VentureBeat

Join our daily and weekly newsletters for the latest updates and exclusive content on industry-leading AI coverage. Learn More Genies, a culture-focused avatar technology company, has launched Parties after developing its foundational technology stack since the last fundraise. Genies equips intellectual property owners and creators with a suite of avatar technology and user-generated content (UGC) tools to co-create and launch their own Parties — gaming platforms driven by AI avatars, user-generated content, and personalized gameplay. Genies raised a whopping $150 million in 2022, and it has been busy ever since. “The ‘next internet’ is already upon us as the convergence of AI, gaming, and XR become more and more realized – mobile apps will soon start to look like ‘mini Robloxes’ powered by two key defining themes: smart avatars and UGC,” said Akash Nigam, CEO of Genies, in a statement. “Traditional usernames and profiles will transform into smart avatar identities and fans will become co-creators of a brand’s IP through UGC. Our Party tech stack, which enables dynamic identities, UGC, and interoperability, is at the core of the Genies’ mission to power the next layer of the internet.” Genies is touting smart avatars. Genies’ technology stack focuses on powering dynamic avatars, unlocking a variety of quality user-generated content, and enabling interoperability so that a user can seamlessly travel from one Party to the next. For IP partners, this technology transforms their online presence into their own Party, unlocking the ability for their audience to build relationships with their Smart Avatars and co-create an infinite supply of UGC games, avatars, fashion, and more, all of which is built around their IP.  “Disney’s $1.5 billion deal with Epic Games is a major indicator of how UGC-driven platforms are becoming a central pillar for brands to engage with their audiences,” said Nigam. “This partnership, along with the growing number of brands activating within platforms like Roblox and Fortnite, signals that more and more IP owners will look to launch their own UGC worlds. Our Party technology stack provides the opportunity to enable limitless creativity and customization for people to unlock this new era of UGC-driven platforms.”  Foundational components that make up a Party UGC creation with Genies. Bringing IP to Life Through Smart Avatars: Any IP can be transformed into interactive smart avatars powered by AI, giving each character a distinct personality and role as a foundational NPC (non-playable character) within a Party. These smart avatars foster deeper engagement by allowing users to build relationships with them and empowering developers to build personalized experiences with them. With AI capabilities driving their personalities, Genies’ avatars enhance gameplay and make Party apps a personalized, interactive experience. Fueled by UGC: User-generated content is at the core of each Party, delivering a continuous influx of new games, avatars, fashion, accessories, props, and more—all created by a community of technical and non-technical users around a brand’s IP. IP owners can let their communities shape the experience and their brand by contributing any content, while they maintain oversight to determine what ultimately gets released. This UGC foundation ensures a constantly evolving, engaging environment for users to contribute and let their imagination run wild.   Ecosystem-wide interoperability: User avatars are created as a single persistent identity that are not only compatible with all other assets across the ecosystem (fashion, props, etc.), but they also can be seamlessly transported across any game or Party app. Whether you’re in a Party that uses anime style or a Party that uses Pixar-level characters and games, a user’s avatar can be ported between either and will automatically ‘evolve’ to adapt and reflect to those environments. Ultimately, everything works with everything.  Distribution driven through Genies’ celebrity network To amplify reach and engagement, Genies offers access to its Celebrity Network, enabling IP partners to collaborate with the biggest icons across music, sports, entertainment, and more. By pairing a brand’s Smart Avatars with figures from Genies’ celebrity roster, brands can create opportunities for the community to build games and content around both IPs, bringing together diverse fan bases for an immersive, shared experience. The tech stack Parties are made possible by the following foundational technologies, all designed to enable interoperability, limitless customization, and personalized engagement. Genies’ Avatar Framework leverages machine learning and advanced computer graphics to support highly customizable and compatible avatars, fashion, and props. By utilizing proprietary AutoRigging technology, the framework enables effortless avatar integration across games and Parties, creating a seamless user experience and fueling a thriving UGC ecosystem. Genies’ avatars are powered by AI, enabling them to evolve as interactive personalities that can communicate and form relationships with users across apps as smart avatars. Through their “Traits Framework,” Genies avatars adapt to user interactions, driving unique and personalized engagement across gaming and social applications. Genies equips developers with a comprehensive toolkit, allowing them to build and publish unique smart games directly into Party apps. Each game is connected through the Genies Login, expanding the content library and increasing user engagement across Party apps. By fostering a collaborative environment, developers can create immersive, branded experiences that keep users coming back. For IP partners, these Parties serve as a scalable and customizable solution, enabling them to launch their own interactive ecosystems that seamlessly blend gaming, social interaction, and co-creation from their audience and fans. source

