AI in Production: What Breaks, What Works, and Who Approves it? | InfoQ Webinar

InfoQ will hold a free, 60-minute live panel titled "AI in Production: What Breaks, What Works, and Who Approves It?" on Wednesday, October 14, 2026, running from 4:00 PM London time, 11:00 AM ET, 8:00 AM PT. The session is a panel discussion, not a single-speaker talk, and attendees can direct the conversation by submitting questions in advance through a free-text field on the registration form or live via Zoom Q&A. Everyone who registers receives the recording afterward, so the live slot is not the only way to watch. The announcement is credited to Artenisa Chatziou.

The framing InfoQ gives the session is the tension between two groups inside the same organization. Platform and coding-agent teams want agents to act faster and do more; privacy and security engineers want tighter limits on what those agents can reach. The panel is positioned around where teams draw that line and what happens when they draw it in the wrong place, whether by granting too much autonomy or by imposing controls that block useful work. InfoQ says the discussion is seeded by questions the moderator and panelists prepare together, and that the aim is to surface reasoning and tradeoffs behind different approaches rather than present one blueprint.

The agenda lists five concrete areas. First, agent autonomy: which actions an agent may take on its own and which still need a human to approve. Second, verification: how teams check AI-generated changes before they enter CI or reach production. Third, data exposure: which sensitive or regulated data a system can access and which threats get modeled first. Fourth, production RAG and platforms: the architectural and operational practices that hold up after a controlled demo ends. Fifth, making the internal case: how engineers argue for stronger controls before an incident forces the decision.

Five practitioners are listed. Hien Luu chairs QCon AI New York 2026, has led machine-learning platform teams at DoorDash and Zoox, wrote MLOps with Ray, and facilitates the InfoQ Certified AI Engineering Program; his stated perspective spans RAG pipelines, agents, AI platforms and reliability. Katharine Jarmul specializes in privacy and security for machine learning and AI systems, wrote Practical Data Privacy for O'Reilly, and facilitates the InfoQ Certified AI Security & Privacy Engineering Program, covering sensitive-data handling, threat modeling, guardrails, observability and governance. Zichuan Xiong is a Principal at Thoughtworks working across domain-driven design, data mesh, organizational change and agentic systems, co-facilitates the InfoQ Certified AI-Assisted Engineering Program, and is slated to discuss the engineering harness around coding agents, including how agent-generated changes are verified before moving through delivery workflows. Premanand Chandrasekaran is a market tech director at Thoughtworks with close to 30 years leading software teams, co-author of Domain-Driven Design with Java, and co-facilitates the same AI-Assisted Engineering Program; his part covers permissions, review gates and governance when agents act across the software delivery lifecycle. Lan Chu is an AI tech lead and senior data scientist with more than seven years building production data and ML pipelines and more than three years building generative AI products; at QCon London 2026 she presented "Beyond the Demo: RAG Is Easy, Production RAG Is Not," drawing on production systems built on more than 10,000 documents.

For developers and product teams, the value here is likely less a reusable blueprint than a set of comparisons. Review gates for agent-written code, permission scoping for tool-using agents, and retrieval over internal documents are all areas where practices are still forming, and the panel mixes platform, security and delivery perspectives that often sit in different meetings. Watching how these five describe their own boundaries, and where they disagree, is the practical takeaway; the Q&A and advance question form are the levers for steering the session toward a specific stack or regulatory context.

Why it matters: teams shipping agents and RAG pipelines are being asked to justify autonomy and data access decisions now, often without settled norms. A free, recorded session with named practitioners gives engineers a low-cost way to benchmark their own controls and to borrow language for the internal argument the agenda explicitly calls out. The main limitation is format: an hour across five panelists and a broad agenda means each topic gets limited depth, so viewers with a narrow problem should submit questions in advance and treat the recording as a starting point rather than a decision document.