AI," though no current model has crossed that threshold.

Zvi Mowshowitz disputes Scott Alexander's case against SURA, the strategy of urging AI labs, their employees and nations to stop building superintelligence unilaterally without waiting for reciprocity, arguing Alexander's claims about consensus, probabilities and alignment are overstated.

Engineer Venkatesh Rao argues in a "Contraptions" essay that fearing AI job losses mistakes intelligence for a scarce resource, much as mercantilists treated gold as finite, and that wider AI access could increase rather than reduce demand for human thinking.

Squarespace is cutting about 19% of its global workforce, roughly 300 employees, as part of an organizational restructuring the company says is meant to simplify operations and sharpen priorities.

The New York Times sued the privacy-focused browser company Brave in federal court in Manhattan, alleging it scraped the outlet's copyrighted content and distributed it to AI companies including Anthropic, Perplexity, Mistral, xAI and Cohere. The Times seeks an injunction and damages.