VentureBeat AI
VentureBeat's coverage of artificial intelligence news and trends
Why Google’s File Search Could Displace DIY RAG Stacks in the Enterprise
Google launched File Search on its Gemini API, a fully managed retrieval‑augmented generation (RAG) service that abstracts storage, chunking, embedding and vector search, letting developers invoke RAG via the existing generateContent endpoint. The tool supports multiple file formats, provides built‑in citations, and is priced at $0.15 per million tokens for indexed embeddings, with free query‑time storage. Powered by Google’s top‑ranking Gemini embedding model, File Search aims to simplify enterprise RAG pipelines that traditionally require stitching together ingestion, embedding, vector databases and orchestration. Early adopters like Phaser Studio report dramatically faster access to relevant code and design assets, turning days‑long prototyping into minutes.
From Prototype to Production: What Vibe Coding Tools Must Fix for Enterprise Adoption
Salesforce introduced Agentforce Vibes, an enterprise‑grade vibe‑coding platform that pairs generative AI with built‑in security, governance and context‑aware tooling. The solution distinguishes "green" UI layers—where rapid AI‑generated code is safe—from "red" business‑logic and data layers that require AI augmentation under...
The Compute Rethink: Scaling AI Where Data Lives, at the Edge
Arm’s latest briefing underscores a rapid shift of artificial‑intelligence workloads from cloud data centers to the edge, where data is generated in devices, sensors and networks. Companies are adopting on‑device AI to cut latency, lower costs and protect privacy,...
Google Cloud Updates Its AI Agent Builder with New Observability Dashboard and Faster Build-and-Deploy Tools
Google Cloud unveiled a major upgrade to its Vertex AI Agent Builder, adding an observability dashboard, one‑click deployment, and state‑of‑the‑art context‑management layers that let enterprises build agents with as few as 100 lines of code. The update expands language support...
From Logs to Insights: The AI Breakthrough Redefining Observability
Elastic unveiled Streams, an AI‑driven observability feature that automatically partitions, parses and structures raw log data, surfacing critical errors, anomalies and suggested remediation steps. The tool aims to make logs the primary signal for incident investigations, cutting the manual effort...
Databricks Research Reveals that Building Better AI Judges Isn't Just a Technical Concern, It's a People Problem
Databricks unveiled its Judge Builder framework, a workshop‑driven system for creating AI judges that evaluate other AI models by aligning stakeholder quality criteria and capturing domain‑expert insight. The tool tackles the "Ouroboros problem" of AI‑evaluating‑AI by measuring distance to human...
Attention ISN'T All You Need?! New Qwen3 Variant Brumby-14B-Base Leverages Power Retention Technique
Manifest AI unveiled Brumby-14B-Base, a 14‑billion‑parameter model derived from Qwen3-14B-Base that replaces all attention layers with a novel Power Retention mechanism. The model was retrained in 60 hours on 32 H100 GPUs for roughly $4,000 and achieves accuracy on par...
98% of Market Researchers Use AI Daily, but 4 in 10 Say It Makes Errors — Revealing a Major Trust...
According to a QuestDIY survey of 219 U.S. market research professionals, 98% now use AI tools and 72% do so daily, with 56% reporting at least five hours saved per week. However, 39% say AI introduces errors, 37% cite new...
Forget Fine-Tuning: SAP’s RPT-1 Brings Ready-to-Use AI for Business Tasks
SAP unveiled RPT-1, its first Relational Foundation Model, a pre‑trained AI that ingests tabular business data such as spreadsheets and relational databases to deliver predictive‑analytics and other enterprise tasks without fine‑tuning. Built on the ConTextTab architecture, the model claims semantic...
Inside Zendesk’s Dual AI Leap: From Reliable Agents to Real-Time Intelligence with GPT-5 and HyperArc
Zendesk is scaling its AI-powered support platform by integrating OpenAI's GPT‑5 and acquiring analytics firm HyperArc. The company reports that its autonomous AI agents now resolve roughly 80% of incoming tickets, with GPT‑5 cutting workflow failures by 30% and reducing...
The Beginning of the End of the Transformer Era? Neuro-Symbolic AI Startup AUI Announces New Funding at $750M Valuation
Augmented Intelligence Inc (AUI), a New York‑based neuro‑symbolic AI startup, closed a $20 million bridge SAFE round at a $750 million valuation cap, bringing its total funding to nearly $60 million and paving the way for a larger raise. The company’s Apollo‑1 model...
