Nobody Has a Good Answer for Agentic Identity Yet (fusionauth.io)

At the recent LDX3 conference in New York City, discussions around AI highlighted significant concerns among engineering managers regarding the implications of AI-written code on team dynamics and developer skill development. With AI tools becoming integral, there’s unease about defining roles, particularly for senior engineers, and how to effectively nurture junior engineers' growth without traditional learning experiences. A notable focus was on "agentic identity"—the representation of agents that operate on behalf of users in systems like Notion, raising questions about access control and authorization. The technical challenges presented revolve around implementing secure access for AI agents in third-party applications. Managers discussed two approaches: using granular access controls (rarely available) or building a proxy gateway to handle token exchanges securely. The latter involves creating a unique agentic identity and managing tokens carefully to ensure that only appropriate access is granted without compromising user data. Alarmingly, a survey conducted at the conference revealed that 69% of teams lack plans to measure the efficacy of their AI tools, and 24% of developers use personal credentials for AI tooling, risking audit trails and accountability when mistakes occur. The overarching message emphasizes the necessity for distinct identities for agents to maintain security and clarity in AI interactions.