Fabrix.ai team attended VentureBeat’s – VB Transform conference returned this week as the premier gathering for enterprise AI leaders, showcasing how artificial intelligence has evolved from experimental chatbots to autonomous agents reshaping entire industries. The event brought together leaders across finance, healthcare, tech, retail, and manufacturing—from the C-suite to heads of product, data, and AI to explore the next frontier: agentic AI systems that can plan, decide, and execute complex tasks independently.
Bottom Line Up Front: The Agentic AI Transformation is Here
Enterprise leaders are done talking about GenAI’s potential. It’s time to build, scale, and prove value. The companies presenting at VB Transform aren’t discussing future possibilities as they’re already deploying autonomous agents, achieving measurable ROI, and scaling across their organizations. The transformation from experimental AI to production-ready autonomous systems is no longer coming but it’s already here.
The Agentic AI Revolution: Beyond Copilots to True Autonomy
The conference’s central theme was clear from the opening session: we’re entering a phase where AI agents don’t just assist, they collaborate, reason, and execute across enterprise systems. Varun Mohan, CEO of Windsurf, kicked off the event by demonstrating how their agentic IDE surpassed one million developers within four months of launch, with the platform now writing over half of the code committed by its user base.
Windsurf: The AI-Native IDE Revolution
Mohan, who previously led teams at Nuro focused on large-scale deep learning infrastructure for autonomous vehicles, transformed his company from a GPU virtualization startup to the leading AI-native IDE through a bold pivot. Windsurf’s IDE is built around what the company calls a “mind-meld” loop—a shared project state between humans and AI that enables full coding flows rather than autocomplete suggestions.
The company’s biggest optimization at enterprise scale isn’t faster token generation or smaller models—it’s relevance. “At scale, the biggest optimization is personalization. Deeply understanding the codebase allows the agent to make maintainable, large-scale changes that reflect user intent,” Mohan explained. Recent developments include embedding a browser inside the IDE, allowing agents to test changes, read logs, and interact with live interfaces directly—much like a human engineer would.
Andrew Ng: The Enterprise S-Curve Shift
Andrew Ng, globally recognized leader in AI and Managing General Partner at AI Fund, provided a crucial perspective on this transition. Ng emphasized that for most businesses, the priority should be building applications with agentic workflows rather than focusing solely on the latest foundational models. His research shows that GPT-3.5 with an agentic workflow can outperform more advanced models like GPT-4 using zero-shot approaches.
Ng outlined four key patterns enabling agentic AI systems: reflection (self-evaluation), tool use (API integration), planning (task breakdown), and multi-agent collaboration (specialized roles). These “patterns” enable AI Agent systems to tackle complex tasks through methodical, step-by-step approaches while maintaining better control and accuracy.
Real-World Enterprise Implementations
Financial Services: Leading the Production Charge
Capital One’s Multi-Agent Mastery
Dr. Milind Naphade, SVP of Technology and Head of AI Foundations at Capital One, joined after senior roles at IBM, NVIDIA, and Cisco. He leads AI efforts including Chat Concierge, a conversational assistant for vehicle comparisons, scheduling, and workflows.
The system leverages multiple AI agents working together to not only provide information but take specific actions based on customer preferences. Breaking workflows into discrete tasks and assigning each to specialized AI agents helps ease cognitive load and create more streamlined experiences. Capital One’s multi-agentic AI workflow was highlighted in NVIDIA CEO Jensen Huang’s keynote, where he noted Capital One as one of the most advanced financial services companies in using technology.
American Express and Intuit’s Advanced Architectures
The financial sector emerged as an early adopter of production-ready agentic systems. American Express and Intuit showcased how intelligent agents are moving beyond generative AI to reshape financial services entirely. Intuit’s GenOS platform demonstrated advanced agentic architectures that handle complex financial workflows autonomously, representing a significant evolution from traditional rule-based systems.
Healthcare: Hallucination-Free AI in Production
Stanford’s ChatEHR: Conversational Medical Records
Stanford Health Care deployed ChatEHR, a groundbreaking AI system that enables clinicians to query patient medical records using natural language. The technology, led by Dr. Nigam Shah, Chief Data Science Officer, and Dr. Michael Pfeffer, Chief Information and Digital Officer, represents one of the first production implementations of LLMs directly embedded within electronic health record workflows.
In early pilot results with 33 Stanford clinicians, emergency physicians experienced 40% reduced chart review time during critical handoffs. This helps decrease physician burnout while improving patient care, building upon decades of work to digitize healthcare data. The system is integrated directly into Stanford’s Epic-based EHR to reduce friction, allowing clinicians to ask questions like “Does this patient have any allergies?” or “What does their latest cholesterol test show?” and receive secure, contextual responses.
The team is evaluating ChatEHR using MedHELM, an open-source framework for real-world LLM evaluation in medicine, and developing features like in-text citations so clinicians can see exactly where information originated in patient records.
