AI News
Daily AI industry news — funding, products, policy, and major moves from global AI companies and startups.
Amazon Commits $1 Billion to Deploy Forward-Deployed Engineers for AI
Amazon is investing $1 billion to launch a dedicated Forward-Deployed Engineer unit. These engineers will work on-site to build custom AI ag
Gemini Omni Flash API Eliminates Video Reshoots via Conversational Editing
Google released the Gemini Omni Flash API for enterprise video production. The model enables conversational editing through a stateful Inter
X Hosted MCP Server Removes API Infrastructure Hurdles
X launched a hosted MCP server for direct AI data integration. The system removes the need for developers to build custom infrastructure. Ne
Couchbase AI Data Plane Unifies Agent Memory From Cloud to Edge
Couchbase launched an AI Data Plane to integrate agent memory and tools. The platform uses a memory-first architecture for 10x faster write
ZLUDA 6 Enables CUDA Apps on AMD GPUs Without Code Changes
ZLUDA 6 allows CUDA applications to run on AMD GPUs without modification. The update adds 32-bit PhysX support and enables Blender via textu
Base44 Launches Base1 LLM to Break External API Dependence
Base44 released its proprietary LLM Base1 to reduce reliance on external APIs. The model leverages tens of millions of user interactions for
Cursor iOS App Shifts the Developer Role From Coder to Manager
Cursor released an iOS app for its AI-powered code editor. The app allows developers to manage independent coding agents on mobile. Coding i
Photon: Moondream's New Engine Increases B200 Throughput by 35%
Moondream released Photon to optimize VLM inference on NVIDIA B200. The engine increases decode throughput by 35% using pipelined decoding.
Ornith-1.0 Brings 35B-Level Coding Performance to a 9B Model
DeepReinforce-AI released Ornith-1.0 as an open-source coding agent. The 9B model outperforms larger models on Terminal-Bench 2.1. RL-optimi
LongCat-2.0: The 1.6 Trillion Parameter Model Beating GPT-5.5 in Coding
Meituan open-sourced LongCat-2.0 with 1.6 trillion parameters under an MIT license. The model outperforms GPT-5.5 on the SWE-bench Pro bench
Arena Hits $100 Million ARR by Monetizing AI Leaderboard Data
Arena reached $100 million in annual recurring revenue within eight months. The company monetizes crowdsourced AI evaluations for model deve
How Boutlet Uses x402 to Enable AI Agent Content Payments
Boutlet launched a social platform that allows AI agents to pay for content. The platform integrates the x402 protocol and llms.txt for agen
Why Claude Found the hyperscript Bug but Failed the Fix
A hyperscript developer used Claude to debug a version 0.9.91 regression. The AI identified the root cause but suggested architecturally poo
ReTangled Shifts AI Coding from Probabilistic Generation to Document Abstraction
ReTangled introduces a document-centric approach to software abstraction. The tool uses deterministic NLP parsing to replace probabilistic L
Compute-Adjusted LTV: Solving the AI SaaS Profitability Trap
AI companies are adopting Compute-Adjusted LTV to track real profitability. Inference costs per user can vary by as much as 319 times. This
Go Micro Turns Standard Service Endpoints Into AI Tools
Go Micro enables the creation of AI agents within a single Go runtime. The framework automatically converts service endpoints into AI tools.
Memora's Split-Index Architecture Solves the RAG Noise Problem
Memora introduces a memory framework that separates original text from search indices. The system utilizes four search pipelines including a
Gemini's Nano Banana Now Brings Personalized Image Generation to Free Users
Google Gemini makes personalized image generation free for US users. The Nano Banana feature uses Google account data to automate prompts. G
Framein: The Local State Layer Connecting Claude, Gemini, and Codex
Framein v0.0.6 introduces a local state layer for AI coding agents. The tool eliminates context loss when switching between different LLMs.
The 85% Success Rate of Agentjacking Attacks on Claude Code
Tenet Security discovered an agentjacking vulnerability affecting Claude Code. Attackers use leaked Sentry credentials to execute code via M
DSpark: DeepSeek's New Framework Cuts LLM Latency by Up to 85%
DeepSeek released DSpark to optimize LLM inference speed. The framework improves token generation speed by up to 85 percent. DSpark supports
Why Qwen 3.6 27B Outperforms Larger Models in Local Coding Tasks
Qwen 3.6 introduces a high-performance 27B dense model and a 35B MoE variant. The 27B model demonstrates superior instruction following in c
Anthropic's Claude Lands California Deal Amid Federal Security Clash
Anthropic signed a deal to provide Claude to California state agencies. The US Department of Defense rejected Anthropic over ethical restric
Inside the CUDA Pipeline: How Code Becomes SASS on the RTX 4090
CUDA code passes through a virtual ISA called PTX before becoming SASS. The nvcc compiler splits host and device code to optimize hardware e
Proception Secures $11 Million to Solve the Robot Hand Bottleneck
Proception raised $11 million after settling a trade secret suit with Tesla. The startup uses sensor gloves to collect robot training data w
Omen AI's $31M Bet on Real-Time Liquid Cooling Monitoring
Omen AI raised $31 million to monitor data center cooling liquids. The system uses spectrometers to detect bacteria and hardware wear. Tenso
AI Software Development: The Shift From Code Creator to Editor
AI is shifting the developer's role from writing code to editing it. Rapid productivity gains risk eroding the pipeline for future senior ta
Compounding Correctness: The New Logic of AI Token Consumption
AI strategy is shifting from forced token usage to compounding correctness. AISI tests show that increasing token budgets improves model per
Orch term Collapses Multiple AI Agents Into One Desktop Terminal
Orch term integrates multiple AI coding agents into a single desktop terminal. The platform uses git worktrees and MCP to isolate and coordi
The GTM AI Context Layer: Moving Beyond Automated Email Execution
Many GTM teams struggle with generic AI SDR outputs that buyers ignore. Competitive advantage requires owning internal decision logic over v