VentureBeat

Want smarter insights in your inbox? Sign up for our weekly newsletters to get only what matters to enterprise AI, data, and security leaders. Subscribe Now Dust, a two-year-old artificial intelligence platform that helps enterprises build AI agents capable of completing entire business workflows, has reached $6 million in annual revenue — a six-fold increase from $1 million just one year ago. The company’s rapid growth signals a shift in enterprise AI adoption from simple chatbots toward sophisticated systems that can take concrete actions across business applications. The San Francisco-based startup announced Thursday that it has been selected as part of Anthropic’s “Powered by Claude” ecosystem, highlighting a new category of AI companies building specialized enterprise tools on top of frontier language models rather than developing their own AI systems from scratch. “Users want more than just conversational interfaces,” said Gabriel Hubert, CEO and co-founder of Dust, in an interview with VentureBeat. “Instead of generating a draft, they want to create the actual document automatically. Rather than getting meeting summaries, they need CRM records updated without manual intervention.” Dust’s platform goes far beyond the chatbot-style AI tools that dominated early enterprise adoption. Instead of simply answering questions, Dust’s AI agents can automatically create GitHub issues, schedule calendar meetings, update customer records, and even push code reviews based on internal coding standards–all while maintaining enterprise-grade security protocols. How AI agents turn sales calls into automated GitHub tickets and CRM updates The company’s approach becomes clear through a concrete example Hubert described: a business-to-business sales company using multiple Dust agents to process sales call transcripts. One agent analyzes which sales arguments resonated with prospects and automatically updates battle cards in Salesforce. Simultaneously, another agent identifies customer feature requests, maps them to the product roadmap, and in some cases, automatically generates GitHub tickets for small features deemed ready for development. “Each call transcript is going to be analyzed by multiple agents,” Hubert explained. “You’ll have a sales battle card optimizer agent that’s going to look at the arguments the salesperson made, which ones were powerful and seem to resonate with the prospect, and that’s going to go and feed into a process on the Salesforce side.” This level of automation is enabled by the Model Context Protocol (MCP), a new standard developed by Anthropic that allows AI systems to securely connect with external data sources and applications. Guillaume Princen, Head of EMEA at Anthropic, described MCP as “like a USB-C connector between AI models and apps,” enabling agents to access company data while maintaining security boundaries. Why Claude and MCP are powering the next wave of enterprise AI automation Dust’s success reflects broader changes in how enterprises are approaching AI implementation. Rather than building custom models, companies like Dust are leveraging increasingly capable foundation models — particularly Anthropic’s Claude 4 suite — and combining them with specialized orchestration software. “We just want to give our customers access to the best models,” Hubert said. “And I think right now, Anthropic is early in the lead, especially on coding related models.” The company charges customers $40-50 per user per month and serves thousands of workspaces ranging from small startups to large enterprises with thousands of employees. Anthropic’s Claude models have seen particularly strong adoption for coding tasks, with the company reporting 300% growth in Claude Code usage over the past four weeks following the release of its latest Claude 4 models. “Opus 4 is the most powerful model for coding in the world,” Princen noted. “We were already leading the coding race. We’re reinforcing that.” Enterprise security gets complex when AI agents can actually take action The shift toward AI agents that can take real actions across business systems introduces new security complexities that didn’t exist with simple chatbot implementations. Dust addresses this through what Hubert calls a “native permissioning layer” that separates data access rights from agent usage rights. “Permission creation, as well as data & tool management is part of the onboarding experience to mitigate sensitive data exposure when AI agents operate across multiple business systems,” the company explains in technical documentation. This becomes critical when agents have the ability to create GitHub issues, update CRM records, or modify documents across an organization’s technology stack. The company implements enterprise-grade infrastructure with Anthropic’s Zero Data Retention policies, ensuring that sensitive business information processed by AI agents isn’t stored by the model provider. This addresses a key concern for enterprises considering AI adoption at scale. The rise of AI-native startups building on foundation models instead of creating their own Dust’s growth is part of what Anthropic calls an emerging ecosystem of “AI native startups”—companies that fundamentally couldn’t exist without advanced AI capabilities. These firms are building businesses not by developing their own AI models, but by creating sophisticated applications on top of existing foundation models. “These companies have a very, very strong sense of what their end customers need and want for that specific use case,” Princen explained. “We’re providing the tools for them to kind of build and adapt their product to those specific customers and use cases they’re looking for.” This approach represents a significant shift in the AI industry’s structure. Instead of every company needing to develop its own AI capabilities, specialized platforms like Dust can provide the orchestration layer that makes powerful AI models useful for specific business applications. What Dust’s $6M revenue growth signals about the future of enterprise software The success of companies like Dust suggests that the enterprise AI market is moving beyond the experimental phase toward practical implementation. Rather than replacing human workers wholesale, these systems are designed to eliminate routine tasks and context-switching between applications, allowing employees to focus on higher-value activities. “By providing universal AI primitives that make all company workflows more intelligent as well as a proper permissioning system, we are setting the foundations for an agent operating system that is future-proof,” Hubert said. The company’s customer base includes organizations convinced that AI will fundamentally change business operations. “The common thread between all

