Agentic AI

QCon San Francisco 2026 features engineers from companies like OpenAI, Netflix, and Airbnb discussing how to build and operate production systems as AI agents assume larger roles. The event focuses on the practical challenges and strategies for evolving infrastructure to support autonomous AI workflows. Speakers will share insights on managing the increased complexity and reliability requirements introduced by these systems.

• Practitioners from major tech firms share agentic system strategies.
• Focus on building production-ready infrastructure for AI agents.
• Discuss operational evolution as agents take on larger roles.
• Event highlights reliability and management of autonomous workflows.

OpenRig is an open-source system that wraps AI coding agents like Claude Code and Codex into persistent, role-based teams. It uses YAML definitions to boot agents as a single managed system, maintaining shared context and work ownership. A lead agent coordinates specialists to deliver structured outcomes rather than isolated terminal sessions.

• Orchestrate multiple AI models as a unified, persistent team
• Define agent roles and structure via simple YAML configs
• Maintains shared context across different agent sessions
• Uses a lead agent to coordinate work and deliver results

Uber Eats has overhauled its search pipeline, achieving a 50% reduction in end-to-end latency. Key architectural changes include parallel hydration, reduced retrieval work, and an infrastructure redesign. The team is also leveraging an agentic coding workflow to accelerate development and optimization efforts.

• Parallel hydration and reduced retrieval work drive the 50% latency cut.
• Agentic coding workflows are accelerating the engineering cycle.
• Experimental optimizations include HTTP multipart streaming and microbatching.
• Product-based retrieval is under exploration for future performance gains.

AI agents can now synthesize public clues to generate working exploits, drastically reducing the time between disclosure and attack. This capability renders traditional disclosure embargoes in open source projects largely ineffective. The window for mitigation is closing, forcing a shift toward rapid patching strategies.

• Traditional embargo periods no longer guarantee safety against automated exploit generation.
• Open source maintainers must prioritize speed over process in patch releases.
• Publicly available vulnerability clues are now sufficient for AI-driven exploitation.
• Security workflows need to assume immediate, automated exploitation upon disclosure.

Security researchers discovered a chain of vulnerabilities in the Zammad ticketing system that enabled attackers to hijack user sessions and execute code. The exploit chain culminated in full root escalation, allowing adversaries to steal sensitive data such as email addresses from security research organizations. This incident highlights critical risks in widely used IT service management platforms when authentication and authorization controls are bypassed.

• Audit Zammad deployments for known chained vulnerability paths immediately
• Enforce strict session management and token rotation policies
• Monitor for unauthorized code execution indicators on backend servers
• Review root access controls and minimize privilege escalation vectors