AI agents are poised to replace standalone LLMs in the enterprise, if Ilan Twig CTO and co-founder of Navan, is right.
“My recommendation to any CXO out there, COOs, CIOs, is you must embrace this technology. There’s no other way,” he said.
Twig shared his perspective in an interview with InformationWeek after his presentation at the recent Momentum AI conference in New York City. During the interview, Twig said Navan, a corporate travel booking SaaS platform, used OpenClaw in the development of its recently announced TravelClaw agentic layer in development.
OpenClaw is an open source, autonomous AI agent. Navan’s TravelClaw is currently only used by Twig himself during development and testing. When released at scale, TravelClaw is expected to proactively contact its users via email and apps to address travel issues, rather than waiting for prompts, as chatbots do. Twig said TravelClaw would function continuously in the background, monitoring trip details.
The agentic wave Twig foresees certainly aligns with the bullish comments from Nvidia CEO Jensen Huang, who has urged CEOs to embrace an OpenClaw vision for their companies. Navan’s aggressive pursuit of agentic and other AI technology is part of Twig’s ongoing strategy to expand what a relatively small team can accomplish with AI tools.
And despite his “LLMs are dead” rhetoric, Twig acknowledged that the introduction of ChatGPT had a seismic effect.
“A lot of people in the company know that I lost eight pounds researching this technology. I didn’t eat. It’s a true story,” Twig said.
He explained how his first step to get his arms around ChatGPT’s capabilities in late 2022 was to procure “a super expensive GPU,” which took a bit of effort. “I reached out to my co-founder. I’m like, ‘I need $30,000 to build a PC.’ And he’s like, ‘What?!'”
After trying to explain the need, Twig said he eventually secured the hardware himself because — as a CTO — he had firsthand experience with ChatGPT before deploying it to production within Navan.
Twig compared his approach to vetting the chatbot to interviewing job candidates before hiring them. This included switching to ChatGPT’s uncensored models, which did not go through the last phase of training on what is right and wrong, or how to be politically correct. “I don’t want to get into politics, but I really wanted to see the nature of it,” he said.
Those experiences helped him find ways to make the technology safer and more secure to protect the organization and its customers.
It also cemented his bullish take on agentic AI.
All-in for the rise of AI agents
The current craze for agents is relatively new in the AI hype cycle. Twig said that in the early days, no one really discussed AI agents, but the term took off in January 2025 at the World Economic Forum Annual Meeting. “I was there in Davos, and that’s what everyone was talking about. Agentic, agentic, agentic,” he said.
Wayfair CTO Fiona Tan also discussed AI agents at the Momentum AI event, and how they are being deployed across retail consumer goods, and other sectors.
Twig urged his peers to familiarize themselves with technology such as OpenClaw, even if it means working with it on an isolated machine — the way Navan’s TravelClaw is being developed. “Do not use LLMs; use agentic systems.”
He’s convinced that OpenClaw could be the start of a substantial AI revolution.
“To me, it is as big as ChatGPT when ChatGPT [first] came out. I just don’t think that most people realize it … Interview me in two years, and you will see that everything is in the OpenClaw spirit,” he said.
There are a number of advantages and advances Twig found in his exploration of OpenClaw. For starters, he described OpenClaw as making an LLM more proactive and autonomous, shifting its capabilities from features to responsibilities.
That “sense” of responsibility taken has been evident as Twig runs Navan’s TravelClaw agent on his own phone, letting it monitor his travel, itineraries and his location. “It knows everything. It does not wait for me to prompt it to do stuff for me. It prompts me,” he said.
For example, TravelClaw knew Twig needed to get to the airport within a few hours of the interview, and it actually got anxious, so to speak, about arriving on schedule. “That’s the shift to responsibility. It wants to make sure that I will get there on time and will not miss my flight.”
Weighing the freedom and authority granted to agents
The autonomy TravelClaw has includes waking itself up whenever it wants to perform functions in context with Twig’s needs. Navan’s AI-powered travel assistant, Ava, also talks to TravelClaw without him knowing about it in order to solve issues on his behalf. For example, if he wanted an aisle seat but it is not available initially, TravelClaw will continue its booking efforts on its own.
“It would wake itself up every day to talk to Ava and check if there is an aisle seat available on the flight,” Twig said.
If such a seat became available, TravelClaw would check with its users via messaging apps such as Telegram, WhatsApp, iMessage or Slack.
Twig also gave TravelClaw a bit of a challenge by asking it to book a restaurant even though the Ava personal assistant does not have that feature yet. So TravelClaw tried to contact a live human agent to complete its task, which led engineers to intervene.
“It was really funny. This is how far the agent would go to solve my problem. It doesn’t care. It has responsibility,” Twig said.
That proactive nature, however, must be controlled. He warned that TravelClaw could destroy his computer, so at this time, he has not granted it freedom across the enterprise. As the world learns to work with agents, Twig said he believes Navan’s early start with the technology has prepared it for future possibilities. “I’m ready for the next thing.”