Want smarter insights in your inbox? Sign up for our weekly newsletters to get only what matters to enterprise AI, data, and security leaders. Subscribe Now Inspiration can come from different places, even for architecting and designing agentic systems.  At VB Transform, Capital One explained how it built its agentic platform for its auto business. Milind Naphade, SVP of Technology and Head of AI Foundations at Capital One, said during VB Transform that the company wanted its agents to function similarly to human agents, in that they problem-solve alongside customers.  Naphade said Capital One began designing its agentic offerings 15 months ago, “before agentic became a buzzword.” For Capital One, it was crucial that, in building its agent systems, they learn from how their human agents ask customers for information to identify their problems.  Capital One also looked to another source of organizational structure for its agents: itself.  “We took inspiration from how Capital One itself functions,” Naphade said. “Within Capital One, as I’m sure within other financial services, you have to manage risk, and then there are other entities that you also observe, evaluate, question and audit.” >>See all our Transform 2025 coverage here<< This same structure applies to agents that Capital One wants to monitor. They created an agent that evaluates existing agents, which was trained on Capital One’s policies and regulations. This evaluator agent can kick back the process if it detects a problem. Naphade said to think of it as “a team of experts where each of them has a different expertise and comes together to solve a problem.” Financial services organizations recognize the potential of agents to provide their human agents with information to resolve customer issues, manage customer service, and attract more people to their products. Other banks like BNY have deployed agents this year.  Auto dealership agents Capital One deployed agents to its auto business to assist the bank’s dealership clients in helping their customers find the right car and car loan. Consumers can look at the vehicle inventories of dealerships that are ready for test drives. Naphade said their dealership customers reported a 55% improvement in metrics such as engagement and serious sales leads. “They’re able to generate much better serious leads through this more conversational, natural conversation,” he said. “They can have 24/7 agents working, and if the car breaks down at midnight, the chat is there for you.” Naphade said Capital One would love to bring this type of agent to its travel business, especially for its customer-facing engagements. Capital One, which opened a new lounge in New York’s JFK Airport, offers a very popular credit card for travel points. However, Naphade pointed out that the bank needs to conduct extensive internal testing. Data and models for bank agents Like many enterprises, Capital One has a lot of data for its AI systems, but it has to figure out the best way to bring that context to its agents. It also has to experiment with the best model architecture for its agents.  Naphade and Capital One’s team of applied researchers, engineers and data scientists used methods like model distillation for more efficient architectures. “The understanding agent is the bulk of our cost because that’s the one that has to disambiguate,” he said. “It’s a bigger model, so we try to distribute it down and get a lot of bang for our buck. Then there’s also multi-token prediction and aggregated pre-fill, a lot of interesting ways we can optimize this.” In terms of data, Naphade said his team had undergone several “iterations of experimentation, testing, evaluation, human in the loop and all the right guardrails” before releasing its AI applications.  “But one of the biggest challenges we faced was that we didn’t have any precedents. We couldn’t go and say, oh somebody else did it this way, so we couldn’t ask how it worked out for them?” Naphade said.  Editor’s Note: As a thank-you to our readers, we’ve opened up early bird registration for VB Transform 2026 — just $200. This is where AI ambition meets operational reality, and you’re going to want to be in the room. Reserve your spot now.  source

Nearly two in three C-suite leaders worldwide recognize that sovereignty over their AI and data is a necessity, not a luxury. Are you running scared or running the table? Presented by EDB A recent study conducted by EDB (February 2025) surveyed executive and technology leaders from some of the world’s largest enterprises across EMEA, North America, and Asia Pacific. Between 11% (France) and 27% (Saudi Arabia and the UAE) understood that sovereignty over their AI and data is a mission-critical requirement, right now. And within three years, most leaders across major global economies will come to the same realization. Based on three-year projections from the initial findings, Germany will lead the way (69%), followed by the U.S. and Saudi, while countries including Japan, Italy, and the UK will continue to lag. So the real question is: will you seize control while the window is open — or let competitors define the rules of the game? The drivers of change When asked to define and rank the key drivers behind the convergence of data, AI, and sovereignty, leaders across all three regions provided remarkably consistent responses: The top choice wasn’t geopolitics or global uncertainty — but one clear priority: breaking data out of silos. By a ratio of 2:1, this beat every other reason. When you look at the total choices (top, second, and third) you get the picture below: Most often stated: 46% are moving workloads to hybrid configurations to break free from data silos. Second most often stated: 40% are shifting AI and data workloads to open-source solutions. Third most often stated: 48% are integrating security and infrastructure for a hybrid world.  This tells us something fundamental: The DNA of AI-led enterprises is built on hybrid architectures, open-source adoption, and tightly integrated security and infrastructure. The financial services tipping point In financial services, banking, and insurance, the collision of these variables is even more pronounced. Among executives in these industries, 50% agree that AI and data are converging, sovereignty is critical, and the goal is to deliver Amazon-level experiences for their customers and back office operations. After nearly a decade of phased digital transformations, we are entering a period where rapid shifts in OPEX, CAPEX, and business models are possible. Enterprises are realizing that instead of relying on fragmented solutions, they must become their own sovereign AI and data platforms. This could be presented as the “Amazon aspiration”. The idea that for all your data to work together at the point or place you need it, you have to gain and sustain sovereign control and security. It has to be achieved in very secure and compliant environments able to deliver the quality of service necessary to deliver value to customers, teams, and ecosystems. You cannot do that without a truly hybrid approach to your data and your AI. The future of AI and data sovereignty: Yours to own The future of enterprise leadership will be defined by those who navigate the complexities of AI and data sovereignty while extracting value from the sprawl of data estates. The leaders surveyed believe we are on the cusp of something monumental — imagine 50 more Amazon-like companies and the creation of $100 trillion in new capital value. These 30% of executive leaders believe it’s possible. EDB will continue monitoring this global shift with a complete report later this year which will cover 13 key geographies representing over $48 trillion in GDP. The research will offer a clearer view of where organizations stand — by region, industry, and platform maturity — and what returns they can expect from their AI and data investments. The findings will include decision-making frameworks to help enterprise leaders prioritize use cases, assess platform needs, and identify the next area of focus. Dig deeper: Read the initial findings from EDB’s Global AI and Data Sovereignty Research here. Michael Gale is CMO at EDB. Sponsored articles are content produced by a company that is either paying for the post or has a business relationship with VentureBeat, and they’re always clearly marked. For more information, contact [email protected]. source

