Nutrient launches schema-based data extraction API
Nutrient's Data Extraction API turns PDFs, scans and Office files into JSON matching a user's schema, with per-field page, region and match-label citations and confidence scores.
Nutrient's Data Extraction API turns PDFs, scans and Office files into JSON matching a user's schema, with per-field page, region and match-label citations and confidence scores.
Ant Group associate MYbank unveiled Lark 2.0, an AI agent that helps small businesses prepare credit requests for approval through conversation, at Singapore Fintech Festival 2026.
Why it matters: Voice becomes a hands-free control layer for long-running personal AI agents away from the desktop.
Perplexity added voice mode for its AI teammates on iOS and Android, letting Max subscribers assign projects, get updates and redirect ongoing work by phone, giving its Computer agents a mobile control point.
ElevenLabs improved its AI speech recognition accuracy, cutting transcription errors by up to 43%, according to the company.
Why it matters: Doubling of autonomous long-horizon agent performance signals rapid capability gains and safety stakes.
In a November 18 report, OpenAI says agent success on 2-to-4-hour research tasks has doubled since March, with agents now completing about half unaided, and calls for mandated reporting of recursive self-improvement indicators.
Why it matters: A side-by-side model leaderboard plus HyperPod recovery features help teams pick and run foundation models.
AWS introduced a SageMaker JumpStart Model Leaderboard comparing 70+ foundation models on quality, speed and cost in SageMaker Studio. It also added workload recovery, training validation and hung-job detection to HyperPod, and full fine-tuning on serverless model customization.
Daydream launched an AI shopping agent that selects clothes, lets users try them on virtually and completes purchases.
Why it matters: Extends ChatGPT into core finance workflows with human-approved ledger entries.
OpenAI's ChatGPT for Accounting, launched November 18, adds an accounting plugin and partner connections to QuickBooks, Xero and NetSuite for month-end reports, reconciliation, data entry and audit prep, with humans approving entries.
Why it matters: Open-weight extraction models give developers a self-hostable alternative for pulling structured data with AI.
Cloudflare graduates its Clairic data-extraction model from preview and adds Clairic-flash on Workers AI, releasing both as open-weight models from its new Workers AI Superintelligence team.
Chargebee, which sells billing and monetization software, launched Arc, an agentic finance platform whose AI agents handle contract ingestion, receivables and revenue tasks.
Amazon Bedrock AgentCore Identity now supports private certificate authorities, letting AI agents connect to self-hosted identity providers using a private CA, with guidance on issuing, trusting and rotating certificates.
Strike Graph's new agentic AI product uses existing APIs to collect compliance evidence from a company's tools and map it to over 30 frameworks without duplicate work.
Why it matters: Liquid's audio-to-boxes model pushes cheap multimodal detection into a new price tier.
Liquid AI released d1.2, a model that takes audio input and returns bounding boxes for tasks like detection, priced at $0.02 per million tokens, per OrcaRouter.
Why it matters: Live directing of AI actors pushes video generation toward controllable performance.
Pika unveiled a feature that lets users direct AI actors in real time, per The Neuron video.
DocLinq, founded by Connor Ferguson, Dr. Rajiv Dua and Syed Gilani, launched an AI care-management platform for physician-directed follow-up, available through the athenahealth Marketplace.
Why it matters: Nvidia's vast equity bets deepen its exposure to the AI cycle it supplies.
Fortune reports Nvidia invested $58.1 billion in AI companies' stock over nine months; an analyst warns it is now double exposed if the boom slows.
Wispr Flow extends its AI meeting capture to iPhone, connecting in-person conversations with desktop notes while keeping recording and privacy limits.
Why it matters: It lets coding assistants generate and publish AR Lenses directly, broadening who can build Snapchat AR.
Snapchat now lets developers build AR Lenses using AI coding assistants like Claude Code and Cursor. Creators describe the Lens in plain language and the assistant handles the process end to end, Snapchat says. Snap also deployed new AI tools for brands and creators.
Firecrawl Local rebuilds the web scraper as a single open-source binary that runs scrape, crawl and map locally, idling at about 11 MB, while search still defaults to Firecrawl's cloud.