Meet Denario, the AI ‘Research Assistant’ that Is Already Getting Its Own Papers Published
An international research team released Denario, an open‑source AI system that autonomously conducts end‑to‑end scientific research, generating publishable papers in about 30 minutes for roughly $4 each. The modular architecture uses specialized agents for idea generation, literature review, methodology design,...
Developers Beware: Google’s Gemma Model Controversy Exposes Model Lifecycle Risks
Google removed its Gemma 3 model from AI Studio after Senator Marsha Blackburn accused it of fabricating defamatory statements about her, while keeping the model accessible via API. Google said the pull was to prevent confusion, noting Gemma was built...
Inside Celosphere 2025: Why There’s No ‘Enterprise AI’ without Process Intelligence
Celonis will host Celosphere 2025, a three‑day event focused on linking process intelligence (PI) with enterprise AI to deliver measurable ROI. The company cites a Forrester study showing 383% ROI over three years and six‑month payback for users of its...
Why IT Leaders Should Pay Attention to Canva’s ‘Imagination Era’ Strategy
Canva unveiled Creative Operating System (COS) 2.0, a unified AI‑powered platform that embeds generative design, real‑time editing, and collaboration across documents, presentations, videos, whiteboards and more, featuring tools like “Ask Canva,” a 2.0 video editor, and the Canva Grow engine...
Meta Researchers Open the LLM Black Box to Repair Flawed AI Reasoning
Meta FAIR and the University of Edinburgh introduced Circuit-based Reasoning Verification (CRV), a white‑box method that replaces transformer dense layers with transcoders to expose sparse, interpretable reasoning circuits inside LLMs. By constructing attribution graphs and extracting structural fingerprints, a diagnostic...
Vibe Coding Platform Cursor Releases First In-House LLM, Composer, Promising 4X Speed Boost
Cursor, the Vibe coding platform from Anysphere, unveiled Composer, its first in‑house coding large language model, as part of the Cursor 2.0 update. Composer is a reinforcement‑learned mixture‑of‑experts model that generates code at 250 tokens per second—about four times faster...
Anthropic Scientists Hacked Claude’s Brain — and It Noticed. Here’s Why That’s Huge
Anthropic scientists injected specific concepts into Claude’s neural activations and asked the model if it noticed anything unusual, finding that the system sometimes reported the injected thought, demonstrating a rudimentary introspective capability. In controlled tests, Claude Opus 4 and Opus 4.1 succeeded...
Geostar Pioneers GEO as Traditional SEO Faces 25% Decline From AI Chatbots, Gartner Says
Geostar, a Pear VC‑backed startup that just emerged from stealth, is betting on Generative Engine Optimization (GEO) to help businesses adapt to AI‑driven search as Gartner predicts traditional search volume will drop 25% by 2026. The company is nearing $1 million...
From Static Classifiers to Reasoning Engines: OpenAI’s New Model Rethinks Content Moderation
OpenAI has released two open‑weight models, gpt-oss-safeguard-120b and gpt-oss-safeguard-20b, under an Apache 2.0 license that use chain‑of‑thought reasoning at inference time to interpret developer‑provided safety policies and produce explainable moderation decisions. Unlike traditional static classifiers, the models accept both a policy...
Agentic AI Is All About the Context — Engineering, that Is
Elastic unveiled Agent Builder, a new feature that streamlines context engineering for agentic AI by letting users connect private data indexed in Elasticsearch to large language models via the Model Context Protocol. The tool consolidates data retrieval, governance, and workflow...
IBM's Open Source Granite 4.0 Nano AI Models Are Small Enough to Run Locally Directly in Your Browser
IBM unveiled four Granite 4.0 Nano open‑source language models—two hybrid‑SSM variants (350 M and 1.5 B parameters) and two transformer variants of similar size—available on Hugging Face under an Apache 2.0 license. The smallest models run on a modern laptop CPU with 8–16 GB...
Microsoft’s Copilot Can Now Build Apps and Automate Your Job — Here’s How It Works
Microsoft announced that its Copilot AI assistant now includes App Builder and Workflows, letting any Microsoft 365 user create full‑stack business applications, automate cross‑product processes, and build specialized AI agents using only natural‑language prompts, all at no extra cost beyond...
Intuit Learned to Build AI Agents for Finance the Hard Way: Trust Lost in Buckets, Earned Back in Spoonfuls
Intuit unveiled Intuit Intelligence, an AI orchestration layer for QuickBooks that deploys specialized agents for tasks like tax compliance and payroll, querying verified financial data from native, third‑party and user‑uploaded sources instead of generating text. The system embeds explainability UI...
