Ascendo AI announced the launch of the Ascendo AI Knowledge Agent, a Product Intelligence capability that automates enterprise technical documentation and knowledge creation. The Knowledge Agent converts fragmented case notes, knowledge bases, chat transcripts and files into structured, persona-driven documentation, complete with schematic diagrams, flow charts, images, video embeds, rich text and tables, using simple templates and customizable personas. Key features: Template-driven document generation (diagrams, flow charts, media, tables); Persona-driven outputs: Yoda the Teacher (Instructional), Martian the Technical Mark (Technical), Carraway the Summarizer (Key Points), Bourne the Brief (Concise), Spock the Analyst (Analytical); Ingests enterprise sources including case notes, knowledge bases, Slack/Teams transcripts, manuals, files and databases to create searchable knowledge assets; New UI to add Knowledge functionality across enterprise workflows with minimal configuration. Early customer impact – Large telecom customer: ~50% reduction in field service bulletin creation time. Leading AI company: ~80% reduction in knowledge creation time for onboarding and customer documentation. Large healthcare organization: ~70% reduction in training and onboarding time with automatically generated training modules. Day Knowledge Challenge – Ascendo AI is running a 7-Day Knowledge Challenge. Participants who create and publish knowledge using the Knowledge Agent will be spotlighted and eligible for prizes.
OpenAI creates an Expert Council on Well‑Being and AI to define healthy AI interactions, advise on sensitive scenarios and guide guardrails alongside clinical input from its Global Physician Network
OpenAI has formed a council to help it define and monitor “what healthy interactions with AI [artificial intelligence] should look like.” The Expert Council on Well-Being and AI is composed of eight researchers and experts focused on how technology affects mental health. OpenAI has consulted with many of these experts in the past, as when the company was developing parental controls and notification language for parents whose teen may be in distress. Moving forward, the council will monitor the company’s approach and will explore topics like how AI should behave in sensitive situations, what kinds of guardrails can support people using ChatGPT, and how ChatGPT can have a positive impact on people’s lives. “Alongside the Expert Council on Well-Being and AI advising on our broader approach to well-being, we’re also working with a multidisciplinary subset of mental health clinicians and researchers within the Global Physician Network to shape our model behavior and policies, and to test how ChatGPT responds in real-world situations. This work spans psychiatry, psychology, pediatrics, and crisis intervention, helping ensure our systems are grounded in clinical understanding and best practices,” the company said.
Experts suggest multi-agent testing orchestration model; with specialized agents handling natural language understanding, test plan execution, application change detection with healing and failure triage automatically routing to developers
C-level executives want their companies to use AI agents to move faster, therefore driving vendors to deliver AI agent-driven software, and every software delivery team is looking for ways to add agentic capabilities and automation to their development platforms. By parallel coding with co-pilots, some pundits are speculating that developers could increase their code output by 10 times. “The only purpose of adopting agents is productivity, and the unlock for that is verifiability,” said David Colwell, vice president of artificial intelligence, Tricentis, an agentic AI-driven testing platform. “The best AI agent is not the one that can do the work the fastest. The best AI agent is the one that can prove that the work was done correctly the fastest.” “When you prompt AI to write a test, one agent will understand the user’s natural language commands, and another will start to execute against that plan and write actions into the test, while another agent understands what changed in the application and how the test should be healed,” said Andrew Doughty, founder and chief executivce of SpotQA, creator of Virtuoso QA. “And then if there is a failure, an agent can look into the history of that test object, and then triage it automatically and send it over to developers to investigate.” “We’ve found that customers don’t need large model-based AIs to do very specific testing tasks. You really want smaller models that have been tuned and trained to do specific tasks, with fine-grained context about the system under test to deliver consistent, meaningful results,” said Matt Young, president, Functionize Inc.
