Twinko AI | Zero Degree Weekly | AI Agents
Introduction
Microsoft is actively advancing its Multi-Agent AI (MAI) strategy, directly challenging OpenAI and DeepMind, while also launching a new personal shopping agent based on Copilot Studio. This week saw a surge in the deployment of AI agents: Box introduced a new AI agent for content management, Notion released an AI agent expected to automate numerous work functions, and Amazon launched an AI agent to assist third-party sellers. Oracle also expanded its Fusion HCM with 13 new AI agents for HR automation. OpenAI anticipates that millions of AI agents will autonomously operate in the cloud within a few years, prompting Xage Security to extend zero trust into the AI domain for safer data access and multi-agent workflows. Google and Coinbase are collaborating to introduce stablecoin payments into AI applications, while publishers and advertisers are grappling with new regulatory challenges posed by these proliferating autonomous agents.
Key News This Issue
Microsoft’s Multi-Agent AI Strategy: How MAI Competes with OpenAI and DeepMind
Date: 2025-09-14 Source: Ts2
Microsoft is strengthening its Multi-Agent AI (MAI) strategy, moving beyond its role as an OpenAI distributor to develop proprietary AI capabilities and unique AI assistant approaches. This initiative was highlighted with the release of MAI-Voice-1 and MAI-1-preview in August 2025, aiming to give Microsoft a complete AI stack, from foundational models to multi-agent orchestration frameworks, optimizing the integration of multifunctional assistants through a unified Copilot interface. This positions Microsoft in competition with OpenAI and Google/DeepMind in key areas such as voice interaction, as Microsoft seeks an advantage through deep integration into PC and enterprise workflows, contrasting with OpenAI’s versatile ChatGPT voice capabilities and Google’s extensive device ecosystem. In software development, Microsoft leads with GitHub Copilot, while OpenAI and DeepMind are also advancing AI coding capabilities, creating a competitive environment that benefits developers through rapidly improved models and features, such as Microsoft’s interactive debugging agent in VS Code. Microsoft’s core strategy involves the universal integration of AI “Copilots” into its product ecosystem (Windows, Office, Edge, Visual Studio), promoting open plugin standards to ensure interoperability and user choice. While their strategies sometimes differ—Microsoft focusing on a specialized, product-driven approach, while OpenAI emphasizes a general-purpose platform API, and DeepMind pursues research-driven work—they all tend to view AI agents as the next computing paradigm. Executives like Satya Nadella, Demis Hassabis, and Sam Altman envision a ubiquitous AI-assisted future, where Microsoft’s MAI emphasizes a range of specialized AIs to provide personalized experiences, OpenAI pushes for smarter general models, and DeepMind integrates agent AI into Google’s vast ecosystem. This competitive-cooperative relationship drives innovation, promising more powerful and convenient technologies while necessitating careful handling of challenges such as hallucinations, privacy, and misuse.
Box Launches New AI Agent for Content Management
Date: 2025-09-12 Source: Techzine
Box is significantly enhancing its content management capabilities by introducing new AI agents—Box Extract, Box Automate, and enhanced Box Apps—aimed at streamlining enterprise business processes. Box Automate facilitates workflow orchestration between various agents and teams, supporting automation of routine tasks and complex cross-system workflows through a visual builder, requiring no programming knowledge. The platform allows for the creation of custom agents for specific functions such as Q&A, composing, extracting, and researching, dynamically routing tasks based on business logic, and seamlessly integrating with Box’s native tools (like Box Forms, Box Doc Gen, and Box Sign), while also extending to custom applications and external systems via API. Box Extract serves as a core component, leveraging advanced AI to accurately extract data from various document types (including contracts, scanned forms, PDFs, spreadsheets, images, and handwritten text), understanding documents and extracting complex, interrelated data points through standard and enhanced extraction agents, with built-in validation features. Additionally, Box Apps now include new AI capabilities, providing no-code dashboards that unify content, metadata, users, workflows, and agents, enabling natural language queries for content exploration, dynamic data visualization, and agent-assisted analysis to identify trends and anomalies. Box offers customers flexibility in choosing AI models through Box AI Studio and Box AI API, including state-of-the-art models from Anthropic, Amazon, Google, IBM, Meta, OpenAI, and xAI, and supports integration with third-party agents via remote Box MCP servers, with existing integrations including Anthropic Claude, Microsoft 365 Copilot, Mistral Le Chat, OpenAI ChatGPT, and Salesforce Agentforce. These three agent solutions will be available to customers with Enterprise Advanced subscriptions in the coming months, marking a strategic step towards AI-driven content automation.
