AI Security Must Focus On Recovery, Not Just Prevention: Rubrik’s Vipul Nayak

The global B2B technology landscape is at an inflexion point as the sector is now pivoting from traditional software architectures to agentic SaaS, or ‘outcome-as-a-service’. However, the economics of this shift remain challenging. 

According to ICONIQ Capital’s 2026 State of AI report, AI-native B2B startups operate with gross margins of just 52%, with model inference costs alone accounting for 23% of revenue. As a result, heavy token consumption is emerging as a major constraint on profitability.

As a result, a number of industry leaders argue that building sustainable agentic products requires more than powerful models. The real differentiator lies in designing robust workflow guardrails that keep costs under control while ensuring reliable outcomes. 

This theme took centre stage at a panel discussion titled ‘How SaaS Companies Are Building Agentic Products Without Breaking Margins’ at the Inc42 AI Summit 2026.

Moderated by Accel VP Anagh Prasad, the session featured MoEngage cofounder and CTO Yashwanth Kumar, KOGO CPO and cofounder Praveer Kochhar, and Rubrik’s head of enterprise sales Vipul Nayak. 

The panellists shared practical insights on building, deploying, and scaling agentic AI systems safely while maintaining healthy unit economics.

From Stochastic Models To Deterministic Guardrails

For agentic AI systems to deliver meaningful enterprise value, they must function as more than sophisticated chatbot interfaces. Left unchecked, autonomous workflows can quickly create operational and financial risks. MoEngage’s Kumar noted that an unsupervised development agent can rack up thousands of dollars in infrastructure costs within a single day. 

To mitigate these risks, enterprises need robust governance frameworks that clearly define which data agents can access and which actions they are authorised to perform, whether that involves reading, modifying, or deleting information.

At the same time, granting agents greater autonomy introduces a new category of security challenges. Rubrik’s Nayak argued that because AI agents operate at machine speed, errors and unintended actions can occur far faster than traditional, human-led security systems can respond. As a result, organisations must focus not only on prevention but also on recovery and resilience.

“90% of AI security is entirely new for everybody,” Nayak said. “Traditional security might give a 99% guarantee, but resilience is fundamentally different. If an agent hallucinates, can you bring the whole infrastructure back to a clean state? A plan to bounce back is what we all have to worry about today.”

The panellists argued that building sustainable, margin-positive agentic products requires a shift in engineering philosophy. Rather than relying solely on prompt engineering, companies must wrap unpredictable foundational models inside predictable software architecture. 

By combining probabilistic reasoning with deterministic guardrails and strict access controls, enterprises can capture the productivity benefits of autonomous agents while limiting operational, security and cost-related risks. 

Ultimately, the discussion concluded that while foundational AI models may increasingly become commoditised, the true competitive advantage will lie in the control layer that governs how those models are deployed, monitored and managed within enterprise environments. 

The post AI Security Must Focus On Recovery, Not Just Prevention: Rubrik’s Vipul Nayak appeared first on Inc42 Media.