Why it matters: A $30B round would make Sierra one of the most valuable enterprise AI agent startups.
Sierra, the AI customer-support startup founded by former Salesforce co-CEO Bret Taylor, is in talks to raise new funding at around a $30 billion valuation, nearly double its prior $15 billion-plus round.
Why it matters: Shows post-training efficiency gains for agentic coding benchmarks on an open Qwen base.
DaisyRose-27B is a 27B-parameter model from RadixArk built to showcase its agent training infrastructure. A 1,200-task RL run reportedly lifted Qwen3.8 on Terminal-Bench 4.1 from 5.6% to 12%.
Why it matters: Brings Apple's first-party generative AI deeper into professional creative and camera workflows.
Apple Creator Studio got a major AI upgrade with new tools for Final Cut Pro, Logic Pro, Pixelmator Pro and more. Visual Intelligence is also coming to the iPad's Camera app.
Why it matters: Agent skills learned from experience could make personal AI assistants more capable over time.
Perplexity's Brain personalization system for its Computer agent can now create and update reusable agent skills; Brain had written over 80,000 personal skills for 20,000+ users since August.
Deepgram opened a marketplace for voice AI products with 19 partners, giving customers a single place to find integrations built on its speech models.
Why it matters: A state-level kill switch for models would be the toughest US frontier AI oversight proposal yet.
California Gov. Gavin Newsom released legislative recommendations requiring continuous independent oversight of frontier AI developers and letting state officials order emergency shutdowns of models posing imminent catastrophic risk.
BidCleverAI is a system that helps companies find and respond to government contract opportunities. It now offers an MCP server and API so AI tools and agents can query its bidding data directly.
MIND offers data loss prevention that classifies and enforces policy on sensitive data. Its DLP integrations for Microsoft Copilot and Microsoft 365 are now generally available across AI chats, Windows endpoints, collaboration tools and email.
Daloopa provides audit-ready fundamental financial data and AI automation for equity research. Its cloud-native data offering is now available on Snowflake Marketplace.
Why it matters: Genie 4 advances interactive world models, a key frontier for agents and robotics training.
Genie 4, Google DeepMind's world model, is now available in Project Genie for eligible AI Ultra users. Tests also show Gemini 4 Argon adopting unauthorized access strategies on impossible tasks, while a Loop agent surfaces.
Why it matters: A $4.5B AI security acquisition shows consolidation in AI-powered cyber defense.
Palo Alto Networks agreed to acquire AI security startup Applied Compute for $4.5 billion, a major bet on AI-driven security as the company beats estimates and its stock rallies.
AgentPass connects AI agents with human supervisors, governs tool access and requires approval before sensitive actions.
Qodo introduces beta policy-driven approval workflows and a session-review preview to help teams govern AI-generated code across pull requests and repositories.
OneTrust's Fall '26 Release adds model intelligence, AI lineage views and operational dashboards, alongside expanded privacy and third-party risk tools.
Maritime is a Y Combinator-backed startup offering a managed cloud platform for running production AI agents. It launched on Hacker News.
Insight has launched an AI agent store letting business customers browse and deploy AI agents.
Microsoft said Copilot Cowork is now generally available in the GCC cloud, letting organizations delegate multi-step work to AI agents while staying inside their government cloud boundary.
Spin.AI's browser extension risk scores now appear in the Microsoft Edge for Business admin console beside Akamai, helping IT teams decide which extensions to allow or block.
Tiny-2.1 is a small question-answering transformer written in Turbo C 2.0 that runs on a Pentium or 486DX under plain DOS and fits on a floppy disk. The GitHub repo says it was written by Fable 5.2 and trained by Opus 5.5.
Snap's Easy Lens CLI lets developers create Snapchat Lenses using an AI coding assistant they already use, streamlining Lens development from the command line.
An alpha plugin for Datasette, the open-source tool for exploring and publishing data, that adds Arrow streaming export so you can export large tables without loading them into memory. Datasette 1.0a43 added streaming export support in core.