TD Securities’ ChatGPT-powered assistant cuts idea generation from hours to minutes by querying the bank’s proprietary research and also provides citations, summaries, text-to-SQL tables and charts for traders and sales desks
TD Securities launched an AI virtual assistant in June that lets traders query the bank’s own equity research, and the early results look positive. “It’s a massive time save,” Dan Bosman, chief information officer at TD Securities, told American Banker. “It’s not unusual for people in the capital markets group to receive ten 20-page PDFs simultaneously in their email inbox. “Even if you’re the fastest reader and you’re really skim reading, you have to take that first 30 minutes in the day to pore over that before you can make your first call,” Bosman said. “Now with the tool, you’re able to get those insights and make those calls within minutes, as opposed to waiting half an hour.” Some people will still read the actual reports when they have time, he said. “The real need is, we put out so much content and folks on a trade floor are constantly bombarded with news and signals,” Bosman said. “Part of what we do is try to reduce that signal-to-noise ratio, getting the right signals to our sales traders and ultimately out to our clients as quickly as possible.” In the broader picture, this initiative is one of many at TD Bank Group, whose CEO Raymond Chun has set ambitious goals for AI at the firm. “Across TD, we’re deploying the capabilities needed to drive speed, such as AI-powered virtual assistants, AI-enabled adjudication, predictive tools and new applications,” Chun said at a recent investor day. “These new capabilities are already driving strong outcomes. We’re approving mortgages in hours instead of days. We’re pre-approving credit cards with data-driven insights for millions of clients. We’re producing reports in minutes versus hours or days, and we’re responding to clients in just a few seconds, significantly shortening call and wait times.” The bank is aiming to get $1 billion in annual value from AI, half through revenue increases and half through cost savings. TD now has 2,500 data scientists, engineers, data analysts and experts building proprietary platforms and applications. “This is huge,” Chun said. “AI is fast becoming fundamental to business and to client experience.” Elsewhere in TD Securities, Bosman’s team has given software developers coding assistants. “A year ago, folks were saying, we need this, and now seeing it’s being woven into their careers and what they’re doing,” he said.
Infinitus partners with Outshift by Cisco to leverage orchestration layer for discoverable AI agents built on MCP and A2A open protocols; streamlining prior authorization and insurance verifications that typically take 24-48 hours into seconds
Infinitus Systems, Announced that the company has partnered with Outshift by Cisco to streamline healthcare operations, leveraging the orchestration layer of discoverable, secure, and distributed AI agents in healthcare. Healthcare clinicians and staff face a flood of time-consuming tasks, from patient communication and follow-ups to prior authorization and insurance verifications. Infinitus AI agents streamline these vital but labor-intensive processes – which can include many thousands of faxes, phone calls, and voicemail exchanges managed by overburdened staff – to improve patient access and outcomes. Once streamlined using Infinitus AI agents, tasks that typically take 24-48 hours can instead be completed in seconds to a couple of hours. As part of this collaboration, Infinitus has contributed to AGNTCY – an open-source framework under the Linux Foundation dedicated to advancing interoperability for the Internet of Agents. AGNTCY’s mission is to establish shared standards, promote agent discoverability, and build trust among AI agents, enabling secure and coordinated multi-agent workflows in mission-critical environments like healthcare. By publishing its MCP implementation to the Agent Directory, Infinitus ensures its healthcare-focused agents can seamlessly integrate with agents from other organizations.
Together AI announces ATLAS adaptive speculator system delivering 400% inference speedup using dual-speculator architecture combining heavyweight static model trained on broad data with lightweight adaptive model learning continuously from live traffic patterns in real-time
Together AI announced research and a new system called ATLAS (AdapTive-LeArning Speculator System) that aims to help enterprises overcome the challenge of static speculators. The technique provides a self-learning inference optimization capability that can help to deliver up to 400% faster inference performance than a baseline level of performance available in existing inference technologies such as vLLM. The system addresses a critical problem: as AI workloads evolve, inference speeds degrade, even with specialized speculators in place. ATLAS uses a dual-speculator architecture that combines stability with adaptation: The static speculator – A heavyweight model trained on broad data provides consistent baseline performance. It serves as a “speed floor.” The adaptive speculator – A lightweight model learns continuously from live traffic. It specializes on-the-fly to emerging domains and usage patterns. The confidence-aware controller – An orchestration layer dynamically chooses which speculator to use. It adjusts the speculation “lookahead” based on confidence scores. The technical innovation lies in balancing acceptance rate (how often the target model agrees with drafted tokens) and draft latency. As the adaptive model learns from traffic patterns, the controller relies more on the lightweight speculator and extends lookahead. This compounds performance gains. Together AI’s testing shows ATLAS reaching 500 tokens per second on DeepSeek-V3.1 when fully adapted. More impressively, those numbers on Nvidia B200 GPUs match or exceed specialized inference chips like Groq’s custom hardware.