Want smarter insights in your inbox? Sign up for our weekly newsletters to get only what matters to enterprise AI, data, and security leaders. Subscribe Now AI agents are one of the hottest topics in tech right now — but how many enterprises have actually deployed and are actively using them?  LinkedIn says it has with its LinkedIn hiring assistant. Going beyond its popular recommender systems and AI-powered search, the company’s AI agent sources and recruits job candidates through a simple natural language interface.  “This is not a demo product,” Deepak Agarwal, chief AI officer at LinkedIn, said onstage this week at VB Transform. “This is live. It’s saving a lot of time for recruiters so that they can spend their time doing what they really love to do, which is nurturing candidates and hiring the best talent for the job.” >>See all our Transform 2025 coverage here<< Relying on a multi-agent system LinkedIn is taking a multi-agent approach, using what Agarwal described as a collection of agents collaborating to get the job done. A supervisor agent orchestrates all the tasks among other agents, including intake and sourcing agents that are “good at one and only one job.”  All communication occurs through the supervisor agent, which receives input from human users regarding role qualifications and other details. That agent then provides context to a sourcing agent, which culls through recruiter search stacks and sources candidates along with descriptions on why they might be a good fit for the job. That information is then returned to the supervisor agent, which begins actively interacting with the human user.  “Then you can collaborate with it, right?” said Agarwal. “You can modify it. No longer do you have to talk to the platform in keywords. You can talk to the platform in natural language, and it’s going to answer you back, it’s going to have a conversation with you.” The agent can then refine qualifications and begin sourcing candidates, working for the hiring manager “both synchronously and asynchronously.” “It knows when to delegate the task to what agent, how to collect feedback and display to the user,” said Agarwal.  He emphasized the importance of “human first” agents that keeps users always in control. The goal is to “deeply personalize” experiences with AI that adapts to preferences, learns from behaviors and continues to evolve and improve the more that users interact with it.  “It is about helping you accomplish your job in a better and more efficient way,” said Agarwal.  How LinkedIn trains its multi-agent system A multi-agent system requires a nuanced approach to training. LinkedIn’s team spends a lot of time on fine-tuning and making each downstream agent efficient for its specific task to improve reliability, explained Tejas Dharamsi, LinkedIn senior staff software engineer.  “We fine-tune domain-adapted models and make them smaller, smarter and better for our task,” he said.  Whereas the supervisor agent is a special agent that requires high intelligence and adaptability. LinkedIn’s orchestrating agent can reason by using the company’s frontier large language models (LLMs). It also incorporates reinforcement learning and continuous user feedback.  Further, the agent has “experiential memory,” Agarwal explained, so it can retain information from recent dialog. It can preserve long-term memory about user preferences, as well, and discussions that could be important to recall later in the process.  “Experiential memory, along with global context and intelligent routing, is the heart of the supervisor agent, and it keeps getting better and better through reinforcement learning,” he said.  Iterating throughout the agent development cycle Dharamsi emphasized that with AI agents, latency has to be on point. Before deploying into production, LinkedIn model builders need to understand how many queries per second (QPS) models can support and how many GPUs are required to power those. To determine this and other factors, the company runs a lot of inference and does evaluations, along with ntensive red teaming and risk assessment.  “We want these models to be faster, and sub-agents to do their tasks better, and they’re really fast at doing that,” he said.  Once deployed, from a UI perspective, Dharamsi described LinkedIn’s AI agent platform as “Lego blocks that an AI developer can plug and play.” The abstractions are designed so that users can pick and choose based on their product and what they want to build.  “The focus here is how we standardize the development of agents at LinkedIn, so that in a consistent fashion you can build these again and again, try different hypotheses,” he explained. Engineers can instead focus on data, optimization and loss and reward function, rather than the underlying recipe or infrastructure.  LinkedIn provides engineers with different algorithms based on RL, supervised fine tuning, pruning, quantization and distillation to use out of the box without worrying about GPU optimization or FLOPS, so they can begin running algorithms and training, said Dharamsi.  In building out its models, LinkedIn focuses on several factors, including reliability, trust, privacy, personalization and price, he said. Models must provide consistent outputs without getting derailed. Users also want to know that they can rely on agents to be consistent; that their work is secure; that past interactions are being used to personalize; and that costs don’t skyrocket.  “We want to provide more value to the user, to do their job back better and do things that bring them happiness, like hiring,” said Dharamsi. “Recruiters want to focus on sourcing the right candidate, not spending time on searches.”  source

