No Priors Ep. 134 | With Palo Alto Networks CEO Nikesh Arora
Nikesh Arora, CEO of Palo Alto Networks, discusses the transformative impact of AI on search, enterprise business models, and cybersecurity. He explores the shift from search to agentic AI, the challenges and opportunities for enterprise adoption, and how AI is fundamentally compressing cyberattack timelines while enabling new, data-centric defense strategies.
The Evolution of Search and the Rise of Agentic AI
The paradigm of search is shifting from information democratization to intelligence democratization. For two decades, the focus was on making all information accessible. Now, users demand synthesis and sense-making from the vast sea of data. This is the core function of modern generative AI models.
The future of search lies in answering a user’s intent, not just their typed query—a vision Larry Page had 15 years ago. While Google is well-positioned to transition its product with its distribution, product capabilities, and AI talent, the primary challenge is the business model. The current ad-based model, built on presenting “ten blue links,” must evolve. The likely transition is towards a model based on consumption or consummated transactions, where AI agents act on behalf of the user.
“Maybe the business model transition is stop giving me leads, give me consummated transactions through agents. Maybe I’ll get paid more to buy you the airline ticket directly than have you be able to find an airline ticket provider.”
The Disruptive Power of AI Agents
AI agents represent a more significant disruption than generative AI itself. For the last 30-40 years, product development has centered on UI design, creating interfaces for humans to interact with backend engineering algorithms. Generative AI and natural language are already changing this, but agentic AI takes it a step further.
If an agent can perform a task directly (e.g., an “Uber agent”), it obviates the need for a human to interact with the UI. This poses a threat to the vast number of applications that are essentially transaction-fulfillment UIs. The most vulnerable businesses are those with low brand loyalty tied to their UI, acting as a thin front-end to a transaction processing system. The central question becomes: who owns the top-level agent that orchestrates these tasks?
AI in the Enterprise: Adoption and Business Models
In the enterprise, there’s a lower tolerance for the inaccuracies that might be acceptable in consumer applications. Giving an LLM the autonomy to perform a critical task like shutting down a server is not yet feasible. Currently, enterprises use AI with a “human in the loop,” primarily for summarization and knowledge work, not for precision tasks or actions.
The enterprise business model for AI is unlikely to be a simple “unit of work” charge until precision and accuracy are guaranteed. A more probable evolution is a form of “AI as a Service” (AIS), where AI-first workflows are designed end-to-end.
For AI applications to have long-term viability in the enterprise, they must move beyond being a simple wrapper on a large model. They need to integrate AI capabilities with a system of record. An LLM itself won’t become the system of record; that will remain a proprietary, locked-down database. The opportunity lies in building AI interfaces that maintain and create these systems of record.
The New Frontier of Cybersecurity in the AI Era
1. Compressing Attack Timelines The single biggest threat from AI in cybersecurity is the compression of attack timelines. A bad actor can use AI agents to scan an enterprise, identify vulnerabilities, simulate attack vectors, and exfiltrate data in minutes.
“When I started seven years ago the average time to identify a target get through it and exfiltrate data was in the three to four day time frame. The fastest we’ve seen right now is 23 minutes.” This drastic reduction in attack time means that defensive response times must, by physics, become near-instantaneous.
2. A Fundamental Shift in Defense Strategy Traditional cybersecurity focuses on stopping the “known bad.” However, major breaches occur from the “unknown bad.” The key is to shift from fragmented, feature-based security to a consolidated data platform approach. Instead of just stopping known threats at the sensor level, the goal is to collect and analyze all data in real-time to understand context and identify anomalous behavior.
This is a move away from providing customers with a list of 5,000 suspicious events to investigate. Instead, by having the full context of the enterprise data, an AI-powered system can determine what actually happened and automate the response.
3. New Attack Vectors and Countermeasures Social engineering attacks, responsible for the vast majority of breaches via credential theft, are becoming more sophisticated with deepfakes and advanced spearphishing. This will render many traditional authentication methods obsolete.
The defense cannot solely focus on preventing identity takeover. The more robust strategy is to focus on detecting anomalous behavior.
“Suddenly Sarah decided to exfiltrate all the data in Elad’s company. Even though she used to do email with him every day, today suddenly she’s logged in and she’s downloading everything onto her laptop… That sounds pretty suspicious.”
This leads to a new security paradigm based on just-in-time rights and continuous behavior analysis, rather than granting persistent access after a one-time check-in.
AI’s Impact on the Workforce and Organizational Efficiency
AI’s impact on jobs will be segmented.
- Product Development: AI will not replace good product developers but will make them faster and more innovative. It will help eliminate technical debt and accelerate development cycles, leading to higher quality products. The anxiety about AI generating more low-quality code is a short-term problem; as the technology matures, it will lead to better, more efficient, and more secure code.
- Sales: High-level enterprise sales, which rely on human relationships and trust, are unlikely to be replaced by AI agents.
- Customer Support & Repetitive Tasks: This is the area most ripe for transformation. Customer support largely exists because of complex or flawed products. AI can dramatically improve product quality, diagnostics, and user onboarding, with the potential to automate 80-90% of current support functions over the next five years. Large-scale, repetitive administrative tasks are also prime candidates for replacement.
Leadership and Strategy for Growth at Scale
- Platform Strategy and Ambition: The cybersecurity industry is fragmented, with enterprises often using over 100 different vendors. Like the ERP or CRM industries, it is destined for consolidation onto a single platform for end-to-end visibility and integrated functionality. The goal is to become that platform of choice.
- M&A as “Distributed R&D”: In a highly innovative market, M&A is treated as a form of product development. The strategy involves acquiring the #1 or #2 player in a category—never a #3 or #4 to “spit and shine”—and making the leaders of the acquired company the leaders of that business unit within the larger organization. This leverages their proven resourcefulness and hustle.
- The Importance of Communication: A leader’s primary job is to set the North Star, get buy-in on the “why,” and then act as a shield to remove friction for their teams. Communication is an underrated and critical component. It’s essential to constantly check that the strategic “why” is understood at all levels of the organization, as it often gets lost in translation.