Plurai wants to help AI agents work better using the same approach that helped get self-driving cars to roads, and more

0:01This is TechCrunch.

0:10This episode is brought to you by Framer. Whenever the end of the year creeps up, things get hectic. You've got holiday sales to plan and projects to carry over the finish line. The last thing you should have to worry about is your website. With Framer, you don't have to. Framer is the AI website builder that helps creators, teams, and businesses ship production-ready sites faster than ever while getting every detail right. Prompt, inspect, edit, and publish in one place at a whole new pace. And it's easy to improve and update things, even after launching. With their AI agents and your team working together, managing a website is a cinch.

0:50So you can focus on other important things, like savoring that last bit of pumpkin pie. From early startups to Fortune 500 companies, thousands are turning to Framer to build a website they'll love. Learn how you can get more out of your site from a Framer specialist or get started building for free today at framer.com slash tcstartups for 30 % off a Framer Pro annual plan. That's framer.com slash tcstartups for 30 % off. Rules and restrictions may apply. Now, it's no secret that AI agents are all the rage, but they're not quite as capable as their purveyors would have you believe. Efforts to benchmark agents tasked with complex work show that they are not typically reliable enough to function without close supervision.

1:39LiveBench 2.0 showed that Claude Fable 5.2 succeeded on 82.9 % of its tasks, while ChatGPT 6 Astra succeeded on 75.6 % of its tasks. and an effort by the AI infrastructure company Eon to test agents in a simulated corporate environment found the most successful model tested, Claude Opus 4.8, performed tasks at 76.8 % accuracy. Now, a startup known as Pluri wants to boost those numbers with the same techniques that put self-driving cars on the street. Simulations using synthetic data based on a company's own digital systems meant to prepare models to tackle difficult edge cases. Co-founders Elon Qadar and Elad Levy previously worked for Nexar, which develops autonomous driving technology.

2:35Qadar worked with Waymo to simulate and evaluate its end-to-end deep learning driving software. And he told TechCrunch, The moment that we saw the wave of AI agents, we saw the same pattern, that it is very easy to build a demo, but hard to get something that will be reliable and cost-effective. Since we saw the same pattern, the same pitch was that we want to build this simulation layer. So, on Wednesday, the startup said it had raised$13 million in seed funding to make that happen. The round was led by Team 8 with participation from UNI, Velocity, and Mercer. Like the end-to-end models trained to drive cars, AI agents are probabilistic systems operating in unpredictable environments.

3:21When pressed by customers, agents tasked with booking travel can ignore spending policies. Or, when tasked with querying one database, agents can reliably provide an answer, but combining information from multiple data sets tends to gum up the works. A long conversation with a customer service bot can overwhelm an agent with too much context and lead to wrong turns. So to fix that, Pluri creates a digital twin of the company's electronic systems, including the knowledge base and tools that an agent will have access to in production. It then maps all the possible flows of the agents and uses that graph to prepare the model with evaluations and fixes for possible edge cases.

4:06The platform works alongside existing agentic tools like ClaudeCode, Codex, Copilot, and Cursor. However, Kadar says the company's real promise is using the data it gathers to post-train small, open-weight models tailored to individual customer needs. Tests performed in cooperation with NVIDIA and Nebius showed the company's models outperforming mid-tier frontier models at lower costs. He said, we could take a large language model, we could take the task, decompose it into a smaller task using the simulation, train an open-weight model, and using the simulation, verify that the quality is higher or comparable to the frontier model, and an order or two orders of magnitude cost reduction.

4:53Pluri hopes analogies between AI agents in physical and digital environments can prove predictive, While powerful general AI models are imbuing robots with fascinating capabilities, they are not yet reliable enough to be deployed widely. Meanwhile, robots powered by AI models built for specific tasks are in the field right now, working for customers and gathering valuable real-world data. Kadar said, My view is that no matter how smart an open AI or anthropic model will be, and they will be much smarter, They will not know your business and your policy. OpenAI has acquired an AI travel agent called Odessia.

5:37That's according to Odessia's founder, Francis Davidson. Davidson said in a post announcing the deal that he's now working for OpenAI, helping the AI lab with the future of travel. Terms of the deal were not disclosed, but this looks like an acquihire. hire. Davidson's announcement focused largely on how he built his AI travel agent Odessia by directing teams of AI agents. The year-old Odessia raised a$6 million pre-seed round led by Sequoia and launched its AI travel agent about five months ago. Davidson said that it had already helped early adopters plan millions of dollars worth of travel, but the tool has already been shut down.

6:21An OpenAI spokesperson confirmed the hire to TechCrunch. The AI lab has not announced new travel-specific products or plans, though. It does, however, count such companies as Bookings.com, Expedia, and Kayak as customers or partners. Travel is also one of the main areas that people are using personal AI assistance to help with. And back in July of 2025, when OpenAI launched its ChatGPT agent, it mentioned travel as one of its potential uses. OpenAI's spokesperson did not have more to share on its plans for travel-related products, except to say that travel is one of the areas that chatbot users frequently ask about.

7:04The spokesperson added that 1.2 billion people turned to ChatGPT each week for help with everyday tasks, from understanding their health, figuring out what to make for dinner, building a budget, and even planning a trip. We see a real opportunity to make these experiences even more useful. Acquiring Odessia is interesting for another reason. Davidson is pretty well known in the travel industry after he founded a travel startup called Sonder as a college student in 2014. It was sort of like Airbnb, but it focused on short-term leasing of upscale apartments in major cities worldwide. The pandemic badly hurt it, and as it climbed out of that rut, it went public via a SPAC, but did not raise as much money as it needed.

7:50When a key partnership with Marriott fell through, the company went into bankruptcy and it wound down. The bankruptcy trustee sold Sonder's name, trademarks, and website domain this year to Travel AI, and that website operates as a curated guide to the kind of travel rental properties that Sonder was once known for. Prior to its downfall, Davidson had built Sonder into a unicorn with $143 million in revenue. While Sonder was not ultimately a success story, Davidson is the kind of battle-scarred founder that Silicon Valley often values. So we'll have to wait and see what OpenAI has in store now with travel, now that he has joined the company.

8:36That's all for now. For the latest in startup tech news, go to techcrutch.com.

9:07executives that understand them. Tune in for insights and a long-term perspective on investing and of course, stock ideas, plenty of them. To quote a listener, it pays to listen. Check us out and subscribe wherever you listen to podcasts.