Want smarter insights in your inbox? Sign up for our weekly newsletters to get only what matters to enterprise AI, data, and security leaders. Subscribe Now Bright Data, the Israeli web scraping company that defeated both Meta and Elon Musk’s X in federal court, unveiled a comprehensive AI infrastructure suite Wednesday designed to give artificial intelligence systems unfettered access to real-time web data — a capability the company argues Big Tech platforms are trying to monopolize. The announcement of Deep Lookup, Browser.ai, and enhanced data collection protocols represents a dramatic expansion for the decade-old company, which has transformed from a specialized web scraping service into what CEO Or Lenchner calls “a unique infrastructure layer for AI companies.” The move comes as artificial intelligence companies increasingly struggle to access current web information needed to power chatbots, autonomous agents, and other AI applications. “The intelligence of today’s LLMs is no longer its limiting factor; access is,” Lenchner said in an exclusive interview with VentureBeat. “We’ve spent the last decade fighting for open access to public web data, and these new offerings bring us to the next chapter in our journey, one characterized by truly accessible data and the subsequent rise of contextually-aware agents.” The launch follows Bright Data’s high-profile legal victories in 2024, when federal judges dismissed lawsuits from both Meta and X alleging the company illegally scraped their platforms. Those rulings established crucial legal precedent defining what constitutes “public data” on the internet — information that can be viewed without logging in and therefore can be legally collected and used. The court cases revealed that both Meta and X had been Bright Data customers even while suing the company, highlighting the contradictory stance many tech giants have taken toward web scraping. The rulings have broader implications for the AI industry, which relies heavily on web data to train and operate language models. “It was revealed in court that both of them were a Bright Data customer, because everyone needs data, everyone, especially those who are building models,” Lenchner explained. “We are the only company that has the financial resources, and I would even say the courage to do that.” Judge William Alsup, who presided over the X case, wrote that giving social media companies “free rein to decide, on any basis, who can collect and use data” risks creating “information monopolies that would disserve the public interest.” The ruling established that data viewable without login credentials constitutes public information that can be legally scraped. Bright Data had previously filed a countersuit against X, alleging the platform violated antitrust laws by trying to create a data monopoly to benefit Musk’s AI company, xAI. However, that case has since been settled. “Though the terms confidential, Bright Data has never backed down from its fundamental belief that public data should be available to the public. Consistent with that belief, we are pleased to report that Bright Data will continue to provide the same industry-leading services that it always has and that our customers have come to expect,” Lenchner said. Deep Lookup and Browser.ai target AI companies struggling with data access The company’s new products address what Lenchner identifies as the three core requirements for AI systems: algorithms, compute power, and data access. While Bright Data doesn’t develop AI algorithms or provide computing resources, it aims to become the definitive solution for the third requirement. Deep Lookup functions as a natural language research engine designed to answer complex, multi-layered business questions in real-time. Unlike general-purpose search engines or AI chatbots that provide summaries, Deep Lookup specializes in comprehensive results for queries beginning with “find all.” For example, users can ask for “all shipping companies that went through the Panama and Suez canals in 2023 whose Q3 revenues declined by over 2 percent.” The system draws from Bright Data’s massive web archive, which currently contains over 200 billion HTML pages and adds 15 billion monthly. By next year, the archive is expected to exceed 500 billion pages. “It’s not just random web pages, it’s actually what the world cares about, because our 20,000 customers represent billions of internet users,” Lenchner noted. Browser.ai represents what the company calls “the industry’s first unblockable, AI-native browser.” Designed specifically for autonomous AI agents, the cloud-based service mimics human behavior to access websites without triggering bot detection systems. It supports natural language commands and can perform complex web interactions like booking flights or making restaurant reservations. The browser infrastructure already processes over 150 million web actions daily, according to the company. “Almost all of them are customers,” Lenchner said of AI agent companies that have raised significant funding. “Because what we figured out, and they figured out, is that we solve that problem of entering a website without being blocked and executing web actions on the website.” MCP Servers (Model Context Protocol) provides a low-latency control layer enabling AI agents to search, crawl, and extract live data in real-time. The protocol allows developers to build AI systems that can act on current information rather than relying solely on training data. Patent portfolio and proxy network create competitive moat against blocking Bright Data’s competitive advantage stems from what Lenchner describes as an “obsession” with overcoming website blocking mechanisms. The company holds over 5,500 patent claims on its technology and operates the world’s largest proxy network with more than 150 million IP addresses across 195 countries. “We have such a good look into the internet,” Lenchner explained. “For a long time now, we have been mapping the internet, and for a long time now, we’re also archiving big chunks of the internet.” The company’s approach involves sophisticated techniques to mimic human behavior, using real devices, IP addresses, and browser fingerprints rather than simple automated scripts. This makes detection and blocking extremely difficult for websites. “The only way to block us, practically, is to put the data behind the login, then we won’t even try,” Lenchner said. “Sometimes there is a new blocking logic that we won’t solve immediately. It will take our

Join the event trusted by enterprise leaders for nearly two decades. VB Transform brings together the people building real enterprise AI strategy. Learn more This article is part of VentureBeat’s special issue, “The Real Cost of AI: Performance, Efficiency and ROI at Scale.” Read more from this special issue. Three years after ChatGPT launched the generative AI era, most enterprises remain trapped in pilot purgatory. Despite billions in AI investments, the majority of corporate AI initiatives never escape the proof-of-concept phase, let alone generate measurable returns. But a select group of Fortune 500 companies has cracked the code. Walmart, JPMorgan Chase, Novartis, General Electric, McKinsey, Uber and others have systematically moved AI from experimental “innovation theater” to production-grade systems delivering substantial ROI—in some cases, generating over $1 billion in annual business value. Their success isn’t accidental. It’s the result of deliberate governance models, disciplined budgeting strategies and fundamental cultural shifts that transform how organizations approach AI deployment. This isn’t about having the best algorithms or the most data scientists. It’s about building the institutional machinery that turns AI experiments into scalable business assets. “We see this as a pretty big inflection point, very similar to the internet,” Walmart’s VP of emerging technology Desirée Gosby said at this week’s VB Transform event. “It’s as profound in terms of how we’re actually going to operate, how we actually do work.” The pilot trap: Why most AI initiatives fail to scale The statistics are