Nimble ads

Today Nimble is announcing our $47M Series B led by Norwest, with participation from Databricks and all of our existing investors, to make the live web accessible and reliable for AI agents and business decision making. AI agents haven't been limited by intelligence — they've been limited by bad data. Until now. Here's the problem: Google built an incredible search engine; for humans. But enterprises still rely on fragile web scraping or expensive data vendors to feed their mission-critical business decisions. Other AI search engines provide high-level, text-based answers from their indexes, but teams struggle to trust them. Nimble changes that. We've built the Web Search Agent platform that lets users stream clean, structured, reliable data straight from the live web into their workflows. No middlemen. No guesswork. No broken scrapers. We unlock the ability to deploy agents you can actually trust, because the data behind them is verified, governed, and real-time. Hundreds of enterprises already run Nimble in production across pricing intelligence, financial due diligence, media optimization, and more. And this is just the beginning. Come check us out for yourself — link in comments. #AIAgents #AgenticAI #AIInfrastructure #EnterpriseAI #SeriesB #DataInfrastructure #RealTimeData #WebData #MachineLearning #StartupFunding Microsoft for Startups Hetz Ventures Square Peg Target Global Slow Ventures J-Ventures Fund R-Squared Ventures

We’re honored to be nominated as “𝗕𝗲𝘀𝘁 𝗥𝗲𝘁𝗮𝗶𝗹 𝗜𝗻𝘀𝗶𝗴𝗵𝘁𝘀 𝗦𝗼𝗹𝘂𝘁𝗶𝗼𝗻” at the 2026 Vendors in Partnership (VIP) Awards! The VIP Awards celebrate the partnerships and innovation that move retail forward and we’re proud to be recognized among this year’s nominees. Our Web Search Agents help retail and CPG teams see what is really happening on the digital shelf in real time, so this recognition means a lot to us. We’d love your support. Join us at Gotham Hall on January 9th to celebrate! For more information about the event, visit 👉 https://lnkd.in/dUaEEevY More to come soon! #VIPNominee #VIPAwardNominee #VIPNYC #VendorsinPartnership

Agentic AI did not start with MCP or the latest OpenAI launch. One of my favorite examples is Eugene Sayan in the late 90s. Eugene's team started building agents for healthcare enrollment. Not toy projects. Production systems that helped millions of people navigate eligibility and coverage. Strict rules, real consequences, very little glamour. These agents sat in front of more than 10 million people trying to keep coverage, so downtime or sloppy decisions were not an option. If an “agent” made a bad decision there, a family could lose health coverage. That is a very different environment from a chatbot that writes emails. Agentic systems have been around for a long time. They just lived far from launch stages and hype threads. Today, this Goldman Sachs map, on the surface, looks like a picture of AI partnerships. Look longer and it feels like a loop for capital: the same few companies providing chips, cloud, services, and funding to each other. The numbers behind it run into trillions for data centers, power, and hardware over the next decade. Platforms like Anthropic with MCP and OpenAI's agent frameworks keep expanding what builders can try, and VCs now light up whenever they see the word “agentic” in a deck. When I put these two pictures side by side, a question keeps coming up in my head: How much of this new AI buildout will end up in Eugene-style systems that quietly handle real work and how much will stay inside the capital loop on this chart? At Nimble we feel that tension in a different corner of the stack. Enterprise data and AI teams want agents that open live pages, read what changed, and act inside existing tools without breaking when layouts or scripts shift. Bigger GPU budgets do not solve that. It lives in the input layer and the reliability work no map like this really shows. When I look at this diagram now, I try to read it less as a list of “who invested in whom” and more as a reminder of what still needs to be built on top of it: agents that live in real workflows, see the real web, and keep working long after the hype cycle moves on. If you are building agents that need reliable live web context, you can reach out to me and my team here to talk through that input layer: https://lnkd.in/dTSXSvdg

I asked ChatGPT to tell me a joke about coffee five times. Same joke. Every. Single. Time. ☕ “Why did the coffee file a police report? It got mugged!” I wondered if that was the ceiling for AI creativity. Turns out, I was asking the wrong question. I tried temperature tweaks, fresh phrasings, clever prompts. Nothing. So I asked a different question. Most teams ask a model for one answer. That is where the creativity goes. A new Stanford-led paper offers a simple fix (link in comments): ask for five responses with their probabilities. They call it 𝗩𝗲𝗿𝗯𝗮𝗹𝗶𝘇𝗲𝗱 𝗦𝗮𝗺𝗽𝗹𝗶𝗻𝗴. It sounds almost silly. I tried it and laughed, because it worked. The reason is human, not mystical. When we train models to match our preferences, raters tend to pick familiar answers. Over time, systems learn to be safe and the same. So change the ask: “Generate 5 jokes about coffee 𝘄𝗶𝘁𝗵 𝘁𝗵𝗲𝗶𝗿 𝗽𝗿𝗼𝗯𝗮𝗯𝗶𝗹𝗶𝘁𝗶𝗲𝘀.” That’s it. No retraining. No API changes. Just a better way to ask. But here’s the part that decides accuracy in the real world: Sampling gives you options. Real-time data chooses the right one. Ask for five, then check what is true right now: the page, the price, the inventory, the log. This is the quiet loop we run at Nimble with our Web Search Agents. Try it on something you’re shipping this week: get five options with probabilities, then verify against current data. That is the gap between interesting and correct. If you’re building a loop like this and want to see how we run it against live web data, you can share your use case and request a walkthrough here 👉 https://t2m.io/49gsXwy

The real race starts when the page changes. In enterprise AI, reliability comes from the runtime around the model: live session and governed context. You feel the difference in small moments. A rule changes mid-flow and the run continues. A layout shifts and drift is still caught. A login path moves and the result still lands where it should. That is the gap between a clever demo and a dependable system. Nimble's Web Search Agents make the live web a first-class data source for enterprise AI, not a best-effort scrape. If you want to see how this could work in your environment, share your use case and request a live demo here 👉 https://t2m.io/49gsXwy

If Finance and Operations are always debating the data, what would it take to get them to see eye to eye? For Uber Eats, the answer was Nimble. Ops was working off promo data from a few stores. Finance had completely different pricing benchmarks. Strategy meetings dragged and were less about action, more about reconciliation. Then came one schema. One cadence. Powered by Nimble's Web Search Agents. Twelve months later, they had collected over 200 million data points across five markets. Finally, clean alignment between every team. Monday mornings started with ZIP-level moves lists covering pricing, promos, availability, delivery fees, and ratings. Midweek updates included source links for every shift. By Friday, teams shared a simple "changes plus shipped" brief. At NRF, Jaime Grau described the impact: "It gave us a clear view of our competitive position and what to push next." We handled the data layer. Uber Eats turned it into competitive advantage. Curious about the specific schema and weekly cadence that made this work? Drop us a note in our Slack: https://lnkd.in/eJTcD7r8