The a16z Show Keycard: 2026 is the Year of Agents
234 snips
Jan 8, 2026 
Ian Livingstone, Co-founder and CEO of Keycard, dives into the future of AI agents, citing 2026 as a pivotal year for enterprise adoption. He outlines the complexities of agent security, highlighting the risks of tool-calling and the limitations of current access protocols. The conversation spans the continuum of agents from copilots to autonomous actors and why identity management for these agents is critical. Livingstone emphasizes the need for dynamic, intent-based access controls and discusses Keycard’s mission to streamline agent governance and compliance.
AI Snips
Chapters
Transcript
Episode notes
Agent Returned Other Companies' Data
- A company deployed an agent that returned other firms' data when asked "give me my data."
- Joel de la Garza uses this story to illustrate agent identity and authz failures in production.
Agents Are A Continuum Of Autonomy
- Agents lie on a continuum from assisted copilots to fully autonomous, long-running actors.
- Ian Livingstone compares levels of agency to driving autonomy to clarify adoption stages and risks.
Tool Calls Multiply Exfiltration Risks
- Tool-calling opens new attack surfaces where benign reads become exfiltration via other tools like browsers.
- Ian warns that agents combining DB access with web tool calls produce complex identity and data-leak risks.