Join our daily and weekly newsletters for the latest updates and exclusive content on industry-leading AI coverage. Learn More Israeli startup xpander.ai has introduced the Agent Graph System (AGS), which it says is a major new approach to building more reliable and efficient multi-step AI agents based on underlying AI models such as OpenAI’s GPT-4o series. The goal is to redefine how AI agents interact with APIs and other tools, making advanced automation tasks more accessible to organizations across industries. From left: Ran Sheinberg, co-founder and chief product officer of xpander.ai and David (Dudu) Twizer, co-founder and CEO of xpander AI. Credit: xpander.ai Solving the challenges of multi-step AI agents Function calling, the backbone of most AI agent workflows, enables models to interact with external systems to perform tasks such as fetching real-time data or executing actions. However, these interactions often falter when faced with complex API schemas or unpredictable responses, leading to inefficiencies and errors. xpander.ai’s Agent Graph System introduces a structured solution to these challenges by using a graph-based workflow that guides agents through appropriate API calls step by step. Instead of presenting all available tools at every stage, AGS intelligently restricts options to only those that align with the current context of the task, significantly reducing out-of-sequence or conflicting function calls. Ran Sheinberg, co-founder and chief product officer at xpander.ai, explained in an interview with VentureBeat: “With AGS, we ensure the agent only uses the relevant tools at each step and follows the correct schema, enforcing precision and efficiency.” Sheinberg previously worked at several other startups and as a principal solutions architecture leader at Amazon Web Services (AWS), leading large-scale compute projects with enterprise customers. Democratizing AI agent development xpander.ai aims to make agentic AI development accessible to a broader audience. “We aimed to create an accessible platform that allows anyone to build AI agents, experiment with the technology, and start automating repetitive tasks to focus on what truly matters,” said David Twizer, co-founder and CEO of xpander.ai, in the same interview. The company also offers AI-ready connectors that integrate easily with NVIDIA NIM (Nvidia Inference Microservices) and other systems. These connectors enrich API tools with detailed documentation, operational IDs, and schemas, reducing the technical burden on developers while enhancing runtime accuracy. “Once the setup is complete, you can connect it to any AI system that supports function calling,” Twizer said. “It was crucial for us to design technology that meets customers where they are and offers flexibility to upgrade models over time.” Twizer also previously worked at AWS as a principal solutions architect and leader of the go-to-market generative AI sales architecture. Key Benefits and Real-World Impact In benchmarking tests, xpander.ai demonstrated that AGS, paired with its Agentic Interfaces, enabled AI agents to achieve a 98% success rate in multi-step tasks, compared to just 24% for agents using traditional methods. These agents completed workflows 38% faster and with 31.5% fewer tokens, underscoring AGS’s ability to reduce costs and improve performance. One real-world example of AGS in action involved a benchmarking task where an AI agent had to research companies across platforms like LinkedIn and Crunchbase, then organize the results in Notion. AGS streamlined the process, ensuring tools were used in the correct sequence and schemas were consistently followed. “We provide a complete AI agent that can create an interface to any system,” Twizer added. “The data interface, for the first time, is native to AI, addressing a major pain point the world is struggling with.” AGS’s role in agentic AI xpander.ai positions AGS as a vital step in the evolution of agentic AI, enabling tools like Nvidia NIM microservices to integrate more seamlessly with enterprise systems. “AI agents will need to use APIs for synchronous use cases involving complex data structures, where traditional UIs just aren’t enough,” Sheinberg noted. Through AGS, xpander.ai transforms how AI agents handle error management and context continuity. By embedding fallback options directly within its graph structures, AGS allows agents to retry failed operations or pivot to alternative workflows without human intervention, preserving task stability. This level of reliability ensures that AGS-equipped agents are not just reactive but adaptive, capable of tackling even the most unpredictable workflows. Building the future of AI workflows xpander.ai’s introduction of AGS, coupled with its Agentic Interfaces, represents a significant leap forward for multi-step AI agents. By enabling structured, adaptive workflows and streamlining complex API interactions, AGS sets a new standard for reliability and efficiency in automation. As the company continues to grow, its tools promise to empower businesses to harness the full potential of AI-driven workflows. source

Join our daily and weekly newsletters for the latest updates and exclusive content on industry-leading AI coverage. Learn More It’s anticipated that by next year, more than 50% of enterprise data will be created and processed outside traditional data centers or clouds. In this age of AI, enterprises need to be able to quickly access and extract value from that edge data — but it can be time-consuming and complicated to do so, and many enterprise leaders are still operating with a cloud mindset.  To help reduce complexity and make edge computing more accessible, Dell is today announcing new advancements to its Dell NativeEdge edge operations software platform. The offering aims to help simplify how enterprises deploy, scale and use AI across numerous types of edge environments.  “The edge is a place where there are a lot of opportunities but a lot of silos and challenges,” Pierluca Chiodelli, Dell’s VP of edge product management, told VentureBeat. “We wanted to create a platform to democratize the edge.”  Supporting inferencing, offering multi-node capabilities  At the far edge, most deployments are on a single node, Chiodelli explained. But this can present challenges when it comes to reliability, scalability and computing power, not to mention cost.  Dell NativeEdge, which is part of the company’s AI Factory, provides multi-node high-availability capabilities. This means that endpoints can be grouped together to act like a single system, allowing enterprises to maintain business processes and edge AI workloads even when there are network disruptions or device failures. It also supports virtual machine (VM) migration — the process of moving a machine from one physical location to another without disrupting availability or performance — as well as automatic app, compute and storage failover.  AI inferencing is increasingly important at the edge, but it can be tedious and time-consuming to deploy AI across “hundreds, if not thousands,” of edge locations, Chiodelli pointed out. To address this, Dell is now offering a catalog of more than 55 pre-built blueprints that automate AI deployment.  The new catalog includes several popular open-source tools as well as a data collector that transfers data from sensors and IoT devices and Geti-branded software that can accelerate the development of computer vision AI models at the edge.  