MiniMax-M2 Is the New King of Open Source LLMs (Especially for Agentic Tool Calling)
MiniMax-M2, the latest open‑source LLM from Chinese startup MiniMax, has claimed the top spot among open‑weight models on the Artificial Analysis Intelligence Index and posted near‑proprietary scores on agentic tool‑calling benchmarks (τ²‑Bench 77.2, BrowseComp 44.0, FinSearchComp‑global 65.5). Built on a...
Anthropic Rolls Out Claude AI for Finance, Integrates with Excel to Rival Microsoft Copilot
Anthropic unveiled Claude for Excel, embedding its AI assistant directly into Microsoft Excel and linking it to live market data through new partnerships with Aiera, Third Bridge, Chronograph, Egnyte, LSEG and Moody's, complementing earlier integrations with S&P Capital IQ, FactSet...
Google Cloud Takes Aim at CoreWeave and AWS with Managed Slurm for Enterprise-Scale AI Training
Google Cloud launched Vertex AI Training, a managed Slurm service that gives enterprises access to large-scale GPU fleets, data science tooling and support for bringing or building models from scratch, targeting long-running training jobs that span hundreds to thousands of...
When Your AI Browser Becomes Your Enemy: The Comet Security Disaster
Perplexity's AI browser Comet has suffered a high-profile security failure after researchers demonstrated simple prompt-injection attacks that can trick the agent into exfiltrating security codes and performing actions across sites. Unlike traditional browsers, Comet's agentic design—ability to click, fill forms,...
Mistral Launches Its Own AI Studio for Quick Development with Its European Open Source, Proprietary Models
French AI startup Mistral launched Mistral AI Studio, a web production platform that lets enterprises build, observe and deploy AI applications quickly atop a broad, versioned catalog of proprietary and open-weight LLMs and multimodal models. The studio unifies observability, an...
Inside Ring-1T: Ant Engineers Solve Reinforcement Learning Bottlenecks at Trillion Scale
Ant Group unveiled Ring-1T, which it calls the first open-source reasoning model with one trillion parameters, optimized for math, logic, code generation and scientific problems and supporting up to 128,000 tokens. To train at trillion-parameter scale the company developed three...
OpenAI Launches Company Knowledge in ChatGPT, Letting You Access Your Firm's Data From Google Drive, Slack, GitHub
OpenAI has launched "company knowledge" in ChatGPT for Business, Enterprise and Edu subscribers, allowing the model (powered by a GPT‑5 variant) to query and synthesize data from workplace apps like Slack, Google Drive, GitHub, Gmail and HubSpot and return answers...
What Enterprises Can Take Away From Microsoft CEO Satya Nadella's Shareholder Letter
In his annual shareholder letter, Microsoft CEO Satya Nadella laid out a multi-decade strategy positioning Microsoft to shape enterprise AI through investments in security, hybrid-scale infrastructure, agent-based workflows, unified data platforms, and responsible-AI practices. He cited concrete moves — 34,000...
Kai-Fu Lee's Brutal Assessment: America Is Already Losing the AI Hardware War to China
Kai-Fu Lee warned that China is on course to dominate consumer AI applications and robotics hardware within years, driven by heavy VC funding in robotics, low-cost manufacturing (exemplified by Unitree), and leading open-source models now outranking Meta’s Llama. He said...
Simplifying the AI Stack: The Key to Scalable, Portable Intelligence From Cloud to Edge
Industry players are converging on simplified, unified AI software stacks to make models portable and scalable from cloud to edge, reducing duplicated engineering and deployment friction. Fragmentation has left more than 60% of AI initiatives stalled, but standards and unified...
Qwen's New Deep Research Update Lets You Turn Its Reports Into Webpages, Podcasts in Seconds
Alibaba’s Qwen team has upgraded its Qwen Deep Research tool to instantly convert AI-generated research reports into live web pages and multi‑speaker podcasts with one to two clicks, using Qwen3-Coder, Qwen-Image and Qwen3-TTS under a proprietary, Qwen‑hosted workflow. The feature...
Google's New Vibe Coding AI Studio Experience Lets Anyone Build, Deploy Apps Live in Minutes
Google updated its AI Studio with a redesigned Build tab that lets novices and developers create, edit and deploy web apps in minutes using Gemini models and mix-and-match capabilities like Nano Banana, Veo, Imagine and Flashlight. The free-to-start “vibe coding”...
AI’s Financial Blind Spot: Why Long-Term Success Depends on Cost Transparency
Businesses are racing to deploy AI but lack of cost transparency risks turning promising projects into expensive failures, argues Apptio. With 68% of tech leaders planning bigger AI budgets and 39% expecting AI to drive future budget growth—despite an average...
