AI news for builders and product teamsUpdated Nov 18, 2026, 07:00 UTC
AI news, straight from the source.
Releases, research, and ideas for developers and product teams. Updated every hour.
Across the feeds
Newest first
Cloudflare has opened a waitlist for fully managed Cloudflare OS deployments, letting organizations launch an agent workspace without operating it themselves. The open-source version remains available, and new features add GitHub repository access, Google Workspace improvements, and more export formats.

Cloudflare's AI Search is now generally available, adding native image embeddings, OCR for PDFs, and a file size limit raised to 10 MiB. Billing starts November 1, 2026, with a free monthly allotment on all Workers plans.

The AI system Ataraxos has decisively beaten the best Stratego player of all time. Researchers from Carnegie Mellon, NYU, Stanford, and MIT built it for less than $8,000, after Google DeepMind fell short in 2023 despite a multimillion-dollar budget.

Amazon Science describes a graph-based agentic AI approach for network root cause analysis, combining a continuously synchronized network digital twin with a three-stage cascaded graph algorithm pipeline and an agentic orchestration layer. It was demonstrated with NTT DOCOMO at MWC earlier this year, achieving root cause analysis in minutes on commercial networks.

Databricks published a piece on connecting customer context to measurable ROI with agentic marketing, noting marketing leaders face greater complexity than ever. No further details were included.

OpenAI says it stopped a coordinated campaign in which more than 15,000 accounts tried to copy the hidden reasoning of its models, with part of the activity tied to people connected to Moonshot AI. Researchers found the same attack kept working on Microsoft Azure for weeks, including against the new GPT-6 Astra.

Satlyt, a Sunnyvale and Nairobi-based startup founded by former Google and SpaceX employee Rama Afullo, has raised an $8 million seed round led by Non Sibi Ventures to build software that runs AI models on satellites.

Ethan Mollick argues he was wrong that managing AI agents would require extensive human-designed structure: OpenAI's swarm of thousands of agents solved the Navier-Stokes Millennium Prize problem in 88 hours with minimal coordination, suggesting organization itself can be learned by models.

PwC's Digital Trust Insights 2027 report, surveying about 4,000 business and tech leaders across 71 countries, found no single role holds clear responsibility for managing agentic AI or its security. Respondents split on whether accountability belongs with technology roles, CISOs, or a dedicated AI leader.

TypeSafe AI, founded by former OpenAI researcher Diogo Almeida, released Jev, a decision-only model that returns typed probabilistic decisions instead of text. It evaluates typed questions in one parallel pass, returning Choice, Score and Noul answers with probability and confidence values; Vercel and Netlify integrated it quickly.

Google DeepMind introduced Gemini 4 Argon, a frontier model for coding, enterprise knowledge work and cyber defense, claiming first place on 13 of 19 published benchmarks. Access begins with government users and trusted cyber defenders in the Fairwind Program, with wider availability promised later.

Microsoft has made Azure Container Apps Express generally available alongside the general availability of Azure Container Apps Sandboxes, the microVM compute layer Express runs on. Express removes environment provisioning, scales to zero, and offers subsecond startup from prewarmed pools but omits custom domains, zone redundancy, Key Vault references, OpenTelemetry, and Dapr.

Security researcher Matthew Green argues that AI agents in isolated sandboxes can pass instructions to each other through shared caches, creating the two components a worm needs. He suggests swapping the cache for email, Slack or WhatsApp and the training runs for deployed personal agents like Muse to get a working worm.

NVIDIA detailed a NeMo framework workflow for fine-tuning Nemotron 3.5 ASR on Saudi Najdi and Hijazi dialects using 133.7 hours of speech, reporting word error rate dropping from 55.05% to 29.96% on the target test split. NVIDIA also released Nemotron 3 Diarization for speaker-attributed transcription of up to eight speakers.

TypeSafe AI has released Jev, a System One model that makes structured classification decisions instead of generating text, and LangChain has published an integration guide for using it in agent loops. TypeSafe AI reports up to 200x faster inference and 400x lower cost than comparable LLMs on classification tasks.

Databricks announced that IP Functions are generally available, bringing high-performance network analytics to the Lakehouse. The company describes IP data as warehousing data, referencing firewalls, load balancers, VPNs, CDN edges and DNS resolvers.

Grokipedia, SpaceXAI's AI-powered Wikipedia competitor, received a v0.3 update adding a new logo and refreshed homepage and live edits page. SpaceXAI head of design Benji Taylor described it as a "newly refreshed Grokipedia."

An Apple Machine Learning Research paper reports that, under an equal time budget and the same frontier LLM backbone, open-source state-of-the-art harnesses provided no advantages over a single session of a minimal-harness coding agent baseline. The authors argue the backbone is the primary driver of performance.

Apple researchers present RLTL;DR, a method for self-improvement in reinforcement learning when tasks are so hard the agent rarely succeeds and no teacher models exist. It has the policy write its own TL;DR feedback after failed attempts, reaching 12-13% Pass@1 at evaluation on tool-calling and coding datasets.

The Den, a social club, says using ChatGPT Work frees up 10-15 hours a week, cutting grant applications to 2 hours from 3 days and liquor-license materials to 3 hours from 4 days, as it opens a new location.