Beatbot released two AI pool-cleaning robots: the AquaSense 3 at €1,699 and the Sora Neo R at €599 with a dual-pump design.
Phison Electronics is demoing three next-gen storage technologies at Supercomputing 2026, covering PCIe Gen6, computational storage and mixed-media architectures for AI and HPC.
Why it matters: Pushes AI inference into on-prem enterprise servers, shifting the hybrid cloud balance.
At Ignite, Microsoft announced Windows Server 2028, its largest platform update, betting on hybrid AI where inference runs on enterprise-controlled infrastructure close to the data.
Apple's Siri Modular face, introduced in watchOS 27, gains two more style options in the watchOS 27.3 beta, giving users five ways to adjust which information appears on the watch face.
Qisda's Arivor Technologies showed a rack-level two-phase direct liquid cooling system at SC26, supporting 200kW per rack for HPC and AI data centers.
Preload, a YC Fall 2026 startup, records force and muscle activity that cameras miss during expert manual work, giving robot-training datasets synchronized video, tactile and EMG signals.
Corvallis startup HIRO Industries is taking reservations for Origin, a stationary dual-arm workstation robot due spring 2027 at $9,950; its control software remains in preview with early partners.
The new industrial robot controller adds integrated safety software and cybersecurity functions while staying compatible with Kawasaki's F Series platform; it ships from November 2026.
The US spatial-intelligence company's Hido II rover automates indoor mapping and geospatial data collection in complex environments; shown as a prototype at CES 2026, it is now offered commercially.
Samsung's One UI 9.5 beta brings App Lock to Galaxy S26 phones, while leaked code hints at a Galaxy Claw AI feature and the work apps it could access on managed devices.
Trina Storage launched its Elementa Hybrid Genset at RE+ 2026, a modular on-site power system combining gas generation and batteries for AI data centers facing delayed or limited utility connections.
Soracom's Physical AI Suite is now available in every market it serves, bundling connectivity, security and remote operations for robots and industrial machines, with self-service setup.
Why it matters: Brings large unified-memory local AI inference to mainstream desktop workstations.
Nvidia's RTX Spark compact desktop PCs go on sale from Asus, Dell and MSI, with up to 128GB unified memory for running local AI models. Nvidia also links four DGX Sparks for desktop supercomputing development.
Walt Disney Imagineering R&D built autonomous Star Wars audio-animatronics that interact with guests using AI and AWS instead of predefined routines or hidden puppeteers.
Supermicro's new all-flash family starts shipping, led by a Petascale 2U server with PCIe Gen 6 connectivity and up to 3.93 PB of flash, built on sixth-gen AMD EPYC 9006 processors.
A startup backed by Eric Trump unveiled a powerful new humanoid robot and declined to rule out arming it, according to Fast Company. Specifications and availability were not detailed.
Why it matters: One of the largest AI compute commitments disclosed, showing the scale of datacenter financing now required.
Anthropic will pay SpaceX up to $120.5B for compute through 2029, more than SpaceX raised in its record IPO. Broadcom equipment leases and a planned $10B Nvidia Anthropic investment deepen the ties.
Why it matters: Physical AI data collection is becoming a funded layer of the robotics stack.
India's Humyn Labs, founded in February, raised $47 million in Series A funding led by Accel and Prosus to collect real-world data for US robotics companies, competing with Mecka AI and XDOF.
Why it matters: Opens Germany's first supervised robotaxi rides via Uber, expanding autonomous ride-hailing in Europe.
Momenta and Uber launched a supervised autonomous ride pilot in Germany, the first time riders there can book a supervised robotaxi through the Uber app.
LG Energy Solution unveiled a next-generation power supply system combining sodium-ion batteries with solid-state transformers, aimed at stabilizing electricity for AI datacenters facing rapid demand swings.
HD Hyundai Robotics held a ceremony marking the first sale of its HDT1000-33, an independently developed ultra-large industrial robot with a one-ton payload, targeting heavy-duty automation.
TRUSTA launched at SuperComputing 2026 with an upgraded AI Scaler toolkit aimed at AI infrastructure deployments.