Chiodelli explained that Dell NativeEdge is consumption-based, and customers pay per each device under management (whether that be a small gateway or a large server). Zero touch, zero trust, adapting with fast-moving AI Chiodelli pointed out that it is important that users have the ability to adapt to changing workload demands across broad environments; they must also be able to adjust on the fly.  “With AI, everything changes everyday,” said Chiodelli. Human users need to not just be able to intervene on day 0 (inception) but also on day 2 (management) to keep up. Zero touch is important to all this because “You don’t need to have IT people going to different locations and trying to deploy things,” said Chiodelli.  Security is also paramount; Dell NativeEdge is built on a zero-trust model, and the platform continually monitors the security of edge infrastructure, Chiodelli exaplained. “You really need zero trust because you are in the land of nowhere, you cannot trust anybody,” he said.  Dell NativeEdge has been deployed by customers across numerous industries. French-headquartered multinational IT company Atos, for instance, used the platform to create Atos business-outcomes-as-a-service (BOaaS). The edge management tool works with AI and machine learning (ML) and helps customers deploy, automate and optimize their edge environments through a single dashboard.  As one example, BOaaS has allowed Atos’s manufacturing customers to see measurable business improvements as the result of predictive maintenance. This in turn has helped them reduce downtime, decrease costs and optimize production.  Another customer is Ontario-based Nature Fresh Farms. While most wouldn’t necessarily consider farms to be all that IT-savvy, the family-owned company has been using edge computing to support yield optimization and to perform real-time environmental monitoring.  Previously, “they had a lot of solutions that were very siloed,” Chiodelli explained. It was a challenge to look at the entire estate and manage updates. “Dell NativeEdge helps us monitor real-time infrastructure elements, ensuring optimal conditions for our produce, and receive comprehensive insights into our produce packaging operations,” said Keith Bradley, VP for IT. In other cases, Dell NativeEdge has been used to perform preventative maintenance of amusement parks and to inspect railways and train tracks, Chiodelli noted. Other companies using the platform include GE, EY, AIShield and Nozomi Networks.  “AI is accelerating new edge workloads and opportunities at an unprecedented rate, and organizations across industries are asking for simpler and more reliable ways to use AI at the edge,” said Gil Shneorson, SVP for Dell’s solutions and platforms. source

Join our daily and weekly newsletters for the latest updates and exclusive content on industry-leading AI coverage. Learn More There is a lot of hype and a lot of promise surrounding AI and what it can bring to an enterprise. For enterprise software vendor OpenText, moving beyond the hype to be able to actually demonstrate the value of AI is now an everyday mission. Today, OpenText announced its latest set of platform updates with the Cloud Editions (CE) 24.4 release. OpenText has an expansive portfolio of enterprise software that include content management, DevOps, messaging, database and security. The OpenText software portfolio also includes enterprise software gained from the $6 billion 2023 acquisition of MicroFocus, which includes the Vertica database and Autonomy IDOL platform. A year ago, the OpenText Aviator technology debuted as part of the Cloud Editions (CE) 23.4 release, providing integrated generative AI capabilities to parts of the OpenText portfolio. With the new update OpenText is providing more Aviator capabilities with deeper integrations as well as AI agent functionality across its portfolio. The company’s expanded AI portfolio, now includes 15 products supported by 102 autonomous agents. The overall goal isn’t to deploy AI for AI’s sake but rather to deliver measurable efficiency gains for enterprise customers. “The biggest ask from customers, I can tell you, it’s almost 95% of the customers I’ve talked to in the past six to eight months have asked for one thing, show me the return on investment,” Muhi Majzoub, EVP and Chief Product Officer at OpenText, told VentureBeat. How OpenText is demonstrating the ROI of enterprise AI In order to help demonstrate the potential value and return on investment from an AI deployment, OpenText has rolled out a program it calls – Earn Your Wings. Majzoub said that the program allows organizations to test OpenText Aviator technology in a controlled environment before committing to a full deployment. The program is a 30-day proof of concept exercise. During this 30-day period, OpenText’s AI experts work closely with the enterprise to deploy a model in the organization’s preferred environment, whether it’s a private cloud, on-premises, or in the OpenText cloud. OpenText guarantees that the model is trained solely on the customer’s data, ensuring the relevance and accuracy of the AI-powered insights. By the end of the 30-day period, the customer can evaluate the tangible benefits of the Aviator platform, such as improved efficiency, faster response times and enhanced productivity. Majzoub said that this hands-on experience allows enterprises to clearly demonstrate the ROI of their AI investments, addressing a critical concern and paving the way for broader adoption. At the OpenText World 2024 event this week, Majzoub said that he will be joined on the keynote stage by numerous customers including German automaker BMW, that have benefited from the Aviator AI capabilities. Majzoub also highlighted the work his company has been doing with global life sciences vendor Catalent, which uses OpenText’s life sciences solution for manufacturing and quality testing. He noted that Catalent might have over a million documents in a clinical trial procedure. That volume of data is not something humans can realistically sift through in a reasonable amount of time, without a lot of people and time. “The Aviator can sift through those and identify the 20 that a doctor should go through in detail,” Majzoub said. Expanding the Aviator AI portfolio At the heart of OpenText’s AI strategy is the continued expansion of its Aviator platform. The company is announcing several new Aviator-powered products and enhancements that showcase the breadth and depth of its AI capabilities. One of the notable additions is the Aviator for XDR (Extended Detection and Response), which Majzoub described as a new platform for managed service providers (MSPs) in the security space.  “It sifts through log files for security that comes from tools like CrowdStrike, LogRhythm, SentinelOne, or it could come from our own ArcSight platform,” he explained. “It sifts through all of these and identifies for you either internal anomalies or external anomalies.” In the content management domain, OpenText is showcasing the integration of Aviator into its internal “Ollie.ai” platform, where the company stores all of its own documents. The Aviator is being used to assist sales teams in generating RFP responses, as well as to support technical analysts in providing faster and more accurate resolutions to customer queries. Majzoub also highlighted the use of Aviator in OpenText’s DevSecOps platform, where the AI agent is being used to automatically document code, freeing up technical writers to focus on creating more engaging user documentation and training materials. Where OpenText is bringing agentic AI to improve enterprise ROI A key component of OpenText’s growing AI capabilities is the integration of agentic AI into the Aviator platform. The AI agents are integrated into OpenText solutions, providing specialized capabilities tailored to the needs of different industries and use cases. “In every business unit, the agents have different meanings and different use cases,” Majzoub said. For example, in the security domain, he said that the agents can be embedded into facial recognition and license plate identification systems to assist law enforcement agencies. In the content management space, the agents can analyze large document repositories, summarizing key information and highlighting critical insights that would be difficult for human analysts to uncover. In the supply chain domain, Majzoub said that the AI agents can map transactions and data flows between different enterprise systems, providing visibility and transparency into complex global supply chains. How OpenText aims to differentiate in a crowded enterprise AI market In the enterprise software space, every vendor today has some degree of generative AI. How OpenText is aiming to differentiate is in several different ways. The platform across the OpenText portfolio, supports multi-cloud deployments. Majzoub also emphasized that OpenText is commonly used alongside other enterprise software and has multiple levels of integrations and partnerships. “The partnerships we have developed with great companies like SAP, Microsoft and Google, allows us to embed our platform in these solutions like no other vendor can deliver today,”  Majzoub claimed.

Join our daily and weekly newsletters for the latest updates and exclusive content on industry-leading AI coverage. Learn More Leading figures in AI, including Anthropic’s Dario Amodei and OpenAI’s Sam Altman, suggest that “powerful AI” or even superintelligence could appear within the next two to 10 years, potentially reshaping our world. In his recent essay Machines of Loving Grace, Amodei provides a thoughtful exploration of AI’s potential, suggesting that powerful AI — what others have termed artificial general intelligence (AGI) — could be achieved as early as 2026. Meanwhile, in The Intelligence Age, Altman writes that “it is possible that we will have superintelligence in a few thousand days,” (or by 2034). If they are correct, sometime in the next two to 10 years, the world will dramatically change. As leaders in AI research and development, Amodei and Altman are at the forefront of pushing boundaries for what is possible, making their insights particularly influential as we look to the future. Amodei defines powerful AI as “smarter than a Nobel Prize winner across most relevant fields — biology, programming, math, engineering, writing…” Altman does not explicitly define superintelligence in his essay, although it is understood to be AI systems that surpass human intellectual capabilities across all domains.  Not everyone shares this optimistic timeline, although these less sanguine viewpoints have not dampened enthusiasm among tech leaders. For example, OpenAI co-founder Ilya Sutskever is now a co-founder of Safe Superintelligence (SSI), a startup dedicated to advancing AI with a safety-first approach. When announcing SSI last June, Sutskever said: “We will pursue safe superintelligence in a straight shot, with one focus, one goal and one product.” Speaking about AI advances a year ago when still at OpenAI, he noted: “It’s going to be monumental, earth-shattering. There will be a before and an after.” In his new capacity at SSI, Sutskever has already raised a billion dollars to fund company efforts. These forecasts align with Elon Musk’s estimate that AI will outperform all of humanity by 2029. Musk recently said that AI would be able to do anything any human can do within the next year or two. He added that AI would be able to do what all humans combined can do in a further three years, in 2028 or 2029. These predictions are also consistent with the long-standing view from futurist Ray Kurzweil that AGI would be achieved by 2029. Kurzweil made this prediction as far back as 1995 and wrote about this in this best-selling 2005 book, “The Singularity Is Near.”  Futurist Ray Kurzweil stands by his prediction of AGI by 2029. The imminent transformation As we are on the brink of these potential breakthroughs, we need to assess whether we are truly ready for this transformation. Ready or not, if these predictions are right, a fundamentally new world will soon arrive.  A child born today could enter kindergarten in a world transformed by AGI. Will AI caregivers be far behind? Suddenly, the futuristic vision from Kazuo Ishiguro in “Klara and the Sun” of an android artificial friend for those children when they reach their teenage years does not seem so farfetched. The prospect of AI companions and caregivers suggests a world with profound ethical and societal shifts, one that might challenge our existing frameworks. Beyond companions and caregivers, the implications of these technologies are unprecedented in human history, offering both revolutionary promise and existential risk. The potential upsides that could come from powerful AI are profound. Beyond robotic advances this could include developing cures for cancer and depression to finally achieving fusion energy. Some see this coming epoch as an era of abundance with people having new opportunities for creativity and connection. However, the plausible downsides are equally momentous, from vast unemployment and income inequality to runaway autonomous weapons.  In the near term, MIT Sloan principal research scientist Andrew McAfee sees AI as enhancing rather than replacing human jobs. On a recent Pivot podcast, he argued that AI provides “an army of clerks, colleagues and coaches” available on demand, even as it sometimes takes on “big chunks” of jobs.  