Microsoft Launches Personal Shopping Agent to Enhance AI-Driven Retail
Date: 2025-09-12 Source: Winbuzzer
Microsoft has launched a preview of its personal shopping agent, an AI tool built on Copilot Studio, designed to empower retailers through a conversational shopping experience. This “headless” B2B solution can be integrated into websites, applications, and in-store tools, acting as a 24/7 “digital store assistant” that provides personalized product discovery and supports human sales staff. The agent is highly customizable, aligning with retailers’ brand guidelines, product catalogs, and policies, exemplified by Ralph Lauren’s “Ask Ralph” in-app styling tool, which its Chief Digital Officer Naveen Seshadri envisions as a means to create a more immersive shopping experience. This initiative aligns with CEO Satya Nadella’s vision of an “AI Agent Era,” where intelligent agents autonomously perform complex tasks, with Microsoft AI CMO Jared Spataro emphasizing that “agents are the new applications in an AI-driven world,” further reinforcing this concept. This release intensifies competition in the AI-driven retail space, with competitors like OpenAI introducing shopping features in ChatGPT, Perplexity enabling one-click purchases through “Buy with Pro,” and Google launching virtual try-on tools and preparing generative “AI modes” for search. Meanwhile, Microsoft is developing the “Copilot Wallet” and testing features like “Copilot Appearance” and “Desktop Share,” while preparing for OpenAI’s GPT-5 model, reflecting Microsoft AI CEO Mustafa Suleyman’s vision for integrated personal AI. However, this push for deeper integration and increased user trust (including access to financial and screen data) occurs against a backdrop of tightening security scrutiny, especially following the “EchoLeak” vulnerability in June 2025, and Gartner’s prediction that by 2028, a quarter of enterprise data breaches will stem from AI agent misuse. Therefore, building user trust is crucial for the success of Microsoft’s agent AI strategy.
Notion’s New AI Agent Will Essentially Do Your Work for You
Date: 2025-09-18 Source: The Verge
Notion has released Notion 3.0, introducing its new AI agent, marking the company’s entry into the “Agent Era” of AI evolution. This “teammate and Notion superuser” is designed to perform tasks traditionally handled by humans within the Notion platform, including automated creation of pages and databases. In addition to the Notion workspace, this AI agent can search for information in connected tools like Slack and the broader internet, plan, and execute up to “20 minutes of autonomous work” across hundreds of pages at once. A key feature is the agent’s ability to “remember” user preferences, such as content citation and archiving methods, storing these “memories” in editable user profiles. While current capabilities allow for the creation of personalized agents with different behaviors under multiple profiles, Notion plans to roll out fully automated and customizable agents in the future. Highlighted practical applications include generating and editing email marketing campaigns, consolidating and analyzing feedback from various platforms into a single report, and converting meeting notes into emails and proposals. Notion co-founder Akshay Kothari demonstrated the agent’s capabilities, showcasing its ability to create trackers and databases based on external data sources like Rotten Tomatoes ratings. This development marks a significant advancement in AI-driven productivity tools, potentially redefining the management and execution of knowledge work within organizations.