Reflection AI’s autonomous coding agent Asimov reads everything from emails to slack messages, project notes to documentation, in addition to the code, to learn everything about how and why the app was created
AI startup Reflection AI has developed an autonomous agent known as Asimov. It has been trained to understand how software is created by ingesting not only code, but the entirety of a business’ data to try to piece together why an application or system does what it does. Co-founder and Chief Executive Misha Laskin said that Asimov reads everything from emails to slack messages, project notes to documentation, in addition to the code, to learn everything about how and why the app was created. He explained that he believes this is the simplest and most natural way for AI agents to become masters at coding. Asimov is actually a collection of multiple smaller AI agents that are deployed inside customer’s cloud environments so that the data remains within their control. Asimov’s agents then cooperate with one another to try and understand the underlying code of whatever piece of software they’ve been assigned to, so they can answer any questions that human users might have about it. There are several smaller agents designed to retrieve the necessary data, and they work with a larger “reasoning” agent that collects all of their findings and tries to generate coherent answers to user’s questions.
Anthropic’s analytics dashboard for Claude coding agent to provide detailed breakdowns of activity by user and cost including lines of code accepted, suggestion accept rates and total spend over time
Anthropic is rolling out a comprehensive analytics dashboard for its Claude Code AI programming assistant. The new dashboard will provide engineering managers with detailed metrics on how their teams use Claude Code, including lines of code accepted, suggestion accept rates, total user activity over time, total spend over time, average daily spend for each user, and average daily lines of code accepted for each user. The dashboard will track commits, pull requests, and provide detailed breakdowns of activity by user and cost — data that engineering leaders say is crucial for understanding how AI is changing development workflows. The feature includes role-based access controls, allowing organizations to configure who can view usage data. The system focuses on metadata rather than actual code content, addressing potential privacy concerns about employee surveillance. The platform has seen active user base growth of 300% and run-rate revenue expansion of more than 5.5 times, according to company data. Unlike some competitors that focus primarily on code completion, Claude Code offers what Anthropic calls “agentic” capabilities — the ability to understand entire codebases, make coordinated changes across multiple files, and work directly within existing development workflows.
Confident Security offers an end-to-end encryption tool that wraps around foundational models, guaranteeing that prompts and metadata can’t be stored, seen, or used for AI training
Startup Confident Security aims to be “the Signal for AI.” The company’s product, CONFSEC, is an end-to-end encryption tool that wraps around foundational models, guaranteeing that prompts and metadata can’t be stored, seen, or used for AI training, even by the model provider or any third party. The company wants to serve as an intermediary vendor between AI vendors and their customers — like hyperscalers, governments, and enterprises. CONFSEC is modeled after Apple’s Private Cloud Compute (PCC) architecture, which “is 10x better than anything out there in terms of guaranteeing that Apple cannot see your data” when it runs certain AI tasks securely in the cloud. Like Apple’s PCC, Confident Security’s system works by first anonymizing data by encrypting and routing it through services like Cloudflare or Fastly, so servers never see the original source or content. Next, it uses advanced encryption that only allows decryption under strict conditions. Finally, the software running the AI inference is publicly logged and open to review so that experts can verify its guarantees. CONFSEC is also well-suited for new AI browsers hitting the market, like Perplexity’s Comet, to give customers guarantees that their sensitive data isn’t being stored on a server somewhere that the company or bad actors could access, or that their work-related prompts aren’t being used to “train AI to do your job.”
Analog Devices AI tool automates the end-to-end machine learning pipeline for edge AI, including model search and optimization using state-of-the-art algorithms and verifies model size against the device’s RAM to enable successful deployment
Analog Devices Inc. (ADI) has introduced AutoML for Embedded, an AI tool that automates the end-to-end machine learning pipeline for edge AI. The tool, co-developed with Antmicro, is now available as part of the Kenning framework, integrated into CodeFusion Studio. The Kenning framework is a hardware-agnostic and open-source platform for optimizing, benchmarking, and deploying AI models on edge devices. AutoML for Embedded allows developers without data science expertise to build high-quality and efficient models that deliver robust performance. The tool automates model search and optimization using state-of-the-art algorithms, leveraging SMAC to explore model architectures and training parameters efficiently. It also verifies model size against the device’s RAM to enable successful deployment. Candidate models can be optimized, evaluated, and benchmarked using Kenning’s standard flows, with detailed reports on size, speed, and accuracy to guide deployment decisions. Antmicro’s Michael Gielda, VP Business Development, said that AutoML in Kenning reduces the complexity of building optimized edge AI models, allowing customers to take full control of their products. AutoML for Embedded is a Visual Studio Code plugin built on the Kenning library that supports: ADI MAX78002 AI accelerator MCUs and MAX32690 devices — deploy models directly to industry-leading edge AI hardware. Simulation and RTOS workflows — leverage Renode-based simulation and Zephyr RTOS for rapid prototyping and testing. General-purpose, open-source tools — allowing flexible model optimisation without platform lock-in