But this measured view of AI’s impact may have an end date. Elon Musk said that in the longer term, “probably none of us will have a job.” This stark contrast highlights a crucial point: Whatever seems true about AI’s capabilities and impacts in 2024 may be radically different in the AGI world that could be just several years away. Tempering expectations: Balancing optimism with reality Despite these ambitious forecasts, not everyone agrees that powerful AI is on the near horizon or that its effects will be so straightforward. Deep learning skeptic Gary Marcus has been warning for some time that the current AI technologies are not capable of AGI, arguing that the technology lacks the needed deep reasoning skills. He famously took aim at Musk’s recent prediction of AI soon being smarter than any human and offered $1 million to prove him wrong. Linus Torvalds, creator and lead developer of the Linux operating system, said recently that he thought AI would change the world but currently is “90% marketing and 10% reality.” He suggested that for now, AI may be more hype than substance. Perhaps lending credence to Torvald’s assertion is a new paper from OpenAI that shows their leading frontier large language models (LLM) including GPT-4o and o1 struggling to answer simple questions for which there are factual answers. The paper describes a new “SimpleQA” benchmark “to measure the factuality of language models.” The best performer is o1-preview, but it produced incorrect answers to half of the questions.  Performance of frontier LLMs on new SimpleQA benchmark from OpenAI. Source: Introducing SimpleQA. Looking ahead: Readiness for the AI era Optimistic predictions about the potential of AI contrast with the technology’s present state as shown in benchmarks like SimpleQA. These limitations suggest that while the field is progressing quickly, some significant breakthroughs are needed to achieve true AGI.  Nevertheless, those closest to the developing AI technology foresee rapid advancement. On a

Join our daily and weekly newsletters for the latest updates and exclusive content on industry-leading AI coverage. Learn More Come the new year, the incoming Trump administration is expected to make many changes to existing policies, and AI regulation will not be exempt. This will likely include repealing an AI executive order by current President Joe Biden. The Biden order established government oversight offices and encouraged model developers to implement safety standards. While the Biden AI executive order rules focus on model developers, its repeal could present some challenges for enterprises to overcome. Some companies, like Trump-ally Elon Musk’s xAI, could benefit from a repeal of the order, while others are expected to face some issues. This could include having to deal with a patchwork of regulations, less open sharing of data sources, less government-funded research and more emphasis on voluntary responsible AI programs.  Patchwork of local rules Before the EO’s signing, policymakers held several listening tours and hearings with industry leaders to determine how best to regulate technology appropriately. Under the Democratic-controlled Senate, there was a strong possibility AI regulations could move forward, but insiders believe the appetite for federal rules around AI has cooled significantly.  Gaurab Bansal, executive director of Responsible Innovation Labs, said during the ScaleUp: AI conference in New York that the lack of federal oversight of AI could lead states to write their policies.  “There’s a sense that both parties in Congress will not be regulating AI, so it will be states who may run the same playbook as California’s SB 1047,” Bansal said. “Enterprises need standards for consistency, but it’s going to be bad when there’s a patchwork of standards in different areas.”  California state legislators pushed SB 1047 — which would have mandated a “kill switch” to models among other government controls — with the bill landing on Gov. Gavin Newsom’s desk. Newsom’s veto of the bill was celebrated by industry luminaries like Meta’s Yann Le Cunn. Bansal said states are more likely to pass similar bills.  Dean Ball, a research fellow at George Mason University’s Mercatus Center, said companies may have difficulty navigating different regulations.  “Those laws may well create complex compliance regimes and a patchwork of laws for both AI developers and companies hoping to use AI; how a Republican Congress will respond to this potential challenge is unclear,” Ball said.  Voluntary responsible AI  Industry-led responsible AI has always existed. However, the burden on companies to be more proactive in being accountable and fair may heighten because their customers demand a focus on safety. Model developers and enterprise users should spend time implementing responsible AI policies and building standards that meet laws like the European Union’s AI Act.  During the ScaleUp: AI conference, Microsoft Chief Product Officer for Responsible AI Sarah Bird said many developers and their customers, including Microsoft, are readying their systems for the EU’s AI act.  But even if no sprawling law governs AI, Bird said it’s always good practice to bake responsible AI and safety into the models and applications from the onset.  “This will be helpful for start-ups, a lot of the high level of what the AI act is asking you to do is just good sense,” Bird said. “If you’re building models, you should govern the data going into them; you should test them. For smaller organizations, compliance becomes easier if you’re doing it from scratch, so invest in a solution that will govern your data as it grows.” However, understanding what is in the data used to train large language models (LLMs) that enterprises use might be harder. Jason Corso, a professor of robotics at the University of Michigan and a co-founder of computer vision company Voxel51, told VentureBeat the Biden EO encouraged a lot of openness from model developers.  “We can’t fully know the impact of one sample on a model that presents a high degree of potential bias risk, right? So model users’ businesses could be at stake if there’s no governance around the use of these models and the data that went in,” Corso said. Fewer research dollars  AI companies enjoy significant investor interest right now. However, the government has often supported research that some investors feel is too risky. Corso noted that the new Trump administration might choose not to invest in AI research to save on costs.  “I just worry about not having the government resources to put it behind those types of high-risk, early-stage projects,” Corso said. However, a new administration does not mean money will not be allocated to AI. While it’s unclear if the Trump administration will abolish the newly created AI Safety Institute and other AI oversight offices, the Biden administration did guarantee budgets until 2025. “A pending question that must color Trump’s replacement for the Biden EO is how to organize the authorities and allocate the dollars appropriated under the AI Initiative Act. This bill is the source for many of the authorities and activities Biden has tasked to agencies such as NIST and funding is set to continue in 2025. With these dollars already allocated, many activities will likely continue in some form. What that form looks like, however, has yet to be revealed,” Mercatus Center research fellow Matt Mittelsteadt said.  We’ll know how the next administration sees AI policy in January, but enterprises should prepare for whatever comes next.  source