Amazon Launches AI Agent to Help Sellers Complete Tasks and Manage Business
Date: 2025-09-17 Source: TechCrunch
Amazon has significantly enhanced support for third-party sellers by launching a constantly available AI agent (an update to its existing seller assistant tools). This advanced AI is designed to proactively manage and automate a wide range of tasks, from daily operations to complex business strategies, allowing sellers to focus on innovation and growth. The seller assistant can now monitor account health and inventory, devise strategies, and take authorized actions. For example, it can identify slow-moving products to prevent long-term storage fees, suggest whether to retain, discount, or remove items, and analyze demand patterns to prepare shipping recommendations. Additionally, the AI agent continuously monitors accounts for potential issues, such as inventory listings that violate new product safety regulations, ensuring sellers comply with regulations in all operating countries. This move aligns with the broader industry trend towards agent-driven business development, where AI agents initiate transactions and purchases on behalf of customers, as evidenced by Google’s recent release of new payment protocols for agent transactions. Amazon is also extending agent AI capabilities into advertising, enabling sellers to develop marketing campaigns through conversational prompts. This initiative builds on Amazon’s previous AI tools for sellers, including video generators for ads and generative AI tools for improving product listings, highlighting the company’s commitment to leveraging AI to enhance seller productivity and platform efficiency.
Xage Security Extends Zero Trust to AI, Promising Safer Data Access and Multi-Agent Workflows
Date: 2025-09-12 Source: Industrialcyber
Xage Security has launched a unified zero trust platform specifically designed to protect AI environments, extending its mature critical infrastructure protection principles to address the growing security challenges of AI adoption. The platform provides fine-grained, enforceable controls over AI data access, tool usage, and multi-agent workflows, aiming to eliminate jailbreak risks and alleviate AI adoption anxiety. Industry experts emphasize the critical need for robust controls as AI integration accelerates across sectors. Mark Gudiksen, managing partner at Piva Capital, stresses that the long-term success of AI depends on strict governance, not just innovation. Frank Dickson, IDC’s global vice president of security and trust, highlights the necessity of rethinking AI identity, advocating for encrypted verifiable identities, scoped permissions, and adherence to the principle of least privilege for AI agents. Xage’s identity-first zero trust architecture enforces real-time, context-aware controls across the entire AI and data center stack (from physical infrastructure to digital workloads). The platform integrates jailbreak prevention controls at the network level, utilizing the Model Context Protocol (MCP) to prevent AI data leaks and protect application-to-application interactions. It also ensures rogue AI containment through least privilege restrictions, enabling secure deployment of AI chatbots like Copilot or Claude with sensitive data. Xage Security CEO Duncan Greatwood notes the shift from vulnerable prompt filters to true zero trust enforcement at the network protocol level, ensuring prevention of data leaks and rogue AI behavior. This solution is particularly suited for high-risk industries such as government, energy, healthcare, and food and beverage, which, despite complex security environments, are leading AI adoption. Rocko Rodriguez, director of network strategy and mission strategy at SAIC, confirms collaboration with Xage to protect critical operational AI applications and data centers for the government.
Oracle Expands Fusion HCM with 13 New AI Agents for HR Automation
Date: 2025-09-16 Source: Siliconangle
Oracle Corp. has significantly expanded its Fusion Cloud Human Capital Management (HCM) suite by introducing 13 new AI agent extensions, bringing the total number of agents in the application to over 100. These autonomous AI software agents are designed to automate entire workflows and perform operations independently with limited human oversight, supporting key HR functions such as internal mobility, performance management, learning and development, payroll, and staffing processes. Yvette Cameron, Oracle’s senior vice president of global HCM product strategy, emphasizes that these agents go beyond routine queries to achieve workflow automation, present real-time insights, and eliminate friction in daily tasks. For example, the learning mentor agent recommends suitable training courses to employees based on metadata and inferred skills. Customers can customize these agents through retrieval-augmented generation (RAG) or by using Oracle’s AI Agent Studio to fine-tune them to specific internal policies and data types, including structured and unstructured data. These agents also provide interoperability with external systems through APIs and AI Agent Studio, supporting emerging standards like Agent2Agent and Model Context Protocols, allowing integration with non-Oracle platforms. Security, privacy, and data management are handled by Oracle Cloud Infrastructure. Notably, Oracle offers these delivered agents for free, allowing unlimited use, contrasting with competitors that typically charge extra. Oracle plans to continue adding new agents to its application suite quarterly, underscoring its strategic commitment to leveraging AI to enhance HR efficiency and strategic decision-making.