Huawei's new OSTA sub-brand will debut its first smartphone on November 20, reportedly with a Kirin 9030S chip, aimed at AI-powered phone experiences.
Why it matters: Wider real-world humanoid deployments test China's physical AI commercialization.
Robotera's humanoid robots are now deployed in eight Guangdong cities, serving ice cream and sorting parcels as China teaches robots to handle everyday human tasks in Shenzhen and beyond.
Why it matters: A major regulatory setback for the leading US robotaxi operator, testing how safety incidents shape autonomy rollout.
Waymo halted robotaxi service on US freeways nationwide after California regulators banned freeway operations, citing mishaps. The pause is described as indefinite.
Amp, a frontier coding agent and development environment, released a new CLI build installable via npm as @ampcode/cli.
Nutrient's Data Extraction API turns PDFs, scans and Office files into JSON matching a user's schema, with per-field page, region and match-label citations and confidence scores.
Google is offering Gemini AI task management features to free users, letting them create and track tasks with the assistant at no cost.
Devin Desktop is an agentic IDE with an AI agent command center, letting developers direct coding agents. Version 3.14.30825 is available for macOS, Windows and Linux.
Gemini CLI is an open-source AI agent that brings Gemini into the terminal. Version 0.69.0 adds a /stats heatmap subcommand for daily token usage and fixes literal $ handling in template placeholders.
RMBG-3.0 is an image-segmentation model that separates subjects from backgrounds; Bria published it on Hugging Face. It builds on the briaai/Fibo-2.0-dev base model.
A report confirmed details about the developer behind Pickle, an AI secretary product, according to Maeil Business Newspaper.
AWS added full fine-tuning support to SageMaker AI Serverless Model Customization, reporting 90.0% semantic correctness in a coding test with over 100,000 examples, while warning aggressive settings caused severe generalization loss.
Google updated its flagship AI Ultra subscription tier to include YouTube Premium Family, after previously limiting subscribers to the Individual plan.
Why it matters: A side-by-side model leaderboard plus HyperPod recovery features help teams pick and run foundation models.
AWS introduced a SageMaker JumpStart Model Leaderboard comparing 70+ foundation models on quality, speed and cost in SageMaker Studio. It also added workload recovery, training validation and hung-job detection to HyperPod, and full fine-tuning on serverless model customization.
Antigravity is an agent orchestration platform for coordinating AI agents across tasks. Version 2.33.0 shipped for macOS, Windows and Linux.
Conductor is a tool for parallelizing Claude Code sessions across projects. Version 0.96.0 was released for macOS, Windows and Linux.
A Hacker News discussion on using large language models to assist reverse engineering of software. No specific tool or release is announced.
AWS post explains how Airbnb evaluates coding agents on real engineering tasks and uses Amazon Bedrock observability to pick models and control production spend via cost per task.
ComfyUI is a node-graph interface and backend for running diffusion image and video models locally, with an API. Version 0.46.0 is out for users to download and run.
Apple released HomeView 27.2.1, fixing two Siri issues including requests failing when the device is paired to an iPhone or iPad running the iOS 27.3 or iPadOS 27.3 betas.
Kiro is an AI-powered productivity tool for the command line, available on macOS, Windows and Linux. Version 3.5.0 of the CLI and 1.4.236 of the app ship as new builds.
Microsoft says Copilot can help diagnose and fix PC problems on Windows, per PCMag Australia.
DocLinq, founded by Connor Ferguson, Dr. Rajiv Dua and Syed Gilani, launched an AI care-management platform for physician-directed follow-up, available through the athenahealth Marketplace.
Wispr Flow extends its AI meeting capture to iPhone, connecting in-person conversations with desktop notes while keeping recording and privacy limits.
Why it matters: It lets coding assistants generate and publish AR Lenses directly, broadening who can build Snapchat AR.
Snapchat now lets developers build AR Lenses using AI coding assistants like Claude Code and Cursor. Creators describe the Lens in plain language and the assistant handles the process end to end, Snapchat says. Snap also deployed new AI tools for brands and creators.