Join our daily and weekly newsletters for the latest updates and exclusive content on industry-leading AI coverage. Learn More Microsoft has quietly built the largest enterprise AI agent ecosystem, with over 100,000 organizations creating or editing AI agents through its Copilot Studio since launch – a milestone that positions the company ahead in one of enterprise tech’s most closely watched and exciting  segments. “That’s a lot faster than we thought, and it’s a lot faster than any other kind of cutting edge technology we’ve released,” Charles Lamanna, Microsoft’s executive responsible for the company’s agent vision, told VentureBeat. “And that was like a 2x growth in just a quarter.” The rapid adoption comes as Microsoft significantly expands its agent capabilities. At its Ignite conference starting today, the company announced it will allow enterprises to use any of the 1,800 large language models (LLMs) in the Azure catalog within these agents – a significant move beyond its exclusive reliance on OpenAI’s models. The company also unveiled autonomous agents that can work independently, detecting events and orchestrating complex workflows with minimal human oversight. (See our full coverage of today’s Microsoft’s agent announcements here.)  These AI agents – software that can reason and perform specific business tasks using generative AI – are emerging as a powerful tool for enterprise automation and productivity. Microsoft’s platform enables organizations to build these agents for tasks ranging from customer service to complex business process automation, while maintaining enterprise-grade security and governance. Building an enterprise-grade foundation Microsoft’s early lead in AI agents stems from its focus on enterprise requirements that often get overlooked in the AI hype cycle. While its new autonomous agents and LLM flexibility grabbed headlines at Ignite, the company’s real advantage lies in its enterprise infrastructure. The platform integrates with over 1,400 enterprise systems and data sources, from SAP to ServiceNow to SQL databases. This extensive connectivity lets organizations build agents that can access and act on data across their existing IT landscape. While enterprises can build custom agents from scratch, Microsoft has also launched ten pre-built autonomous agents targeting core business functions like sales, service, finance, and supply chain – to accelerate adoption for common enterprise use cases. The company did not provide any more detail about which types of agents customers are finding the most popular. But Lamanna said that aside from apps that IT departments are building for specific core tasks, there was a second category of apps that is more bottoms-up. This is where employees create Copilot agents to share their documents or presentations with their team or other partners, so that others can interact with the content and ask questions about it.  Security and governance features, often afterthoughts in AI deployments, are built into Microsoft’s core architecture. The platform’s control system ensures agents operate within enterprise permissions and data governance frameworks. “We think it will show up everywhere,” Lamanna told VentureBeat, “because whenever you have a technology that makes something possible that was previously impossible, all of you kind of are always shocked by how broadly it ends up being used.” He compared it with the Internet, where connectivity extended from the browser to the operating system, and fundamentally changed client-server architecture.  The LLM made a big breakthrough, Lamanna explains, in that it understands unstructured content – language or video or audio – and has shown the beginnings of reasoning, to make conclusions or judgments based on this data, Lamanna said. “So the browser, word processor, the core operating system experience, and the way you do sales processes and customer support processes – they all have to be reevaluated now that this capability exists…I don’t think there’ll be really any part of the stack in computing that doesn’t have some component reimagined as a result of all the agent and AI capabilities.” Early adopters are already seeing results. McKinsey reduced its project intake workflows from 20 days to just 2 days using automated routing agents. Pets at Home deployed fraud prevention agents in under two weeks, saving millions annually. Other companies using Copilot Studio include Nsure, McKinsey, Standard Bank, Thomson Reuters, Virgin Money, Clifford Chance and Zurich, Microsoft told VentureBeat. The Agent mesh: Microsoft’s vision for enterprise AI At the heart of Microsoft’s strategy is what Lamanna calls the “agent mesh” – an interconnected system where AI agents collaborate to solve complex problems. Rather than operating in isolation, agents can pass tasks, messages, and knowledge seamlessly across the enterprise. Copilot Studio has been associated so far with agents that are triggered via chat, but now Microsoft is emphasizing any kind of actions. Imagine an enterprise where agents collaborate seamlessly: A sales agent triggers an inventory agent to check stock availability, which then notifies a customer service agent to update the client. This architecture includes: Autonomous agents that detect events and trigger actions without human oversight An orchestration layer that coordinates multiple specialized agents Real-time monitoring tools that provide transparency into agent workflows Microsoft’s research arm recently released the Magnetic-One system based on the company’s Autogen framework, which establishes a sophisticated agent hierarchy: a managing agent maintains task checklists in an “outer loop” while specialized agents execute work in an “inner loop.” This architecture could potentially soon embrace tools like Microsoft’s OmniParser that let agents interpret UI elements, and showcases Microsoft’s technical lead in computer-using agents — matching capabilities being developed by Anthropic and Google. The company said it is working to bring this research into production, but did not specify how and when. Image: Microsoft Research’s Magentic-One multi-agent system, aims to solve open-ended web and file-based tasks, using two loops, and outer loop and an inner loop.   Microsoft’s approach addresses a key enterprise challenge: scaling from hundreds to potentially millions of agents while maintaining control. The platform enables companies to coordinate multiple specialized agents through its orchestration capabilities – an approach that aligns with a broader industry trend toward multi-agent systems. The platform’s pricing model reflects this enterprise focus. Rather than charging per token like most AI providers, Microsoft