OpenAI Predicts Millions of AI Agents Will Operate “Somewhere in the Cloud” in the Coming Years
Date: 2025-09-16 Source: Business Insider
OpenAI envisions that millions of AI agents will autonomously operate in the cloud within a few years, creating significant economic value under human supervision. This strategic direction was articulated by co-founder and president Greg Brockman and Codex engineering lead Thibault Sottiaux on the “OpenAI Podcast,” with the core idea being that AI agents will become mature collaborators in software engineering and broader business operations. The company recently released GPT-5 Codex, a significant improvement over earlier versions capable of running for hours on a range of complex software projects, including extensive code refactoring. This development signals a shift towards highly persistent and capable AI systems. OpenAI predicts that millions of these agents will work within enterprise and public data centers, performing useful tasks. This vision underscores a profound transformation in software development practices, business operation models, and the growing demand for cloud infrastructure, prompting executives to consider AI integration strategies and the implications for workforce augmentation and technology investment.
Google and Coinbase Collaborate to Introduce Stablecoin Payments into AI Applications
Date: 2025-09-19 Source: Geeky-gadgets
Google and Coinbase are collaborating on an initiative to integrate stablecoin payments into AI applications, aiming to establish an AI-driven economy where autonomous AI agents can negotiate transactions, manage resources, and execute trades independently. At the core of this collaboration is Coinbase’s x402 payment protocol, a blockchain-based system designed for seamless, low-cost stablecoin microtransactions, optimized for the high-frequency, low-value interaction characteristics of AI agents. This protocol, along with Google’s Agent Payments Protocol (AP2) and Agent-to-Agent (A2A) protocol, facilitates the development of an agent marketplace where AI agents can autonomously discover and interact with services. While AP2 is currently open to developers and partners for experimentation rather than widespread consumer deployment, the initiative establishes an early controlled environment for AI-driven transactions, refining agent capabilities while minimizing risks. Potential applications range from supply chain optimization and automated financial services to personalized digital assistants. This collaboration addresses key challenges such as monopolies, security, and equitable access, while marking a significant step in redefining digital commerce and financial interactions within the AI ecosystem, providing early adopters with a potential competitive advantage.
Publishers and Advertisers Face New Regulatory Challenges from AI Agents
Date: 2025-09-14 Source: Digiday
The surge of autonomous AI agents presents significant governance and accountability challenges for publishers and advertisers, as these systems increasingly operate independently and guide other agents. Platform providers like Salesforce, Adobe, Microsoft, and Optimizely are launching agent AI tools that make decisions and take actions on behalf of users. Protocols like the Model Context Protocol and Google’s open-source A2A (Agent-to-Agent protocol) facilitate agent logins to websites and API usage, enabling automated tasks like dynamic ad creative changes. This shift has already impacted web traffic, with TolBit reporting a 9.4% decrease in human visitors between Q1 and Q2, while autonomous headless browsing activities from AI engines like Perplexity have increased, showing up in website logs as human visits. Huge’s CTO Marc Maleh emphasizes the urgent need for robust governance frameworks to prevent multi-agent systems from making unauthorized decisions that could lead to brand and financial liabilities. Discussions with clients like NBCUniversal and Planet Fitness often involve model decision documentation, traceability mechanisms (such as audit trails and data logs), and accountability for biased or harmful agent outputs. Protecting consumer data privacy and preventing unintended consequences from third-party agent interactions are also critical. While agent orchestration platforms like Adobe’s Agent Orchestrator and Microsoft’s Copilot Studio include governance features, effective governance requires policies that go beyond mere automation. Clive Henry, Adobe’s partner solutions director, points out that consumer data privacy issues, similar to those that prompted GDPR and CCPA, may drive new regulations for agent workflows. For media companies, agent browsing of streaming applications raises questions about ad control and data localization. David Berkowitz, founder of the AI Marketers Guild, stresses that CMOs must engage early in these discussions, as agent systems could fundamentally change marketing by creating messages for bots rather than humans. The ability of AI agents to impersonate human users (as demonstrated by Perplexity’s Comet autonomous browsing appearing as a standard Chrome user with residential IP) underscores the necessity of these safeguards.
– THE END –