AI news for builders and product teamsUpdated Nov 18, 2026, 03:01 UTC
AI news, straight from the source.
Releases, research, and ideas for developers and product teams. Updated every hour.
Across the feeds
Newest first
Meta released two new AI models, Muse Wonder and Muse Wonder Full, available today for Muse Code subscribers, with Wonder also accessible via the Meta Model API. Meta claims they rival the industry's best and plans to eventually power its Muse AI agent with Wonder.

Simon Willison profiled his web-edit app using the new Python 3.15 profiler, generating flame graphs and source heatmaps as demonstrations of the tooling.

Databricks published an installment of its Brickster Voices series profiling Clara Kurian, part of a recurring feature spotlighting employees and their personal career journeys.

NVIDIA and Princeton's Dao AI Lab have released ReplaySSM, a decoding technique that caches recent state-space model inputs instead of full recurrent state, cutting SSM state memory traffic roughly in half and speeding up hybrid-model decoding. It is already integrated into TensorRT-LLM, vLLM, SGLang and FlashInfer and runs in production in GLM-5.3-Flash and Kimi-K3.

Microsoft Research introduced StudentSim, a research system that trains personalized AI student simulators from public de-identified learner records so AI tutors can be evaluated and refined before human studies. In a chess tutor reinforcement-learning test, StudentSim-based reward reached 90.5% accuracy versus 75.7% with no RL and 71.6% using GPT-5.4 as the simulated student.

Simon Willison released web-edit 0.1a0, a Python web app that provides a browser UI for viewing and editing files in a folder, intended to run over Tailscale for phone-based editing. He built it with Codex and GPT-6 Astra Extra High after wanting to edit credentials without routing them through an LLM.

AWS has published a walkthrough for serving llama.cpp decision models on Amazon SageMaker AI, including a SemIf adapter that returns probabilities over caller-declared options instead of generated text. A benchmark of 231 JevBench tasks showed a GPU endpoint at 0.352s p50 versus 1.715s on CPU.

AWS has introduced a multi-node Ray LLM Deep Learning Container that bundles vLLM, Ray Serve LLM, NIXL and the EFA stack so one large language model can be split across two GPU nodes on Amazon EKS behind a single OpenAI-compatible endpoint. The walkthrough deploys Qwen3.5-9B with prefill on one node and decode on the other.

Honeyjar, a startup founded by PR veterans Michelle Masek and Nadia Jamshidi, is launching AI tools that automate communications work such as drafting announcements, building media lists, and finding speaking opportunities. It drew 10,000 waitlist signups in 24 hours, and beta testers include Nextdoor and Bloomberg Beta.

Microsoft president Brad Smith called for urgent global AI regulation on a BBC podcast, arguing the UK is better positioned than the US or EU to lead, and that voluntary self-regulation fails because only responsible companies comply.

Blueprints AI, a Santa Monica startup founded by Al Drams and Coco Cao, is launching a second-generation tool that generates code-compliant construction blueprints from a sketch, photo or PDF, with changes propagating across architecture, structural, MEP and interior drawings. Its first-generation software costs $3,200 to $8,500 per month, and licensed architects still seal documents before permit submission.

Flo Health used Databricks AI Runtime to build open generative AI models for its 81 million monthly users, prioritizing safety and usefulness in a health context with a high bar.

At the OCP Global Summit in San Jose, NVIDIA contributed Vera Rubin rack and compute tray designs to the Open Compute Project via its MGX architecture, launched new open source DSX AI Factory components, and published open AI factory standards with AMD, Google, Meta, Microsoft and Cisco.

TechCrunch Disrupt 2026 opens October 13 at 8 a.m. PT at Moscone West in San Francisco, running three days with 10,000+ attendees. Discounted passes are available for less than 24 hours before the event begins.

Spotify reports that AI-assisted development more than doubled its monthly merged code changes year over year, from roughly 8,100 to 17,000 in August, forcing changes to testing, observability and rollback capacity. The company says incident reviews found no material direct contribution from AI-authored code.

The White House has restored the UK AI Security Institute's early access to frontier models from Anthropic, Google DeepMind and OpenAI, reversing a September block after diplomatic pressure from UK AI Minister Kanishka Narayan.

Google has announced the Fitbit Edge, a $180 smartwatch with a slimmer 11.5mm profile and a hybrid OLED touchscreen. It is available to pre-order now and ships everywhere on October 27.

Microsoft announced Maia 300, its third-generation in-house AI chip, intended as the primary inference system for its cloud products, with production rollout planned for 2027. Microsoft says it should deliver 40 percent more tokens per megawatt than leading GPU-based systems.

Andrej Karpathy, OpenAI co-founder now at Anthropic, estimates only about 5,000 people worldwide have access to internal frontier AI systems, while roughly 75 percent of the global population has barely touched LLMs. He describes a widening gap between public perception of chatbots and what frontier systems can do.

Ambi, founded by former Plaud executive Peter Mo, has raised $26 million across multiple rounds and launched always-listening AI apps for iPhone, Mac, and Apple Watch that act as meeting notetakers and proactive assistants. The company also plans a desktop app and its own hardware device.