85% of enterprises are running AI agents. Only 5% trust them enough to ship.

Cisco President and Chief Product Officer Jeetu Patel recently discussed the challenges facing enterprises when it comes to moving AI agents from pilot programs into production. In an exclusive interview at the RSA Conference 2026, Patel highlighted the importance of trust in bridging the gap between pilot and production stages for AI agents. He emphasized that establishing trust is crucial for ensuring the success of AI agents in enterprise settings, as it can mean the difference between market dominance and bankruptcy.

According to a recent Cisco survey, 85% of enterprises have AI agent pilot programs in progress, but only 5% have successfully transitioned these agents into production. Patel pointed out that the main issue hindering this transition is the lack of a robust trust architecture. He compared AI agents to teenagers, noting that while they possess intelligence, they lack the maturity and accountability required to make critical business decisions. Without proper guardrails and oversight, AI agents can easily make mistakes with significant consequences.

To address the trust deficit, Cisco introduced several new products and initiatives at the RSA Conference 2026. These included AI Defense Explorer Edition, an open-source tool for red-teaming AI agents, and the Agent Runtime SDK for embedding policy enforcement into agent workflows. Additionally, Cisco collaborated with Nvidia to launch OpenShell, a secure container for open-source agent frameworks, which integrates seamlessly with Cisco’s Defense Claw framework for enhanced security.

One of the key announcements from Cisco was the shift towards a zero-human-code engineering mandate. Patel revealed that Cisco is aiming to have 70% of its products built entirely by AI by the end of 2027. This ambitious goal signifies a significant cultural shift within Cisco’s engineering organization, where employees are expected to code with AI or risk not working for the company.

In terms of security, Patel outlined five strategic advantages that winning enterprises should focus on, including sustained speed, trust and delegation, token efficiency, human judgment, and AI dexterity. He emphasized the importance of telemetry in verifying the effectiveness of these strategic advantages, noting that enterprises need to be able to differentiate between agent-initiated actions and human-initiated actions to ensure proper oversight and control.

Overall, Patel’s vision for the future of AI agents in enterprise settings revolves around building a strong trust foundation, leveraging innovative technologies like AI and open-source frameworks, and empowering security teams to verify and validate the effectiveness of their security measures. By following a proactive approach and embracing the principles outlined by Patel, enterprises can enhance their security posture and successfully navigate the complexities of the agentic workforce.