sobering. Industry research shows that 85% of AI projects never make it to production, and of those that do, fewer than half generate meaningful business value. The problem isn’t technical—it’s organizational. Companies treat AI as a science experiment rather than a business capability. “AI is already cutting some product-development cycles by about 40 percent, letting companies ship and decide faster than ever,” said Amy Hsuan, chief customer and revenue officer at Mixpanel. “But only for companies that have moved beyond pilots to systematic deployment.” The failure patterns are predictable: scattered initiatives across business units, unclear success metrics, insufficient data infrastructure and—most critically—the absence of governance frameworks that can manage AI at enterprise scale. Initial evaluation is also something too many organizations overlook, Sendbird head of product Shailesh Nalawadi emphasized at this week’s VB Transform. “Before you even start building [agentic AI], you should have an eval infrastructure in place. No one deploys to production without running unit tests. And I think a very simplistic way of thinking about eval is that it’s the unit test for your AI agent system.” Simply put, you can’t build agents like other software, Writer CEO and co-founder May Habib said at VB Transform. They are “categorically different” in how they’re built, operated and improved, and the traditional software development life cycle doesn’t cut it with adaptive systems. “Agents don’t reliably follow rules,” Habib said. “They are outcome-driven. They interpret. They adapt. And the behavior really only emerges in real-world environments.” The production imperative: A framework for systematic AI deployment The companies that have succeeded share a remarkably consistent playbook. Through interviews with executives and analysis of their AI operations, eight critical elements emerge that distinguish pilot-phase experimentation from production-ready AI systems: 1. Executive mandate and strategic alignment Every successful AI transformation begins with unambiguous leadership commitment. This isn’t ceremonial sponsorship—it’s active governance that ties every AI initiative to specific business outcomes. At Walmart, CEO Doug McMillon established five clear objectives for AI projects: enhancing customer experience, improving operations, accelerating decision-making, optimizing supply chains and driving innovation. No AI project gets funded without mapping to these strategic pillars. “It always comes back to basics,” Gosby advised. “Take a step back and first understand what problems do you really need to solve for your customers, for our associates. Where is there friction? Where is there manual work that you can now start to think differently about?” “We don’t want to just throw spaghetti at the wall,” explained Anshu Bhardwaj, Walmart’s SVP of Global Tech. “Every AI project must target a specific business problem with measurable impact.” JPMorgan Chase’s Jamie Dimon takes a similar approach, calling AI “critical to our future success” while backing that rhetoric with concrete resource allocation. The bank has over 300 AI use cases in production precisely because leadership established clear governance from day one. Practical implementation: Create an AI steering committee with C-level representation. Establish 3-5 strategic objectives for AI initiatives. Require every AI project to demonstrate clear alignment with these objectives before funding approval. 2. Platform-first infrastructure strategy The companies that scale AI successfully don’t build point solutions—they build platforms. This architectural decision becomes the foundation for everything else. Walmart’s “Element” platform exemplifies this approach. Rather than allowing teams to build isolated AI applications, Element provides a unified machine learning infrastructure with built-in governance, compliance, security and ethical safeguards. This allows teams to plug in new AI capabilities quickly while maintaining enterprise-grade controls. “The vision with Element always has been, how do we have a tool that allows data scientists and engineers to fast track the development of AI models?” Parvez Musani, Walmart’s SVP of stores and online pickup and delivery technology, told VentureBeat in a recent interview. He emphasized that they built Element to be model agnostic. “For the use case or the query type that we are after, Element allows us to pick the best LLM out there in the most cost-effective manner.” JPMorgan Chase invested $2+ billion in cloud infrastructure specifically to support AI workloads, migrating 38% of applications to cloud environments optimized for machine learning. This wasn’t just about compute power—it was about creating an architecture that could handle AI at scale. Practical implementation: Invest in a centralized ML platform before scaling individual use cases. Include governance, monitoring, and compliance capabilities from day one. Budget 2-3x your initial estimates for infrastructure—scaling AI requires substantial computational resources. 3. Disciplined use case selection and portfolio management The most successful companies resist the temptation to pursue flashy AI applications in favor of high-ROI use cases with clear business metrics. Novartis CEO

Presented by EdgeVerve Despite substantial investment, AI in the enterprise often stalls at the proof-of-concept stage — trapped in silos and limited in scale. Without a cohesive strategy, organizations often encounter scalability challenges, governance gaps and data fragmentation. Successful pilots in customer service automation or predictive analytics, may not translate into organization-wide value if AI systems operate in isolation. This is where enterprise-grade AI platforms play a transformative role. Modern AI platforms create a connected ecosystem across business units, enabling seamless data flow, standardized model deployment, and unified governance frameworks. They facilitate interoperability across disparate systems — CRM, ERP, SCM — ensuring that AI models have access to holistic, high-quality data critical for effective predictions and decisions. By integrating various data sources and AI models, these platforms enable organizations to break down silos and achieve more efficient, cross-functional operations, ultimately driving better business outcomes. By combining AI with automation and orchestration capabilities, platforms also allow enterprises to move from isolated efficiencies to systemic transformation. This shift from AI experiments to AI-enabled enterprises (or the “foundry to factory” approach) is foundational to realizing a sustainable competitive advantage and unlocking newer growth opportunities. Embedding contextual intelligence across the enterprise AI’s true value emerges when intelligence is deeply rooted in business context, not when it operates in abstraction. A predictive maintenance model is only effective if it understands the nuances of a specific manufacturing process. A customer service AI solution must be trained on sector-specific vocabulary and sentiment to provide meaningful assistance. Modern AI platforms empower enterprises to build domain-aware models that can interpret signals, behaviors, and risks