Join our daily and weekly newsletters for the latest updates and exclusive content on industry-leading AI coverage. Learn More AI agents are the talk of the enterprise right now. But, business leaders want to hear about tangible results and relevant use cases — as opposed to futuristic, not-quite-there-yet scenarios — and demand tools that are easy to deploy and use and, further, that support their preferred model(s).  Microsoft claims to have all these concerns covered with new no-code and low-code capabilities in Microsoft 365 Copilot. Today at Microsoft Ignite, the tech giant announced that users can now build their own custom autonomous agents or deploy out-of-the-box, purpose-built agents. And, they can do this via a bring-your-own setup that provides them access to the 1,800-plus models in the Azure AI catalog. (See our separate story today about how Microsoft has quietly assembled the largest AI agent ecosystem — and no one else is close). “Companies have done a lot of AI exploration and really want to be able to measure and understand how agents can help them be more efficient, improve performance and decrease cost and risk,” Lili Cheng, corporate VP of the Microsoft AI and research division, told VentureBeat. “They’re really leaning into scaling out their copilots.” Supporting bring-your-own-knowledge, bring-your-own-model According to IDC, in the next 24 months, more and more companies will build custom, tailored AI tools. Indeed, vendors — from tech giants such Salesforce and Snowflake to smaller players like CrewAI and Sema4.ai — are increasingly pushing platforms to market that promise to revolutionize enterprise operations.  Microsoft introduced Copilot in February 2023, and has now infused it with a suite of new capabilities to support agentic AI. Autonomous capabilities now in public preview allow users to build agents that act on their behalf without additional prompting. This means agents can work and act in the background without human oversight.  Users can use templates for common scenarios (such as sales order and deal accelerator agents) in Copilot Studio. Or, more advanced developers can take advantage of a new Agent SDK (now available in preview) to build full-stack, multichannel agents that integrate with various Microsoft services and can be deployed across Microsoft, third-party and web channels.  New integrations with Azure AI Foundry will support bring-your-own-knowledge (custom search indices can be added as a knowledge source) (now in preview) and bring-your-own- model (now in private preview). This will allow users to pull from the 1,800-some-odd models (and counting) in Azure’s catalog.  This element is critical, as users are demanding the ability to securely use proprietary data and combine and test different models without getting locked in to one or the other. “People want a variety of models, they want to be able to fine-tune models,” said Cheng.  Ready-made agents for HR, translation, project management But not all tasks require a custom solution; already-built models can be useful across enterprises. Microsoft is releasing several ready-made agents in Copilot that can handle simple, repetitive tasks or more complex multi-step processes. These include:  Agents in SharePoint, which allows users to create their own tailored agents that they can give names and personalize. Users can ask questions and receive real-time answers and share agents across emails, meetings and chats. Microsoft emphasizes that agents follow existing SharePoint user permissions and sensitivity labels to help ensure that sensitive information isn’t overshared. Employee self-service agent, which answers common workplace policy-related questions and takes action on HR and IT-related tasks. For instance, employees can retrieve benefits and payroll information, request a new device or start a leave of absence form.  Facilitator agent, which takes real-time notes in Teams and chats and provides a summary of important information as the conversation is unfolding.  Interpreter agent, which provides real-time translation in Teams meetings in up to nine languages. Participants can also have Interpreter simulate their voice. Project Manager agent, which automates processes in Planner, handling projects from creation to execution. The agent can automatically create new plans from scratch or use templates; it then assigns tasks, tracks progress, sends notifications and provides status reports.  Further, a new Azure AI Foundry SDK offers a simplified coding experience and toolchain for developers to customize, test, deploy and manage agents. Users can choose from 25 pre-built templates, integrate Azure AI into their apps and access common tools including GitHub or Copilot Studio.  Cheng pointed to the importance of low-code and no-code tools, as enterprises want to accommodate teams with a range of skills. “Most companies don’t have big AI teams or even development teams,” she said. “They want more people to be able to author their copilots.” The goal is to greatly simplify the agent-building process so that enterprises “build something once and use it wherever their customers are,” she said. Tooling should be simple and easy to use so that app creators don’t even know if things are getting ever more complicated on the back end. Cheng posited: “Something might be more difficult, but you don’t know it’s more difficult, you just want to get your job done.” McKinsey, Thomson Reuters use cases Initial use cases have revolved around support, such as managing IT help desks, as well as HR scenarios including onboarding, said Cheng.  McKinsey & Company, for its part, is working with Microsoft on an agent that will speed up client onboarding. A pilot showed that lead time could be reduced by 90% and administrative work by 30%. The agent can identify expert capabilities and staffing teams and serves as a platform for colleagues to ask questions and request follow-ups.  Meanwhile, Thomson Reuters built an agent to help make the legal due diligence process — which requires significant expertise and specialized content — more efficient. The platform combines knowledge, skills and advanced reasoning from the firm’s gen AI tool CoCounsel to help lawyers close deals more quickly and efficiently. Early tests indicate that several tasks in these workflows could be cut by at least 50%.  “We really see people combining more traditional copilots — where you have AI augmenting people skills

