The conversation around artificial intelligence in the corporate world is shifting. For the past two years, the primary focus for executives has been the integration of Large Language Models (LLMs) to handle text generation, summarization, and basic customer interaction. However, a new architectural philosophy is emerging—one that moves beyond the prompt-and-response nature of chatbots toward something far more autonomous.
At the recent Momentum AI conference in New York City, Ilan Twig, the CTO and co-founder of Navan, presented a bullish case for this transition. Twig argued that the era of standalone LLMs in the enterprise is giving way to agentic systems for enterprise, urging corporate leaders to pivot their strategies before they are left behind by the rapid pace of autonomous innovation.
For those unfamiliar with the terminology, the distinction is critical. While an LLM is essentially a sophisticated prediction engine that responds to a user’s input, an agentic system is designed to act. It doesn’t just provide an answer; it pursues a goal, interacts with other software, and makes decisions to complete a complex workflow without needing a human to guide every single step.
Twig’s message to his peers was blunt. “My recommendation to any CXO out there, COOs, CIOs, is you must embrace this technology. There’s no other way,” he stated during the event. His perspective suggests that the ability to execute tasks autonomously will be the primary differentiator for competitive enterprises in the coming years.
From Reactive Chatbots to Proactive Agents
The fundamental limitation of the current LLM-driven landscape is its reactive nature. A user must provide a prompt, and the AI provides a response. In a corporate travel context, this might mean a traveler asking a chatbot if their flight is delayed and then asking how to rebook it. This process, while faster than a phone call, still places the burden of initiation and management on the human user.
Navan is attempting to solve this by developing an agentic layer called TravelClaw. Built using OpenClaw—an open-source, autonomous AI agent—TravelClaw is designed to operate in the background, continuously monitoring trip details rather than waiting for a user to start a conversation.
The goal for TravelClaw is a shift toward proactive resolution. Instead of a traveler discovering a flight cancellation and then prompting an AI for help, the agentic system would identify the issue in real-time and proactively contact the user via email or app to address the problem. By moving the AI from a “tool” that is used to a “system” that monitors and acts, the friction of corporate travel management is significantly reduced.
Currently, TravelClaw is in the development and testing phase, used primarily by Twig himself. However, the vision for its scaled release is to create a seamless, background-operated experience where the AI anticipates needs and resolves conflicts before the user even realizes a problem exists.
The ‘LLMs are Dead’ Rhetoric
Throughout his presentation and subsequent discussions, Twig employed provocative rhetoric, suggesting that “LLMs are dead.” To a casual observer, this seems contradictory given that agentic systems are actually powered by LLMs. However, from a software architecture perspective, Twig is arguing that the LLM should no longer be the final product delivered to the end-user.
In this view, the LLM is the engine, but the agentic system is the vehicle. Using a standalone LLM for enterprise tasks is like buying an engine and expecting it to drive you to work; you still need the chassis, the steering, and the navigation system to make it useful. Agentic systems provide that structure, allowing the AI to use tools, access real-time data, and execute multi-step plans.
Despite this aggressive stance on the evolution of the technology, Twig acknowledged the foundational importance of the current wave of AI. He noted that the introduction of ChatGPT had a “seismic effect,” providing the necessary proof of concept and public interest that paved the way for more complex, autonomous agents.
Strategic Implications for the C-Suite
The push toward agentic AI is not just a technical preference; This proves a strategic move to maximize human capital. Twig indicated that Navan’s aggressive pursuit of these technologies is part of a broader strategy to expand the capabilities of a relatively small team. By delegating complex, repetitive, and monitoring-heavy tasks to autonomous agents, a lean team can achieve the output traditionally associated with a much larger organization.
This vision aligns with broader industry trends. Nvidia CEO Jensen Huang has similarly urged CEOs to embrace a vision centered on OpenClaw and similar autonomous frameworks. The implication is that the “AI infrastructure boom” will eventually move from the hardware layer (chips and servers) to the execution layer (agents that do actual work).
For COOs and CIOs, the transition to agentic systems requires a shift in how they view AI implementation. It is no longer about finding a “use case” for a chatbot to answer FAQs; it is about identifying “workflows” that can be fully automated from start to finish.
Key Differences: LLMs vs. Agentic Systems
| Feature | Standalone LLM (Chatbot) | Agentic System (AI Agent) |
|---|---|---|
| Interaction Model | Reactive (Wait for prompt) | Proactive (Goal-oriented) |
| Operation | Session-based | Continuous background monitoring |
| Capability | Information retrieval/generation | Task execution and tool usage |
| User Effort | High (Must manage the AI) | Low (AI manages the process) |
The Role of Open Source in Autonomous AI
The choice to build TravelClaw on OpenClaw highlights a growing trend toward open-source foundations in the enterprise AI space. Open-source autonomous agents allow companies to customize the “reasoning” and “action” loops of their AI without being entirely locked into a single proprietary ecosystem. This flexibility is essential for companies like Navan, which must integrate with a vast array of third-party travel APIs, airline systems, and corporate policy engines.
By utilizing an open-source framework, developers can more easily audit how the agent makes decisions—a critical requirement for corporate governance, and reliability. In a travel environment, where a mistake in rebooking a flight can result in significant financial loss or operational disruption, the ability to refine the agent’s logic is paramount.
What Happens Next
The transition from the “chatbot era” to the “agent era” is still in its early stages. While leaders like Ilan Twig are already deploying these systems in testing, the broader enterprise market is still grappling with the security and reliability of autonomous agents. The primary challenge remains “hallucinations”—the tendency of LLMs to invent facts—which becomes a much larger risk when an AI has the authority to actually book a flight or move money.
However, the trajectory is clear. The goal is a world where AI does not wait for us to ask for help, but instead informs us that the help has already been provided. As Navan continues to develop and eventually scale TravelClaw, the industry will be watching to see if proactive, background AI can truly replace the traditional prompt-based interface.
The next major milestone for this specific implementation will be the transition of TravelClaw from internal testing by Twig to a scaled release for Navan’s broader user base. This rollout will serve as a real-world test of whether agentic systems can handle the volatility of global travel at scale.
Do you believe proactive AI agents will replace traditional chatbots in your industry? Share your thoughts in the comments below or join the conversation on our social channels.