through the lens of industry and business-specific knowledge. In industries like healthcare or finance, where regulatory changes are frequent, maintaining contextual intelligence becomes even more critical. For example, AI-driven predictive models in healthcare must not only be trained on patient data but also adapt to new regulations regarding privacy and treatment protocols. Platforms with adaptive learning capabilities ensure these models stay compliant, offering a safeguard against both regulatory and operational risks. This means curating domain-specific datasets, layering in contextual metadata, and ensuring model outcomes are tied directly to operational KPIs. AI platforms make this possible by providing the foundation to embed real-world relevance into every output, so results aren’t just technically accurate, but useful and aligned with business priorities. Contextual intelligence also plays a key role in building trust, something that’s becoming more valuable than ever in today’s AI-driven world. Responsible AI is essential and can’t be treated as an afterthought. Core principles like bias detection, explainability, and fairness need to be built into the model lifecycle from the start. Adapting AI to stay relevant in a changing world In fast-moving markets, yesterday’s insights can quickly lose their relevance. Customer expectations shift, supply chains realign, and regulatory landscapes evolve. AI models that remain static in the face of these changes risk becoming obsolete as they use outdated information to inform their decision-making. To stay effective, AI must continuously learn and adapt. That means retraining and refining models based on real-time data, performance feedback, and new external conditions. A unified AI platform can play a pivotal role here by not only integrating data but also transforming and feeding to AI models at desired speed and scale. But continuous learning isn’t just a technical function, it’s an organizational mindset. Enterprises need processes that regularly evaluate model performance and teams empowered to make adjustments that keep AI aligned with business goals. For example, a retail company using AI for demand forecasting must regularly recalibrate its models to reflect changing consumer behaviors, seasonal trends, and new product lines. Aligning those updates with inventory planning and marketing efforts ensures AI continues to drive measurable impact. Core components of continuous learning include: Monitoring model drift: Detecting when a model’s predictions begin to diverge from expected outcomes. Automated retraining pipelines: Streamlining updates by triggering model retraining as new data becomes available — without waiting for manual intervention. Consider a model designed to predict supply chain disruptions. As geopolitical dynamics or supplier performance change, the model should auto-update to reflect emerging risks — ensuring agile, informed decisions. The AI platform powering it must be flexible and scalable, not rigid, to meet evolving enterprise needs. Amplifying human potential with AI Despite the narrative around AI replacing jobs, the most successful enterprises will be those that use AI and automation to amplify human potential, not diminish it. Modern AI platforms integrate decision intelligence systems that assist human decision-makers rather than replacing them. For example: In customer support, AI can suggest next-best actions while humans retain the final say. In financial services, AI can highlight anomalous transactions for human review rather than making unilateral decisions. In supply chain operations, AI can recommend optimal routes based on predictive analytics, empowering managers to make more informed choices. By combining AI with automation in a human-centric design, enterprises can foster more fulfilling work environments, drive higher productivity, and unlock innovation at scale. This symbiotic relationship between human and machine is not just a technological goal, it’s a cultural transformation that defines the future of work. Enterprises should prioritize AI platforms that optimize TCO through a modular, resource-efficient architecture, while maximizing ROI by seamlessly integrating with existing systems — enhancing value from past digital investments with minimal disruption or added cost. Building resilient, adaptive enterprises The journey to becoming an AI-first enterprise is complex. It requires more than just new technologies; it demands reimagined processes, new governance models, leadership commitment, and a willingness to continuously evolve. AI platforms are the technological foundation of this transformation, but the mindset shift they enable is even more important. A resilient, AI-first enterprise is characterized by: Integrated intelligence: AI embedded seamlessly into every operational layer. Contextual relevance: Models that understand the nuances of business processes and customer needs. Continuous evolution: Adaptive systems that grow smarter over time. Responsible governance: Trustworthy AI practices that ensure fairness, transparency, and regulatory compliance. By embracing connected, contextual, and continuous AI, enterprises can build adaptive advantage, responding faster

Join the event trusted by enterprise leaders for nearly two decades. VB Transform brings together the people building real enterprise AI strategy. Learn more >>See all our Transform 2025 coverage here<< When people started to talk about AI agents and assistants, the number one use case revolved around travel. Could someone be watching a video about the Maldives and direct their AI agent to start finding flights and hotels, and book these seamlessly?  We‘re inching closer to a similar future as the travel industry begins to embrace agentic AI. Kayak and Expedia, two of the largest companies in travel booking, said during VB Transform that personalization and changing search patterns mean travel companies can rely on agents to make travel inspiration a reality. Matthias Keller, chief product officer at Kayak, said the company has been experimenting with this idea for a couple of years, even taking advantage of a partnership with Amazon’s Alexa. Kayak no longer launches on Alexa, but that hasn’t stopped the company from offering different search modes for customers. “We are striving for this vision of a travel agent that is always available, that is agentic and powered by AI,” Keller said onstage at VB Transform. “In April, we launched our new testbed for agentic travel booking called Kayak AI; it’s a fully chat-based agentic experience that puts together the power of ChatGPT and many different tools. One is web search, but we also offer tools specifically for flights or hotels.” Keller said Kayak is “working towards our vision of having this fully personalized experienced that does all the heavy lifting for you when you plan travel.” While the idea of an agent proactively guiding potential travelers makes for efficient trip planning, Expedia CTO Ramana Thumu noted that there is a delicate balance to strike. “More and more customer expectations will revolve around a seamless experience, from search to completing transactions,” Thumu said. “But the most important thing, and the shift I see happening, is asking for the balance between the control the traveler has, and the control we give to the agent.” One reason this balance becomes essential is that, increasingly, consumers find travel inspiration everywhere. For one of its AI projects, Expedia decided to capitalize on the growing influence of travel influencers who post their trips on Instagram. Thumu said Expedia’s Trip Matching feature, which launched in June for U.S. customers, allows people to send any travel-related public Instagram Reel to Expedia, and the platform can build an itinerary based on them.  