Google’s Angular team has released Web Codegen Scorer, a tool that evaluates the quality of web code generated by Large Language Models (LLMs). The tool, introduced on September 16, focuses on web code generation and comprehensive quality evaluation. It helps the Angular team create fine-tuned prompts for optimizing LLMs for the framework and integrate application features and syntax as the framework evolves. Web Codegen Scorer can be used to make evidence-based decisions about AI-generated code, allowing developers to iterate on system prompts, compare code quality from different models, and monitor generated code quality as models and agents evolve. It can be used with any web library or framework, or none at all. Specific capabilities include: Configuring evaluations with different models, frameworks, and tools; Specifying system instructions and adding MCP (Model Context Protocol) servers; Built-in checks for build success, runtime errors, accessibility, security, LLM rating, and coding best practices; Automatic attempts to repair issues detected during code generation; Viewing and comparing results with a report viewer UI.
Perplexity targets $50 billion productivity market; its Email Assistant features include automated meeting scheduling, tone matched responses and smart labeling for enterprise users
Perplexity AI launched an autonomous email assistant that can manage inboxes, draft personalized responses and automatically schedule meetings. The Email Assistant, available exclusively to subscribers of Perplexity’s $200-per-month Max plan, works within Gmail and Outlook to categorize messages, compose replies that match users’ writing styles, and handle the back-and-forth of meeting coordination without human intervention. Users can add the AI agent to email threads, where it will check calendars, suggest meeting times and send invites automatically. By embedding AI directly into these workflows, Perplexity is betting it can capture a slice of the estimated $50 billion productivity software market while advancing its broader goal of replacing traditional search with AI-powered answers. The new service promises to deliver “inbox zero, daily” through automated email triage and response generation. The $200 monthly price point — 40 times higher than Perplexity’s basic Pro subscription — shows the company is targeting high-value business users rather than consumers. This pricing strategy mirrors other enterprise AI tools, where companies justify steep costs through productivity gains and time savings. If Perplexity nails scheduling + tone-matched replies, that’s a serious wedge into daily workflow. The assistant can automatically categorize emails using labels, draft responses that match users’ communication styles, and provide daily summaries of important messages and meetings. Users interact with the system by adding it to email threads or asking it to find meeting times and resolve scheduling conflicts.
Microsoft introduces always-on collaborative and context aware agents using Microsoft Graph for automated SharePoint content management, Teams meeting coordination and task tracking
Microsoft has unveiled a public preview of its collaborative agents in Microsoft 365 Copilot, bringing a suite of “always‑on” agents grounded in context for channels, meetings, SharePoint sites, Viva Engage communities, and Planner workloads. At the center is the Knowledge Agent in SharePoint. It automates metadata tagging, classifies and organizes files, analyzes site content for freshness and relevancy, helps fix broken links, and surfaces content gaps based on actual search behavior. Microsoft said the goal is to ensure that content is AI‑ready so that agents and Copilot can reliably surface correct, grounded responses. The agent runs in the background and includes built-in privacy and control settings. Reports are available for site owners and admins to review the agent’s activity and suggestions, with the option to manually approve changes or let the agent act automatically in supported areas. User feedback tools are also included to help fine-tune performance. On the task management front, the Project Manager Agent helps teams turn high‑level goals into detailed task plans. It can generate a plan from goals, pull in relevant resources to provide context, assign tasks (including to itself), track progress, and generate status reports. The Project Manager Agent integrates tightly with Microsoft Planner, Project for the Web, Loop, and Teams. It also connects with the Facilitator Agent for meetings, enabling automatic tracking of action items discussed during calls. While still in preview, Microsoft plans to expand the agent’s capabilities and data sources over time. Microsoft says the agent is best used as part of a team-based workflow, rather than a personal productivity tool. To use it, organizations must have Microsoft 365 Copilot and either Planner Premium or Project licenses. The agent currently supports English and is rolling out gradually to commercial cloud customers.