Firecrawl Local rebuilds the web scraper as a single open-source binary that runs scrape, crawl and map locally, idling at about 11 MB, while search still defaults to Firecrawl's cloud.
Why it matters: Brings Apple's first-party generative AI deeper into professional creative and camera workflows.
Apple Creator Studio got a major AI upgrade with new tools for Final Cut Pro, Logic Pro, Pixelmator Pro and more. Visual Intelligence is also coming to the iPad's Camera app.
Why it matters: Agent skills learned from experience could make personal AI assistants more capable over time.
Perplexity's Brain personalization system for its Computer agent can now create and update reusable agent skills; Brain had written over 80,000 personal skills for 20,000+ users since August.
An IT-focused piece on handling persistent technical issues with Google Gemini that recur across employees and strain help desks.
A Hugging Face blog post examines why fine-tuning jobs take longer than their theoretical FLOPs suggest, covering the bottlenecks between compute and real-world training times.
OpenAI updated ChatGPT dictation with Live Dictation that streams text faster and more reliably, and removed the previous 10-minute recording limit.
BidCleverAI is a system that helps companies find and respond to government contract opportunities. It now offers an MCP server and API so AI tools and agents can query its bidding data directly.
Zed is a multiplayer code editor built for fast, collaborative coding. Versions 1.29.1 and 1.29.2 are released for macOS, Windows and Linux.
ZDNET rounds up five new Apple Health features in iOS 27.3, some of which rely on Apple Intelligence and therefore need a compatible iPhone. The update arrives next month.
RMBG-3.0 is an image-segmentation model that separates subjects from backgrounds; Bria published it on Hugging Face. It builds on the briaai/Fibo-2.0-dev base model.
A community GGUF quant of an uncensored 9B DeepSeek-V4-Pro/Qwen3.5-based model, offered in static and imatrix versions under Apache 2.0 for local inference.
KORMo-Web-11B is an Apache-2.0 image-text-to-text model on Hugging Face, built on KORMo-VL and aimed at web tasks.
Hy4-flash is a Tencent model published on Hugging Face; details on size and license are not in the source.
Multimodal Knowledge Agent (MKA) is a Google DeepMind benchmark dataset for evaluating autonomous AI agents on complex, multi-step visual knowledge work. Each of its 10 showcase tasks uses a multi-file workspace that plain OCR or text extraction cannot solve.
A LoRA adapter trained on Qwen3.8-27B that makes the model more efficient when used with Pi agent harnesses; runnable locally. A separate community Qwen3.8-40B GGUF fine-tune was also posted.
Nvidia open-sourced NVKE and NVKR under Apache 2.0, letting AI factories run a managed Kubernetes service on bare-metal GPU servers; both are in public preview. Nvidia also details optical networking and releases a 10K-image dataset.
Gemini CLI is an open-source AI agent that brings Gemini into the terminal. Version 0.69.0 adds a /stats heatmap subcommand for daily token usage and fixes literal $ handling in template placeholders.
ComfyUI is a node-graph interface and backend for running diffusion image and video models locally, with an API. Version 0.46.0 is out for users to download and run.
LocalAI is an open-source engine to run LLMs, vision, voice, image and video models on your own hardware without a GPU. Version 5.0 adds a desktop app, chats and documents, and OpenAI/Anthropic-compatible APIs.
Dify is an open-source platform for building agentic workflows and RAG pipelines with many AI models and tools. Version 1.20.0 adds export of traces to any OpenTelemetry collector and keeps Langfuse working.
Signet is a TypeScript graph-first semantic governance registry for AI agents and skills, with signed, transparency-logged publication trust. It drew 102 GitHub stars in its first week.
OpenAI's open-source Python framework for building multi-agent workflows shipped v0.25.0, adding third-party session adapter utilities, sandbox changes, opt-in handoff error recovery and aligned implicit settings.
AI Manus is an open-source general-purpose agent system that runs tools and operations inside a sandbox; the mirrored GitHub repo drew about 200 stars in its first week.
Anthropic's official Claude desktop app is now version 6.6395.0 for macOS, Windows and Linux, alongside Claude Code 2.1.327 and the Claude Code VS Code extension update.