Mayo Clinic’s Revolutionary Anti-Hallucination Architecture
Mayo Clinic delivered perhaps the most technically impressive healthcare AI implementation, presenting their blueprint for hallucination-free, multimodal AI systems. Their reverse-RAG pipeline—extract first, then tether every fact back to its source—paired with the CURE clustering algorithm has driven retrieval errors nearly to zero, addressing healthcare’s biggest AI concern.
LinkedIn’s Production-Grade Hiring Agent
LinkedIn launched its first AI agent, Hiring Assistant, designed to automate up to 80% of the pre-offer recruitment workflow. The system is now live with enterprise customers including Siemens, Canva, AMD, and Zurich Insurance. Built on LinkedIn’s vast data set of over 1 billion members, 68 million companies, and 41,000 skills, the agent can generate role qualifications, develop candidate pipelines, and engage with prospects through AI-assisted messaging.
Early results show hirers using AI-assisted messages experience 44% higher acceptance rates and 11% faster responses from job seekers compared to traditional methods. The system includes “experiential memory” that learns from recruiter feedback to improve performance over time.
Retail and Travel: Customer-Facing Autonomous Agents
Walmart presented on “Trust in the Algorithm,” revealing how the retail giant uses agentic systems to solve real-world customer and associate pain points at global scale. Their systems demonstrate how AI can maintain consumer confidence while delivering personalized experiences across 100+ million customers.
The travel industry showcased sophisticated autonomous agents, with KAYAK and Expedia presenting systems that handle the sector’s unique challenges. These agents negotiate seat inventory, re-route disrupted trips, and provide instant personalized recommendations—all while managing constantly changing fares and complex multi-supplier itineraries.
The Infrastructure Revolution: Building for Autonomy
Anthropic’s Claude 4: Seven-Hour Autonomous Work Sessions
Anthropic’s release of Claude Opus 4 and Claude Sonnet 4 marked a quantum leap in AI capabilities. Claude Opus 4 achieved a 72.5% score on SWE-bench, outperforming OpenAI’s GPT-4.1 (54.6%), and demonstrated the ability to work autonomously for nearly seven hours on complex coding projects. Scott White, Anthropic’s product lead, explained that the model can handle larger, more complex projects for about a full workday operating independently, without additional prompts from humans.
During testing at Rakuten, Claude Opus 4 maintained focus on a complex open-source refactoring project for nearly seven hours—a breakthrough that transforms AI from a quick-response tool into a genuine collaborator capable of tackling day-long projects. This represents a dramatic shift from the minutes-long attention spans of previous AI models.
Rethinking Enterprise Architecture
IBM’s Armand Ruiz emphasized that agentic AI isn’t just a model upgrade—it demands a complete rethinking of enterprise architecture. Organizations must design, monitor, and govern systems that act independently, requiring new approaches to infrastructure and security.
Critical infrastructure components highlighted at the conference included:
- Observability for Agentic Systems: New Relic’s sessions focused on tracking, trust, and transformation in autonomous AI environments, addressing how to maintain control over systems that operate independently
- Storage Solutions: Solidigm and PEAK:AIO addressed AI’s storage bottleneck, a crucial but often overlooked constraint in large-scale deployments
- Security Frameworks: Multiple sessions covered AI red teaming and adversarial testing for stress-testing AI security defenses
The Economics of AI Inference
A standout panel featuring leaders from SemiAnalysis, Cerebras Systems, and Groq explored how AI inference is reshaping enterprise economics. The race to the bottom on cost-per-token, driven by hyperscale build-outs, is creating both opportunities and challenges for enterprises planning their AI infrastructure investments.
Open Source vs. Proprietary: The Strategic Model Debate
The conference featured extensive discussions on model strategy, with most enterprises now running several LLMs in parallel—some open, some proprietary, many embedded in third-party tools. Key strategic considerations included:
- Customization vs. Speed to Market: When to invest in fine-tuned open models versus using proprietary solutions
- Security and Compliance: Balancing innovation with enterprise security requirements
- Cost Management: Managing the economics of multiple model deployments across different use cases
Development and Engineering Transformation
The Coding Revolution: From 20% to 90% AI-Generated Code
Varun Mohan predicts that AI will write over 90% of software code in the future, a vision aligned with industry leaders like Anthropic’s Dario Amodei. This shift will redefine engineering, moving it from code-writing to strategic decision-making. “AI is going to handle the vast majority, if not all, of the ‘solving it’ part,” Mohan explained, referring to the act of coding once the problem and approach are clear.
Genspark’s Record-Breaking “Vibe Working” Revolution
Perhaps the most audacious presentation came from Kay Zhu, CTO of Genspark AI, who challenged fundamental assumptions about enterprise workflows. Genspark achieved an unprecedented $36M ARR in just 45 days with only a 20-person team—setting what may be the fastest growth rate in startup history.
Founded by former Baidu executives Eric Jing and Kay Zhu, Genspark made a bold pivot from their successful AI search engine (5 million users) to their “Super Agent” platform. Their “Less Control, More Tools” philosophy argues that rigid workflows fundamentally limit what AI agents can accomplish for complex business tasks.