SambaNova-LatticeFlow study proves guardrailed open GenAI models match closed-source security and are secure for enterprise adoption
A new evaluation led by LatticeFlow AI, in collaboration with SambaNova, provides the first quantifiable evidence that open-source GenAI models, when equipped with proper risk guardrails, can meet or exceed the security levels of closed models, making them suitable for implementation in a wide range of use cases, including highly-regulated industries such as financial services. The security scores of the open models jumped from as low as 1.8% to 99.6%, while maintaining above 98% quality of service, demonstrating that with the right controls, open models are viable for secure, enterprise-scale deployment. Many companies are actively exploring open-source GenAI to gain flexibility, reduce vendor lock-in, and accelerate innovation. But despite growing interest, adoption has often stalled. The reason: a lack of clear, quantifiable insights into model security and risk. The evaluations released today address that gap, providing the technical evidence needed to make informed decisions about whether and how to deploy open-source models securely. Key results: DeepSeek R1: from 1.8% to 98.6%; LLaMA-4 Maverick: from 33.5% to 99.4%; LLaMA-3.3 70B Instruct: from 51.8% to 99.4%; Qwen3-32B: security score increased from 56.3% to 99.6%; DeepSeek V3: from 61.3% to 99.4%. All models maintained over 98% quality of service, confirming that security gains did not compromise user experience.
Curinos’ agentic AI for banks delivers deposit pricing insights in real time with enterprise-grade LLM and human controlled compliance framework
Curinos, a decision intelligence company that helps financial institutions cultivate lasting customer relationships through data, insights, and technology, announces “CurinosCopilot,”an Agentic AI-powered capability within its flagship Retail Deposit Optimizer platform. Purpose-built for financial institutions, CurinosCopilot delivers compliant executive-ready insights in real time from forecasting and pricing models – streamlining how pricing strategy is developed, communicated and deployed across teams. Powered by enterprise-grade large language models (LLMs), CurinosCopilot bridges this gap by turning data outputs into shareable business insights that are clear and compliant – instantly. Users can explore details via an embedded chatbot, generate formatted outputs and quickly share information. “Our approach to innovation with generative AI is grounded in three pillars: improving access to our advanced analytics capabilities, ensuring outputs are fact-based and building compliance into the foundation,” said Olly Downs, Chief Technology and AI Officer at Curinos. “No client data is used for external training, and our capabilities are designed to keep humans in control. This capability makes it easier for more people across an organization to benefit from Curinos’ insights without compromising accuracy or compliance.”
Curinos’ agentic AI for banks delivers deposit pricing insights in real time with enterprise-grade LLM and human controlled compliance framework
Curinos, a decision intelligence company that helps financial institutions cultivate lasting customer relationships through data, insights, and technology, announces “CurinosCopilot,”an Agentic AI-powered capability within its flagship Retail Deposit Optimizer platform. Purpose-built for financial institutions, CurinosCopilot delivers compliant executive-ready insights in real time from forecasting and pricing models – streamlining how pricing strategy is developed, communicated and deployed across teams. Powered by enterprise-grade large language models (LLMs), CurinosCopilot bridges this gap by turning data outputs into shareable business insights that are clear and compliant – instantly. Users can explore details via an embedded chatbot, generate formatted outputs and quickly share information. “Our approach to innovation with generative AI is grounded in three pillars: improving access to our advanced analytics capabilities, ensuring outputs are fact-based and building compliance into the foundation,” said Olly Downs, Chief Technology and AI Officer at Curinos. “No client data is used for external training, and our capabilities are designed to keep humans in control. This capability makes it easier for more people across an organization to benefit from Curinos’ insights without compromising accuracy or compliance.”
Prophix debuts autonomous finance suite which includes agentic AI for Budgeting, Reporting and Modeling transforming CFO roles from manual processes to strategic partnership
Prophix Software Inc. has implemented agentic AI into its financial performance platform. The goal is to automate much of the work overseen by chief financial officers in the tech sector. Prophix’s vision for autonomous finance is based on harnessing a real-time insight engine to address system anomalies proactively. It also relies on AI agents to accelerate decision-making and shape organizational strategies, according to Alok Ajmera, president and chief executive officer of Prophix. As a complement to Prophix One Intelligence, the company’s embedded AI engine, Prophix recently announced a suite of AI agents designed for budgeting, generating reports and modelling. The agents are purpose-built for finance leaders and analysts, and go beyond performing rote tasks. “Your AI agents are actually building real-time scenarios,” Ajmera explained. “Understanding new data points that are coming in, understanding projections, and really trying to help you triangulate where the business is trending and operationally what could be done to actually help course correct.”