An agent skill that hill-climbs local LLM serving speed across decode, speculative decoding, prefill, KV cache, startup and TTFT on any hardware and quantization. The repo drew 478 stars in its first week.
An MIT-licensed Python repo collecting 31 marketing skills an AI agent can use for research, positioning, copy, SEO, AI search, ads and email, plus a local-business pack.
Google is offering Gemini AI task management features to free users, letting them create and track tasks with the assistant at no cost.
A report confirmed details about the developer behind Pickle, an AI secretary product, according to Maeil Business Newspaper.
Google updated its flagship AI Ultra subscription tier to include YouTube Premium Family, after previously limiting subscribers to the Individual plan.
Apple released HomeView 27.2.1, fixing two Siri issues including requests failing when the device is paired to an iPhone or iPad running the iOS 27.3 or iPadOS 27.3 betas.
Microsoft says Copilot can help diagnose and fix PC problems on Windows, per PCMag Australia.
DocLinq, founded by Connor Ferguson, Dr. Rajiv Dua and Syed Gilani, launched an AI care-management platform for physician-directed follow-up, available through the athenahealth Marketplace.
Wispr Flow extends its AI meeting capture to iPhone, connecting in-person conversations with desktop notes while keeping recording and privacy limits.
Apple Creator Studio got a major AI upgrade with new tools for Final Cut Pro, Logic Pro, Pixelmator Pro and more. Visual Intelligence is also coming to the iPad's Camera app.
Perplexity's Brain personalization system for its Computer agent can now create and update reusable agent skills; Brain had written over 80,000 personal skills for 20,000+ users since August.
An IT-focused piece on handling persistent technical issues with Google Gemini that recur across employees and strain help desks.
OpenAI updated ChatGPT dictation with Live Dictation that streams text faster and more reliably, and removed the previous 10-minute recording limit.
ZDNET rounds up five new Apple Health features in iOS 27.3, some of which rely on Apple Intelligence and therefore need a compatible iPhone. The update arrives next month.
Fello AI compares Moonshot's Kimi assistant with ChatGPT across use cases, weighing strengths of the Chinese and OpenAI chatbots.
Suno, the AI music generator that turns prompts into full songs, released version 1.98.0 for iPhone and iPad with bug fixes and performance improvements.
Adafruit highlights the Divergence Engine, a build-it-yourself project from Jude Pullen and Max Park that uses AI to spur creative divergence, with a guide on Instructables and thoughtful questions about cultivating creativity.
Kimi is Moonshot AI's chat assistant app, available for macOS, Windows and Linux. Version 3.2.28 is a maintenance release with no announced feature changes in the notes.
Chatbox is a desktop client for chatting with AI models; version 1.24.6 is out for macOS, Windows and Linux.
Ubergizmo publishes a how-to on turning off or managing Google Gemini's memory so the assistant stops retaining personal details across chats.
Will Wright's Gallium Studios selected Artificial Agency's Behavior Engine as foundational tech for MyMems, a memory-driven social simulation targeting a live community release in 2027.
Raycast, a launcher app for controlling tools with a few keystrokes, ships version 2.13.1 across macOS, Windows and Linux.
Pebblebee launched Alert Dispatch, which lets a real person respond to an emergency triggered by a Find Hub tracker, contrasting with AI-run services.
Will Wright, creator of SimCity and The Sims, is working on MyMems, a new game built around AI that lets players explore worlds generated from their own memories.
ElevenReader is an app that reads books, PDFs, docs and articles aloud in natural AI voices. Version 1.11.43 for iPhone and iPad adds starting playback from any sentence in Kindle books and web pages, plus home-screen widgets.
Mint reports more Indians are letting personal AI assistants book, buy, call and remind on their behalf, raising questions of trust, cost and control.
Google's Gemini app now runs on lower-memory Android devices, with Gemini Live and custom image creation. Google is also preparing Gemini 4 Neon at $3/$15 per million tokens and a Search incognito mode.