Thumu said Expedia can build this type of AI product because of its extensive database amassed over 30 years. Both Thumu and Keller underscored the importance of data in building out these personalizations, a task that can be challenging.  Personalization can go beyond planning a trip based on inspiration or previous preferences, as Keller said; eventually, their platforms and AI agents can also start recommending things to do based on the weather in your planned location during your stay.  AI helps simplify the complexity One use case where companies like Kayak and Expedia find AI to be helpful is for “snackers,” or individuals who search for flights or hotels without any intention of booking. These are usually people who just want to check the price of a flight or find out how much a hotel costs.  AI systems can help snackers find their answers, and may even encourage them to go on that trip, because much of the tediousness of finding accommodation, transportation and activities can be presented to them right there on either the Expedia or Kayak front end.  “What I find interesting when I pitch Kayak AI to somebody [is] I can say that we can get this complex trip planned without running all of the searches,” Keller said. “Every hotel booking site out there can tell you that a hotel has a pool, but you have to go deep to find an infinity pool. That’s the type of question that ChatGPT does a great job with, so it’s something we have to adapt and deliver.” source

Join the event trusted by enterprise leaders for nearly two decades. VB Transform brings together the people building real enterprise AI strategy. Learn more Editor’s note: Kumo AI was one of the finalists at VB Transform during our annual innovation showcase and presented RFM from the mainstage at VB Transform on Wednesday. The generative AI boom has given us powerful language models that can write, summarize and reason over vast amounts of text and other types of data. But when it comes to high-value predictive tasks like predicting customer churn or detecting fraud from structured, relational data, enterprises remain stuck in the world of traditional machine learning.  Stanford professor and Kumo AI co-founder Jure Leskovec argues that this is the critical missing piece. His company’s tool, a relational foundation model (RFM), is a new kind of pre-trained AI that brings the “zero-shot” capabilities of large language models (LLMs) to structured databases. “It’s about making a forecast about something you don’t know, something that has not happened yet,” Leskovec told VentureBeat. “And that’s a fundamentally new capability that is, I would argue, missing from the current purview of what we think of as gen AI.” Why predictive ML is a “30-year-old technology” While LLMs and retrieval-augmented generation (RAG) systems can answer questions about existing knowledge, they are fundamentally retrospective. They retrieve and reason over information that is already there. For predictive business tasks, companies still rely on classic machine learning.  For example, to build a model that predicts customer churn, a business must hire a team of data scientists who spend a considerably long time doing “feature engineering,” the process of manually creating predictive signals from the data. This involves complex data wrangling to join information from different tables, such as a customer’s purchase history and website clicks, to create a single, massive training table. “If you want to do machine learning (ML), sorry, you are stuck in the past,” Leskovec said. Expensive and time-consuming bottlenecks prevent most organizations from being truly agile with their data. How Kumo is generalizing transformers for databases Kumo’s approach, “relational deep learning,” sidesteps this manual process with two key insights. First, it automatically represents any relational database as a single, interconnected graph. For example, if the database has a “users” table to record customer information and an “orders” table to record customer purchases, every row in the users table becomes a user node, every row in an orders table becomes an order node, and so on. These nodes are then automatically connected using the database’s existing relationships, such as foreign keys, creating a rich map of the entire dataset with no manual effort. Relational deep learning Source: Kumo AI Second, Kumo generalized the transformer architecture, the engine behind LLMs, to learn directly from this graph representation. Transformers excel at understanding sequences of tokens by using an “attention mechanism” to weigh the importance of different tokens in relation to each other.  Kumo’s RFM applies this same attention mechanism to the graph, allowing it to learn complex patterns and relationships across multiple tables simultaneously. Leskovec compares this leap to the evolution of computer vision. In the early 2000s, ML engineers had to manually design features like edges and shapes to detect an object. But newer architectures like convolutional neural networks (CNN) can take in raw pixels and automatically learn the relevant features.  Similarly, the RFM ingests raw database tables and lets the network discover the most predictive signals on its own without the need for manual effort. The result is a pre-trained foundation model that can perform predictive tasks on a new database instantly, what’s known as “zero-shot.” During a demo, Leskovec showed how a user could type a simple query to predict whether a specific customer would place an order in the next 30 days. Within seconds, the system returned a probability score and an explanation of the data points that led to its conclusion, such as the user’s recent activity or lack thereof. The model was not trained on the provided database and adapted to it in real time through in-context learning.  “We have a pre-trained model that you simply point to your data, and it will give you an accurate prediction 200 milliseconds later,” Leskovec said. He added that it can be “as accurate as, let’s say, weeks of a data scientist’s work.”  The interface is designed to be familiar to data analysts, not just machine learning specialists, democratizing access to predictive analytics. Powering the agentic future This technology has significant implications for the development of AI agents. For an agent to perform meaningful tasks within an enterprise, it needs to do more than just process language; it must make intelligent decisions based on the company’s private data. The RFM can serve as a predictive engine for these agents. For example, a customer service agent could query the RFM to determine a customer’s likelihood of churning or their potential future value, then use an LLM to tailor its conversation and offers accordingly. “If we believe in an agentic future, agents will need to make decisions rooted in private data. And this is the way for an agent to make decisions,” Leskovec explained. Kumo’s work points to a future where enterprise AI is split into two complementary domains: LLMs for handling retrospective knowledge in unstructured text, and RFMs for predictive forecasting on structured data. By eliminating the feature engineering bottleneck, the RFM promises to put powerful ML tools into the hands of more enterprises, drastically reducing the time and cost to get from data to decision. The company has released a public demo of the RFM and plans to launch a version that allows users to connect their own data in the coming weeks. For organizations that require maximum accuracy, Kumo will also offer a fine-tuning service to further boost performance on private datasets. source

