The rogue agent problem: A conversation with Gil Elbaz at the Hg Digital Summit

The rogue agent problem: A conversation with Gil Elbaz at the Hg Digital Summit

In this episode, John Cranmer is joined by Gil Elbaz, Co-Founder and Chief AI Officer at Onyx Security, the company building the secure AI control plane for the agentic era. Gil, formerly in the CTO office at Nvidia working on multi-agent infrastructure, argues that the AI attack surface has fundamentally changed. The old boundary between what is yours and what is not has collapsed, and every agent, every MCP, every tool connection is now part of the perimeter. The conversation moves from the rise of shadow AI inside enterprises, where employees reach for any tool that makes work faster, to the harder problems underneath: rogue agents taking destructive actions they were never asked to take, multi-agent communication happening invisibly through Slack and email rather than direct A2A channels, and offensive AI systems like Mythos finding more zero-days in weeks than internal red teams find in years. Gil shares his playbook for getting your arms around it: visibility first, then policy, then runtime protection. As he puts it, the right seatbelt is what lets you drive faster.

Patrick Debois on why context is the new code: A conversation from the Hg Digital Summit

The inventor of DevOps explains why AI needs engineering rigour and human collaboration. The engineer who coined "DevOps" in 2009 thinks he is watching the same pattern play out again, only faster. In this episode of Orbit, Patrick Debois joins Nathaniel Barnes, Hg's portfolio CTO, at our annual Digital Summit in Paris, to map the parallels between the DevOps movement and what is happening right now with AI agents in software teams. The framing question: what if context is the new code? Patrick unpacks the four phases of his context development lifecycle, generate, evaluate, distribute, observe, and explains why the maturity of a company's CI/CD pipeline is the single best predictor of how well it will absorb AI. He and Nathaniel dig into context drift, customer-facing "vibe coding" as a discovery engine, the silos that agents are quietly breaking down, and why the companies winning right now are the ones that skipped the 27-step plan and just started. As Patrick puts it: "We are at the age where it doesn't need to be perfect. We just need to be there." Essential listening for any technology leader navigating this moment.

Jonathan Sanders, CEO of Light: fear is not a strategy

When Jonathan Sanders pitched Light to investors, the reaction was unanimous: rebuilding ERP would take a decade and a hundred million dollars. He did it in two years, with AI at the core. Today Light powers AI unicorns including Lovable and Legora, and Jonathan joins Hg director Soren Holt to explain how. The conversation goes well beyond automation. It's a builder's account of a first-principles redesign. Jonathan and Soren cover the move from template-based OCR to context-aware agents, why eighty percent of AI's value comes from doing things that were previously impossible rather than making old tasks faster, and how Light's customers now sweep entire transaction populations during audits rather than sampling. Jonathan walks through Light's hackathon programme with CFOs, the operating model behind a remote engineering organisation, and the shift from finance as operator to finance as orchestrator. The line that lands hardest: "Fear is not a strategy." An essential listen for any finance leader thinking about what 2030 actually looks like.

Evan Goldberg of NetSuite: 3 decades and 2 platform shifts

Evan Goldberg co-founded NetSuite in 1998 with a five-minute phone call to Larry Ellison when his "graphics stuff" wasn't going great. He pioneered cloud computing before SaaS existed, sold to Oracle for $9 billion, and stayed at the helm for a decade more. Now at 40,000 organisations and $4 billion ARR, Evan has presided over two tectonic platform shifts and says AI dwarfs both. Recorded at Hg's 'Office of the CFO' dinner in New York back in February, Goldberg explains why autonomous close agents that find exceptions while you sleep are "never going back" territory, his perfect metaphor for why AI still needs integrated data ("cell phones don't help you speak French"), and why customers "have no interest in becoming ERP hobbyists." He reveals he quit Oracle for MIT's AI lab 30 years too early, why mixing deterministic systems with probabilistic AI is the answer, and his tactical advice for founders: step back from strong growth numbers to spot destructive trends, then make painful changes even if they crater quarterly results. From a coffee-stained computer to global success, this is founder wisdom about building companies that last through multiple technology revolutions.

Marjorie Janiewicz of Mistral AI: flipping the adoption curve - why SaaS with data can win in an AI world

Marjorie Janiewicz has sold enterprise software through every platform shift for three decades: Oracle, MySQL, SAP, MongoDB, HackerOne. Now as Chief Revenue Officer at Mistral AI, she's taking a French startup from zero to $400 million in 18 months, closing deals with ASML and HSBC in under a year - timelines that used to take half a decade. Recorded in New York as Mistral announced its Finance offering, this conversation addresses what's working versus what's theatre in enterprise AI. Marjorie supports the MIT hypothesis: 95% of AI projects never reach production. Chatbots drive adoption but don't change businesses. The 5% that work? They start with one high-impact use case, they customise models with proprietary data, and they deploy on-prem where regulated data lives. She explains why the adoption curve flipped, why SaaS companies sitting on data can win if they treat AI as transformation not automation, and why Mistral bet on 400 forward deployment engineers instead of just shipping models. From prototypes done in 48 hours to why "sovereignty is just marketing, independence is what matters," this is pattern recognition from someone who's been in the room when the shift happens repeatedly. Whether you're a SaaS company worried about agents or trying to sell AI to enterprises struggling with ROI, Marjorie's earned her perspective.