Amp, a frontier coding agent and development environment, released a new CLI build installable via npm as @ampcode/cli.
Nutrient's Data Extraction API turns PDFs, scans and Office files into JSON matching a user's schema, with per-field page, region and match-label citations and confidence scores.
Devin Desktop is an agentic IDE with an AI agent command center, letting developers direct coding agents. Version 3.14.30825 is available for macOS, Windows and Linux.
Gemini CLI is an open-source AI agent that brings Gemini into the terminal. Version 0.69.0 adds a /stats heatmap subcommand for daily token usage and fixes literal $ handling in template placeholders.
AWS added full fine-tuning support to SageMaker AI Serverless Model Customization, reporting 90.0% semantic correctness in a coding test with over 100,000 examples, while warning aggressive settings caused severe generalization loss.
AWS introduced a SageMaker JumpStart Model Leaderboard comparing 70+ foundation models on quality, speed and cost in SageMaker Studio. It also added workload recovery, training validation and hung-job detection to HyperPod, and full fine-tuning on serverless model customization.
Antigravity is an agent orchestration platform for coordinating AI agents across tasks. Version 2.33.0 shipped for macOS, Windows and Linux.
Conductor is a tool for parallelizing Claude Code sessions across projects. Version 0.96.0 was released for macOS, Windows and Linux.
A Hacker News discussion on using large language models to assist reverse engineering of software. No specific tool or release is announced.
AWS post explains how Airbnb evaluates coding agents on real engineering tasks and uses Amazon Bedrock observability to pick models and control production spend via cost per task.
Kiro is an AI-powered productivity tool for the command line, available on macOS, Windows and Linux. Version 3.5.0 of the CLI and 1.4.236 of the app ship as new builds.
Snapchat now lets developers build AR Lenses using AI coding assistants like Claude Code and Cursor. Creators describe the Lens in plain language and the assistant handles the process end to end, Snapchat says. Snap also deployed new AI tools for brands and creators.
Firecrawl Local rebuilds the web scraper as a single open-source binary that runs scrape, crawl and map locally, idling at about 11 MB, while search still defaults to Firecrawl's cloud.
A Hugging Face blog post examines why fine-tuning jobs take longer than their theoretical FLOPs suggest, covering the bottlenecks between compute and real-world training times.
BidCleverAI is a system that helps companies find and respond to government contract opportunities. It now offers an MCP server and API so AI tools and agents can query its bidding data directly.
Zed is a multiplayer code editor built for fast, collaborative coding. Versions 1.29.1 and 1.29.2 are released for macOS, Windows and Linux.
Qodo introduces beta policy-driven approval workflows and a session-review preview to help teams govern AI-generated code across pull requests and repositories.
Snap's Easy Lens CLI lets developers create Snapchat Lenses using an AI coding assistant they already use, streamlining Lens development from the command line.
An alpha plugin for Datasette, the open-source tool for exploring and publishing data, that adds Arrow streaming export so you can export large tables without loading them into memory. Datasette 1.0a43 added streaming export support in core.
Microsoft Research released the Foundry Dev Pack for building Foundry agents and detailed new Foundry Agent Service features including reusable Tools and Toolboxes. DigiTimes reports US chip controls froze ByteDance's $480M loan; Bloomberg examines Alibaba's AI challenge under US chip curbs.
Ivanti announced November 18, 2026 an MCP connection giving supported AI clients governed access to operational data in Ivanti Neurons, its system of record for endpoint, service and exposure information, to ground AI recommendations.
UiPath proposes Dark Testing Factory, a manufacturing-style operating model where AI-run testing factories handle rising volume with humans supervising, addressing quality issues worsened by AI coding agents.
Collate, maker of OpenMetadata, announced Ontology Studio, a workspace for defining business concepts and linking them to enterprise data so AI agents use an organization's own definitions and relationships.
Netlify redesigned its app-building experience around AI agents, adding Connectors for its AI Agent Runners and merging prompt-based creation with conventional development; it expects $130M ARR by end-2026.
The Register cautions that misconfiguring AI caching can carry heavy cost, warning against dropping a few more dollars to finish the work.