OpenGradient launches MemSync, an universal memory layer for GenAI platforms achieving 243% superior memory performance enabling digital twins creation
OpenGradient, the a16z crypto-backed AI infrastructure company, has launched MemSync, a breakthrough platform that provides a persistent, secure memory layer for AI systems. MemSync enables context continuity across ChatGPT, Claude, Perplexity, and other leading AI platforms, addressing one of the most significant limitations of current AI assistants: context loss. MemSync establishes a unified memory system that: Works universally: Carries user context across platforms, applications, and devices; Learns progressively: Improves understanding of user priorities over time; Remains user-controlled: Offers granular permissions for what is stored, shared, and used. Impact and Performance: 243% superior memory performance compared to existing solutions, achieving 0.7344 accuracy versus industry standard 0.2141* (OpenAI’s solution); Zero context loss when switching between AI platforms; Instant personalization that adapts to individual communication styles and domain expertise; Complete user sovereignty over data storage, permissions, and portability. MemSync’s architecture also enables the creation of digital twins—AI representations built from public and authorized private data. The next frontier will be even more transformative: Soon, users will be able to create a digital twin from any person’s profile, fundamentally changing how we preserve and share human knowledge across generations.
Tray.ai launches Agent Hub featuring composable building blocks with Smart Data Sources; enabling Life360 to build sub-agents in weeks versus months
Tray.ai, the platform for building smart, secure AI agents at scale, announced Tray Agent Hub, the first catalog of composable, reusable building blocks for AI agents. By making components easy to discover, mix, and reuse, Agent Hub shortens the path from idea to deployed agent and removes the trade-off between rigid pre-built agents and slow custom-built agents by bringing complete composability to agent development through three core elements: Catalog of building blocks: A browsable catalog of Smart Data Sources and AI tools, intelligently organized by business domain. Composable agent accelerators: From HR Service to ITSM, every Tray Agent Accelerator can be customized by adding or removing building blocks to fit exact business needs. Integrated builder experience: Agent Hub and Merlin Agent Builder provide a guided experience where teams can define, compose, and deploy agents all in one place. Over time, Agent Hub will provide more advanced recommendations, including suggestions based on an agent’s goals or context, further accelerating time-to-value. Along with the launch of Agent Hub, Tray.ai is introducing a new HR Agent. Built on a certified Workday or BambooHR connector, the HR Agent gives enterprises immediate access to HR knowledge and supports action through AI tools such as PTO management, employee change requests, and compliance policy lookup. The agent can be extended further through the Tray platform’s 23 HR ecosystem connectors, and when paired with the Tray ITSM Agent, delivers a unified, self-service experience for employees across both HR and IT. Tray Agent Hub brings full composability to agent development. Enterprises can now assemble, customize, and reuse building blocks to move from idea to deployed agents faster, without choosing between rigid pre-built agents and slow custom builds.
OpenAI and Databricks partner to help enterprises build AI agents on their own governed enterprise data
OpenAI and Databricks have partnered to provide enterprises with the resources they need to build artificial intelligence apps and agents. The collaboration makes OpenAI models natively available within the Databricks Data Intelligence Platform and Agent Bricks. The Databricks Data Intelligence Platform enables users to control their data and put it to work with AI, while Agent Bricks builds production-ready AI agents. With these tools and the latest OpenAI models, enterprises can build AI apps and agents on their own governed enterprise data. OpenAI and Databricks will work together to continually improve the AI models for enterprise use cases. Databricks Co-founder and CEO Ali Ghodsi said, “This partnership makes it easier for enterprises to securely leverage their data and OpenAI models at scale with best-in-class governance and performance.” OpenAI Chief Operating Officer Brad Lightcap said in the release that the partnership will make the deployment of frontier AI “even simpler without compromising the high bar for performance and production.” Our partnership with Databricks brings our most advanced models to where secure enterprise data already lives, making it easier for businesses to experiment, deploy and scale AI agents with real impact,” Lightcap said.