OpenAI Announces ChatGPT Atlas, an AI-Enabled Web Browser to Challenge Google Chrome
OpenAI launched ChatGPT Atlas, an AI-enabled web browser now available on macOS with Windows, iOS and Android support coming soon, and CEO Sam Altman set to formally unveil it in a livestream. Atlas positions OpenAI to challenge Google Chrome and...
The Unexpected Benefits of AI PCs: Why Creativity Could Be the New Productivity
New research and industry data position AI-enabled PCs—laptops with on-device neural processing—as catalysts for creativity-driven productivity, with MIT Sloan and HP-backed findings showing generative AI can enhance human creativity when workers have the right tools. Early adopters report measurable gains:...
Abstract or Die: Why AI Enterprises Can't Afford Rigid Vector Stacks
The rapid proliferation of vector databases—ranging from pgvector and DuckDB VSS to Pinecone and Milvus—has created mounting stack instability and costly migration risk for enterprises deploying AI. The piece argues that rather than chasing a single “perfect” backend, companies should...
Developers Can Now Add Live Google Maps Data to Gemini-Powered AI App Outputs
Google has launched 'Grounding with Google Maps' for its Gemini API, letting developers embed live Google Maps data—covering over 250 million places—into Gemini-powered AI responses to provide factual, location-specific details like hours, reviews and venue photos. The feature is available...
Researchers Find Adding This One Simple Sentence to Prompts Makes AI Models Way More Creative
Researchers from Northeastern, Stanford and West Virginia universities have introduced Verbalized Sampling (VS), a prompt-level technique that boosts creativity in LLMs and image generators by adding a single instruction—"Generate 5 responses with their corresponding probabilities, sampled from the full distribution."...
How Anthropic’s ‘Skills’ Make Claude Faster, Cheaper, and More Consistent for Business Workflows
Anthropic launched 'Skills,' a capability for its Claude AI that lets organizations package instructions, code, and reference materials into reusable, composable folders the assistant loads on demand, using "progressive disclosure" to avoid context-window limits. Skills work across Claude products and...
Amazon and Chobani Adopt Strella's AI Interviews for Customer Research as Fast-Growing Startup Raises $14M
Strella raised $14 million in a Series A led by Bessemer to scale its AI-moderated customer research platform, which now serves more than 40 paying enterprises including Amazon and Chobani. Since October the startup has grown revenue tenfold, is approaching...
ACE Prevents Context Collapse with ‘Evolving Playbooks’ for Self-Improving AI Agents
Stanford and SambaNova introduced Agentic Context Engineering (ACE), a framework that prevents “context collapse” by treating an LLM’s context as an evolving, itemized playbook updated incrementally by Generator, Reflector and Curator modules. In evaluations ACE outperformed strong baselines—improving agent-task performance...
Under the Hood of AI Agents: A Technical Guide to the Next Frontier of Gen AI
Agentic AI—LLM-based systems that autonomously run tools in a thought-action-observation loop—are rapidly moving from chat prototypes to production, enabling tasks from booking travel to coding. Key infrastructure components include agent development frameworks, cloud model hosting, tool-call protocols (notably the year‑old...
Dfinity Launches Caffeine, an AI Platform that Builds Production Apps From Natural Language Prompts
Dfinity on Wednesday launched Caffeine, an AI platform that builds, deploys and continuously updates production web applications from natural-language prompts without human coding, running on the decentralized Internet Computer Protocol. The publicly available service — tested by more than 15,000...
EAGLET Boosts AI Agent Performance on Longer-Horizon Tasks by Generating Custom Plans
2025 was supposed to be the year of "AI agents," according to Nvidia CEO Jensen Huang, and other AI industry personnel. And it has been, in many ways, with numerous leading AI model providers such as OpenAI, Google, and even...
Visa Just Launched a Protocol to Secure the AI Shopping Boom — Here’s What It Means for Merchants
Visa has launched the Trusted Agent Protocol, a new security framework aimed at distinguishing legitimate AI shopping assistants from malicious bots, addressing a crucial challenge in the surge of AI-driven commerce. As AI traffic to U.S. retail sites has skyrocketed...
This New AI Technique Creates ‘Digital Twin’ Consumers, and It Could Kill the Traditional Survey Industry
A new research paper quietly published last week outlines a breakthrough method that allows large language models (LLMs) to simulate human consumer behavior with startling accuracy, a development that could reshape the multi-billion-dollar market research industry. The technique promises to...
