Latest Interviews
Showing 1–9 of 9 transcripts.
Clear all filters- RAISE Summit21 min
It Takes AI to Break AI: Securing the Agentic Attack Surface | Alice | RAISE Summit 2026
Alice Lead of Product and Engineering Avi Golan explains how enterprises face escalating security and safety risks as AI evolves from static chatbots to autonomous agents with tool-calling capabilities. He details a specialized Agentic AI Red Teaming methodology that moves beyond traditional static scanning to continuously simulate adversarial attacks, ensuring robust threat modeling before production deployment. This approach addresses the unique vulnerabilities of non-deterministic models, such as reasoning manipulation and indirect prompt injection, which have previously led to significant operational failures and legal liabilities across major brands.
- RAISE Summit21 min
The Uncomfortable Math of Frontier AI in Production | Kavitha Mariappan | RAISE Summit 2026
At RAISE, approximately 110 exhibitors, including 18% startups and 30% scale-ups, convened to address the shift from perimeter-based security to resilience against autonomous AI agents that now possess economic agency and emergent capabilities. Recent threat intelligence reveals that 53% of recent attacks are AI-related, with adversaries exploiting embedding space geometry and open-source models to bypass traditional deterministic defenses while 88% of enterprises lack the ability to roll back unauthorized agent actions. Experts recommend a strategic pivot toward assuming breach status by inventorying non-human identities, rotating machine tokens, and prioritizing recovery protocols over prevention to counter the asymmetric advantages held by autonomous threats.
- RAISE Summit23 min
Agentic Development Security | Manoj Nair, Snyk | RAISE Summit 2026
Presenting critical findings on the surge of security risks in AI-driven development, this analysis reveals a 108% increase in vulnerabilities linked to agentic workflows and highlights that current frontier models still produce only 60% secure code autonomously. With threat actors already executing attacks within weeks and open models projected to match proprietary capabilities in six months, the session outlines a strategic blueprint requiring independent validation, pre-build context integration, and the eradication of technical debt before scaling adoption. Urging organizations to shift from fragile remediation to resilient, load-bearing architectures, the discussion emphasizes immediate board-level action to secure supply chains and comply with upcoming regulatory frameworks like the EU AI Act.
- RAISE Summit29 min
The Physics of Thought & Architecture of Intelligence | Amin Vahdat, Google | RAISE Summit 2026
The speaker characterizes the current epoch as the dawn of an "intelligence era" where unbounded human ambition and the Jevons Paradox will drive exponential demand for cognitive computing power, surpassing the historical pace of the Industrial Revolution. To meet this surging need, Google is leveraging specialized hardware like the eighth-generation TPU and adopting a relaxed reliability model to enable agentic computing swarms that coordinate autonomous tasks at unprecedented scale. This technological shift is already accelerating scientific breakthroughs, such as rapid protein folding prediction, while necessitating responsible management of the economic turbulence inherent in this transition.
- RAISE Summit23 min
Alex Ratner, CEO & Co-Founder, Snorkel AI: Data Centric AI in the Agentic Era
Alex, co-founder of Snorkel.ai, argues that specialized agentic systems require robust evaluation benchmarks composed of prompt datasets, rubrics, and verifiers to serve as functional product requirements. His company provides data operations platforms to major US banks and federal agencies while demonstrating that custom-trained models using programmatic evaluators can outperform GPT-4 on complex insurance underwriting tasks. This approach establishes high-quality data and verification logic as the foundational prerequisites for optimizing Reinforcement Learning in enterprise environments.
- RAISE Summit23 min
Atindriyo Sanyal, CEO of Galileo: Making Enterprise Gen AI “Reliable”
Galileo, an AI evaluation platform founded by a former Siri architect, addresses the critical "measurement problem" plaguing generative AI by introducing a specialized observability framework designed to productionize complex RAG and multi-step agents. The company's proprietary Luna models provide sub-200ms real-time inference for dynamic guardrails and qualitative-to-quantitative error analysis, effectively solving the latency and accuracy trade-offs that typically hinder enterprise deployment. This approach enables Fortune 500 organizations to move beyond basic statistics and systematically identify nuanced failure patterns in data quality, tool execution, and security across high-stakes workflows.
- RAISE Summit21 min
Keynote by Caspar Herzberg, AVEVA Systems CEO | RAISE Summit 2024 | Paris
Caspar Herzberg, Kasper Hesberg
Aveva CEO Kasper Hesberg is set to launch in June an industrial AI assistant that merges predictive analytics with Large Language Models to help engineers overcome data silos and access the insights of veterans with 12 years of experience. This solution, already validated in cases like Duke Energy where it prevents $90 million in losses per event, addresses the critical challenge of 70% unused time-series data by summarizing technical manuals and guiding remote collaboration through 3D asset visualizations. By strictly limiting its scope to verified telemetry and documentation, the system delivers high-accuracy maintenance predictions while mitigating the hallucination risks common to general-purpose AI.
- RAISE Summit23 min
'Production Ready RAG' by David Leconte from DataStax | RAISE Summit 2024 | Paris
David Leconte, a Solution Engineer at Datastack, presented a comprehensive framework for transitioning Retrieval-Augmented Generation applications from experimental prototypes to stable, enterprise-grade production environments. The proposed RAG Stack mitigates common industry risks such as library fragmentation and compliance gaps through curated, daily-validated integrations and provides AstraDB, a serverless vector database that delivers 10-millisecond latency and 74% faster retrieval speeds than dedicated competitors. By combining these core components with the newly acquired Longflow workflow engine, Datastack enables organizations to deploy compliant, high-throughput generative AI systems via single-command installation while reducing development time and operational liability.
- RAISE Summit23 min
Production Ready RAG | RAISE Summit 2024 | Paris
David Leconte, a Solution Engineer at DataStax, outlines a strategy to bridge the gap between experimental RAG prototypes and production deployment by introducing a curated solution stack. This approach combines the new RackStack library management tool, the high-performance AstraDB vector database, and the Longflow workflow engine to address critical stability, compliance, and integration challenges faced by major enterprises. By offering verified packages, 10-millisecond latency, and automated support, the platform enables organizations to safely scale generative AI applications in regulated sectors without the risks associated with fragmented open-source frameworks.