AI news for builders and product teamsUpdated Nov 18, 2026, 09:00 UTC
AI news, straight from the source.
Releases, research, and ideas for developers and product teams. Updated every hour.
Across the feeds
Newest first
LangChain published details of its Paid Media Agent, a Slack-based long-running agent it open-sourced, reporting that paid media grew from 0 to 20% of marketing pipeline in six months, cost per qualified lead fell 30% from June to August, and an early reporting workflow became about 40x cheaper and 13x faster.

NASA astronaut Christina Koch speaks with Google's James Manyika in a Dialogues on Technology and Society episode covering her spaceflight career and the roles of robotics and AI in exploration.

NVIDIA reports that its Transformer Engine with JAX raises DeepSeek-V3 MoE training throughput from 103 to 1,068 TFLOPS/GPU on GB200, a 10.4x gain. The stack sustains 97% scaling efficiency at 1,024 GPUs on GB300 NVL72.

Google's DevFest 2026 runs October 1 through December 31, with more than 800 events across 115 countries and nearly a million developers expected to take part. The theme is "Build, Secure, Scale: Developers and Builders in the Agentic Era."

Dario Amodei published a new essay titled "We Must Pace the Frontier," proposing three measures to slow the rate at which AI capabilities increase. Anthropic unilaterally committed to the first proposal, embedded evaluators, and OpenAI followed, with Elon Musk and Demis Hassabis endorsing the overall proposal.

Gary Marcus wrote about a September 24 date he ties to Trump and AI policy, citing falling AI stocks, anti-AI public opinion, and reports that Steve Bannon and Bernie Sanders will push for human-controlled AI. Marcus also described an Economist essay on a possible Trump-Xi AI deal.

IEEE Spectrum AI lists an interactive webinar titled Responsible AI for Higher Education, presented by IBM, covering types of AI, concerns with AI, IBM's approach to Responsible AI, and guidance for students in IEEE chapters.

OpenAI unveiled Jalapeño, its first AI accelerator, on 25 August, citing up to 13.4 petaflops of 4-bit compute and claiming up to 3.6 times lower end-to-end latency than Nvidia's GB300. The chip went from architecture concept to first silicon in under 20 months, a timeline OpenAI credits partly to its own LLMs.

Builders of adversarial clothing are now selling garments designed to confuse computer vision, but the protection is fragile: patterns are tuned to specific models and can be defeated by retraining or by other identifying signals.

Andon Labs, a San Francisco AI safety company, runs real businesses managed by AI agents—including a San Francisco store and a Stockholm café—as testbeds to measure how much real-world responsibility current agents can handle. Cofounder Lukas Petersson calls the single-store experiments "weak science" and primarily useful for discovering failure modes.

Atlas Building Composites, an MIT spinout, has developed an AI-powered robotic platform that turns single-use plastic into durable building materials without water. Its recycled composite trusses were used by the U.S. Army Corps of Engineers for a 40-foot bridge in a Massachusetts wetland.

MIT researchers developed HardFlow, a deployment-time algorithm that lets pretrained generative models satisfy hard safety, physical, or task-specific constraints in their final output while still producing high-quality solutions, without retraining. The research appears in the IEEE Transactions on Pattern Analysis and Machine Intelligence.

LangChain announced Connections, a feature in Managed Deep Agents v0.7.0 and later that stores credentials in a LangSmith workspace instead of .env files and supports per-caller identity via user-owned OAuth grants. Connections can be created with the mda CLI and read at run time through connections.get().

Credit Genie adopted LangChain's open-source OpenWiki to automatically generate and update repository documentation, aggregating it into a searchable GitHub Pages portal. The team reports reduced tribal knowledge, faster context access, and better onboarding, and plans to connect OpenWiki with its internal cross-repository knowledge graph.

LangChain detailed how it built a GTM agent on Deep Agents that drafts personalized sales emails for rep approval. The company reports a 250% lead-to-qualified-opportunity conversion increase from December 2025 to March 2026 and 40 hours per month saved per rep.

The Federal Laboratory Consortium selected AI-GUIDE, a portable ultrasound-based medical device developed by MIT Lincoln Laboratory and Massachusetts General Hospital, for its 2026 Excellence in Technology Transfer Award. The prototype is being transferred to the startup AutonomUS Medical Technologies, Inc.

Ben Tossell launched Design Words, a work-in-progress site at bensbites.com/design-words that lets people pick UI styles and components and get text they can paste into a coding agent to reproduce that look. He built it over two days using several AI models.

Nathan Lambert has published an updated reading list on open models, dated 26 Oct. 2026, grouping essays, papers and talks into foundation topics, US-China competition, technical details and cyber risk. It cites a roughly 4-6 month gap between leading open and closed models, with Chinese labs holding the open frontier since about 2024.

Together AI has expanded its Together Fine-Tuning service with support for new open-weight models, live experiment tracking, Expert LoRA adapters, early stopping, dataset previews, pre-flight validation, and training price cuts of 30% to 70% on selected models.

NVIDIA says its NIM 2.0.12 optimized serving stack delivers up to 2.5x higher system throughput on a 4xB200 system versus a baseline without NIM optimizations, reaching 1,997 tokens per second at a 50 TPS per user target for Nemotron 3 Ultra.