The Technical Architecture Behind the Magic
Genspark’s Super Agent uses a “Mixture-of-Agents” system integrating nine differently sized LLMs with over 80 specialized tools and 10+ premium proprietary datasets. Built on Claude’s planning and reasoning capabilities, the system achieves autonomous problem-solving over deterministic execution paths, allowing agents to intelligently backtrack and find alternative approaches when unexpected situations arise.
The platform’s capabilities range from AI slides and sheets to video generation and even “AI Call For Me”—a feature that has revealed unexpected use cases. “Some of the Japanese users are using this to call to resign from their company,” Zhu noted, “and some people are using call for me agents to break up with their boyfriend and girlfriend.”
Redefining the Future of Work
Zhu’s vision extends beyond mere automation to what he calls “vibe working”—bringing the Cursor experience for developers to the workspace for everyone. “Everyone here should be able to do vibe working… it’s not only the software engineer that can do vibe coding,” Zhu explained. The system uses LLM judges to evaluate every agent session, feeding data back through reinforcement learning for continuous improvement.
One demonstration showed five minutes of automated work equaling three hours of manual effort, fundamentally changing how users approach complex information tasks. This represents a philosophical shift from structured workflow orchestration to autonomous agent-driven problem-solving.
Replit’s Economic Disruption
In one of the most striking presentations, Replit’s CEO Amjad Masad demonstrated how AI agents are eliminating traditional enterprise tool costs. Their example of shipping a full CPQ (configure-price-quote) platform with role permissions and Slack integration—replacing $50K+ SaaS contracts—illustrated the disruptive potential of agentic AI.
The Human Element: Leadership and Workforce Evolution
Women in AI: Driving Inclusive Innovation
The conference’s Women in AI awards and breakfast sessions highlighted the crucial role of diverse leadership in AI development. The 7th annual VentureBeat Women in AI Awards honored women leaders, mentors, researchers, and entrepreneurs who are transforming the AI industry with their groundbreaking efforts.
BCG’s Workforce Research
BCG’s research with 3,000 knowledge workers revealed how agentic AI is moving beyond prompt-response interactions into full collaboration, fundamentally reshaping how people plan, decide, and execute work across finance, healthcare, and industrial sectors.
Key Takeaways for Enterprise Leaders
1. The Production Imperative
The message was clear: 2025 is the year to move from AI pilots to production deployments. Companies still stuck in proof-of-concept mode risk falling behind competitors who are already scaling autonomous systems.
2. Infrastructure Investment is Non-Negotiable
Successful agentic AI requires purpose-built infrastructure for observability, security, and multi-agent coordination. Half-measures won’t suffice for production-grade autonomous systems.
3. Architectural Transformation is Mandatory
Traditional enterprise architectures aren’t designed for autonomous agents. Organizations need to fundamentally rethink their technical infrastructure and governance models.
4. Trust and Safety are Paramount
With AI systems making autonomous decisions, trust becomes paramount. Enterprises must implement robust testing, monitoring, and safety mechanisms from day one.
5. Workforce Evolution Strategy is Essential
Agentic AI isn’t just changing technology—it’s reshaping roles, responsibilities, and organizational structures. The most critical skill in an AI-driven world is agency—the drive to identify problems and act independently.
Innovation Showcase: The Next Generation
The Innovation Showcase featured seven companies to present their generative AI products which includes CTGT’s AI risk management platform, Catio’s AI-powered tech stack optimization, and Solo.io’s cloud-native application networking solutions with their new kagent framework for building AI agents in Kubernetes.
Looking Ahead: The Autonomous Enterprise Economy
The conference painted a picture of an approaching “$1 trillion post-SaaS economy” where autonomous agents don’t just augment human work but they fundamentally reimagine what enterprise software can be. The companies presenting weren’t talking about future possibilities, they’re deploying these systems today, achieving measurable ROI, and scaling across their organizations.
The Speed Advantage
As Windsurf’s Mohan noted, “If we don’t innovate and do crazy things, we’re going to die,” emphasizing the startup’s need to push boundaries. The company organizes into lean squads of three or four engineers, each focused on testing narrow product hypotheses, allowing rapid iteration in a space where foundational AI models and user needs evolve at breakneck speed.
The Competitive Imperative
As industry analysis shows, leading financial institutions like JPMorgan Chase and Capital One are dominating the AI arms race through significant talent acquisition and infrastructure investments. As these AI leaders continue deploying at scale, they’re gaining competitive advantages that will become increasingly difficult to match.
The Bottom Line
The conference made clear that the question isn’t whether agentic AI will transform enterprises, but how quickly organizations can adapt their infrastructure, processes, and culture to harness its potential. Those who move decisively in 2025 will likely define the competitive landscape for the next decade.
As the conference theme emphasized: “AI hype is cheap. Execution is rare.” VB Transform showcased organizations that have moved beyond the hype to real, measurable business impact through agentic AI systems. The transformation from experimental AI to production-ready autonomous agents isn’t a future prediction—it’s the current reality for industry leaders who are already reaping the benefits of this technological revolution.