This article is part of VentureBeat’s special issue, “The Real Cost of AI: Performance, Efficiency and ROI at Scale.” Read more from this special issue. AI has become the holy grail of modern companies. Whether it’s customer service or something as niche as pipeline maintenance, organizations in every domain are now implementing AI technologies — from foundation models to VLAs — to make things more efficient. The goal is straightforward: automate tasks to deliver outcomes more efficiently and save money and resources simultaneously. However, as these projects transition from the pilot to the production stage, teams encounter a hurdle they hadn’t planned for: cloud costs eroding their margins. The sticker shock is so bad that what once felt like the fastest path to innovation and competitive edge becomes an unsustainable budgetary blackhole – in no time.  This prompts CIOs to rethink everything—from model architecture to deployment models—to regain control over financial and operational aspects. Sometimes, they even shutter the projects entirely, starting over from scratch. But here’s the fact: while cloud can take costs to unbearable levels, it is not the villain. You just have to understand what type of vehicle (AI infrastructure) to choose to go down which road (the workload). The cloud story — and where it works  The cloud is very much like public transport (your subways and buses). You get on board with a simple rental model, and it instantly gives you all the resources—right from GPU instances to fast scaling across various geographies—to take you to your destination, all with minimal work and setup.  The fast and easy access via a service model ensures a seamless start, paving the way to get the project off the ground and do rapid experimentation without the huge up-front capital expenditure of acquiring specialized GPUs.  Most early-stage startups find this model lucrative as they need fast turnaround more than anything else, especially when they are still validating the model and determining product-market fit. “You make an account, click a few buttons, and get access to servers. If you need a different GPU size, you shut down and restart the instance with the new specs, which takes minutes. If you want to run two experiments at once, you initialise two separate instances. In the early stages, the focus is on validating ideas quickly. Using the built-in scaling and experimentation frameworks provided by most cloud platforms helps reduce the time between milestones,” Rohan Sarin, who leads voice AI product at Speechmatics, told VentureBeat. The cost of “ease” While cloud makes perfect sense for early-stage usage, the infrastructure math becomes grim as the project transitions from testing and validation to real-world volumes. The scale of workloads makes the bills brutal — so much so that the costs can surge over 1000% overnight.  This is particularly true in the case of inference, which not only has to run 24/7 to ensure service uptime but also scale with customer demand.  On most occasions, Sarin explains, the inference demand spikes when other customers are also requesting GPU access, increasing the competition for resources. In such cases, teams either keep a reserved capacity to make sure they get what they need — leading to idle GPU time during non-peak hours — or suffer from latencies, impacting downstream experience. Christian Khoury, the CEO of AI compliance platform EasyAudit AI, described inference as the new “cloud tax,” telling VentureBeat that he has seen companies go from $5K to $50K/month overnight, just from inference traffic. It’s also worth noting that inference workloads involving LLMs, with token-based pricing, can trigger the steepest cost increases. This is because these models are non-deterministic and can generate different outputs when handling long-running tasks (involving large context windows). With continuous updates, it gets really difficult to forecast or control LLM inference costs. Training these models, on its part, happens to be “bursty” (occurring in clusters), which does leave some room for capacity planning. However, even in these cases, especially as growing competition forces frequent retraining, enterprises can have massive bills from idle GPU time, stemming from overprovisioning. “Training credits on cloud platforms are expensive, and frequent retraining during fast iteration cycles can escalate costs quickly. Long training runs require access to large machines, and most cloud providers only guarantee that access if you reserve capacity for a year or more. If your training run only lasts a few weeks, you still pay for the rest of the year,” Sarin explained. And, it’s not just this. Cloud lock-in is very real. Suppose you have made a long-term reservation and bought credits from a provider. In that case, you’re locked in their ecosystem and have to use whatever they have on offer, even when other providers have moved to newer, better infrastructure. And, finally, when you get the ability to move, you may have to bear massive egress fees. “It’s not just compute cost. You get…unpredictable autoscaling, and insane egress fees if you’re moving data between regions or vendors. One team was paying more to move data than to train their models,” Sarin emphasized. So, what’s the workaround? Given the constant infrastructure demand of scaling AI inference and the bursty nature of training, enterprises are moving to splitting the workloads — taking inference to colocation or on-prem stacks, while leaving training to the cloud with spot instances. This isn’t just theory — it’s a growing movement among engineering leaders trying to put AI into production without burning through runway. “We’ve helped teams shift to colocation for inference using dedicated GPU servers that they control. It’s not sexy, but it cuts monthly infra spend by 60–80%,” Khoury added. “Hybrid’s not just cheaper—it’s smarter.” In one case, he said, a SaaS company reduced its monthly AI infrastructure bill from approximately $42,000 to just $9,000 by moving inference workloads off the cloud. The switch paid for itself in under two weeks. Another team requiring consistent sub-50ms responses for an AI customer support tool discovered that cloud-based inference latency was insufficient. Shifting inference closer to users via colocation not