Join our daily and weekly newsletters for the latest updates and exclusive content on industry-leading AI coverage. Learn More 2025 will undoubtedly be the year AI agents get real. Many early entrants to the market, though, either tend to be singularly-purposed and less flexible, or more horizontal yet IT and developer-driven (and thus not always business user friendly).  Startup Sema4.ai says it has the differentiating factor that future-thinking enterprises need: The company has put a “tremendous amount of intelligence” into its platform to make it suitable for a wide variety of business use cases.  “We think it’s much better to have a horizontal platform that enterprises can build their agents for, versus coming in with a single purpose,” Rob Bearden, Sema4.ai co-founder and CEO, told VentureBeat.  Today Sema4.ai is announcing the general availability of its full-stack enterprise AI agent platform. In less than 9 months, the startup has come out of stealth, piloted its platform with six of the Fortune 2000, secured $30.5 million in funding and acquired open-source automation company Robocorp. And, it has already been featured in two Gartner hype cycles.  “Agents are going to drive the biggest transformation in business models and efficiencies that the enterprise has seen since the launch of the internet,” said Bearden.  AI agents outside DevOps and IT teams Sema4.ai’s no-code agent platform was designed to “speak industry language” and integrate with existing business processes and applications. It has seven key components:  Studio: Users can quickly build, test and deploy AI agents. Runbooks: Users can build and maintain agents with natural language runbooks and pre-built actions. Control Room: Features complete lifecycle management as well as security and scalability.  Actions: An automation framework that allows agents to integrate with apps including SharePoint, SAP and APIs using automation-as-code and Python. Work Room: Users can find, work with and supervise enterprise agents. Document Intelligence: Provides accurate document interpretation. Dynamic Data Access: Gives agents zero-copy access to past, present and future data. It is critical to shift the current operating model from “programmatically driven by DevOps and IT” to the business user, Bearden emphasized. This is because business users deeply understand specific processes and procedures and best practice outcomes, as well as potential problems and remediation methods.  In Sema4.ai, business users can define parameters and expected outcomes in runbooks that calibrate AI; agents, possessing an understanding of the data they need and best reasoning paths, then construct automations and software development kits (SDKs).  “It’s all guided by the business user in natural language,” said Bearden, the former CEO of data platform company Cloudera. “Agents will disintermediate the legacy ERP applications and even the SaaS applications. They will put the power into the hands of the business user versus the DevOps and IT teams.” Sema4.ai’s platform is architected to be interoperable with whatever large language model (LLM) is most cost-effective for the enterprise use case — currently including Claude, OpenAI, Azure and Bedrock, but that will be expanded, Bearden explained.  “Bring your own LLM, we’ll make sure that we interoperate with it at the highest standard,” he said.  Use case: Koch invoice reconciliation Customers have used Sema4.ai’s platform for a range of use cases — from simple scenarios requiring just one agent for a specific use case, to “15, 18, 20-plus” working collaboratively to manage entire business processes, Bearden explained. Agents (at least for now) are best in areas where work is procedural, high volume, human-intensive, understood, measurable and has definitive outcomes.  “It tends to be high ROI kind of work,” said Bearden. “It’s measurable. It’s auditable.” Six Fortune 2000 companies are piloting the platform in early proof-of-concept (PoC). Bearden explained that these partners are using agents to automate invoice processing, payment reconciliation, employee onboarding and regulatory compliance. In two of the PoCs, Sema4.ai’s platform is autonomously performing more than 80% of knowledge work tasks.  One early adopter is industrial giant Koch, which is using agents to automate one of its invoice reconciliation processes, Koch Labs director Tanner Gonzalez told VentureBeat. Previously, he explained, this involved manually reviewing invoices that can be 80 pages or longer. Sema4.ai allows them to use natural language processing (NLP) to create automated workflows that extract relevant data and validate invoices.  The key benefit of the platform is that it provides an easy-to-maintain, document-like interface for building and updating gen AI workflows. “Compared to previous robotic process automation tools we’ve used, Sema4.ai is much more user-friendly and doesn’t require specialized technical skills to manage over time,” said Gonzalez.  Using natural language, employees — finance analysts, accountants, operations engineers or other non-technical individuals — interact with the platform similar to how they would describe their workflow in a Word document, “explaining their logic and the tasks they complete again and again,” Gonzalez explained. In more complex use cases, the platform provides capabilities for data scientists to deploy custom AI models, and for data engineers to connect new data sources for read and write functions.  Looking ahead, Koch sees potential to expand the use of the platform to other areas such as market research analysis or external communications for commercial teams, said Gonzalez. “The flexibility and low-code nature of the platform makes it well-suited to tackle a variety of automation and conversational AI use cases across our organization,” he said.  A horizontal approach to address a variety of business needs When looking to adopt AI agents, Koch analyzed many alternatives in the market, Gonzalez noted. They found others to be too narrowly focused on specific industries, building their own foundation models or limited on integrations.  The key highlights for Sema4.ai, he said, are 1.) flexibility, “meaning we’re not tied to a specific model as new ones emerge”; 2.) ease of use for business users that can write out their steps as opposed to coding or learning a new tool; and 3.) the ability to implement closed-loop automation, driving real agent automation and monitoring progress periodically for new anomalies. Navin Chaddha, managing partner at Mayfield Fund, one of Sema4.ai’s top backers, said the startup is