Nimble

A 2021 Tesla Model S listed for $9,900 on one platform and $29,000 on another for the same year and model. That spread only becomes visible when used EV listings from eBay Motors and Carvana are aggregated, normalized, and scored against the same criteria. Using Nimble, we pulled 818 used EV listings from both marketplaces and scored 807 of them across five factors: price, mileage, year, condition, and title status. The result was a dataset that did not exist before the app ran. Cross-source comparisons that are invisible when you browse either platform manually. Getting there meant combining two live marketplaces with different site structures, rendering requirements, and extraction logic. Nimble’s Web Search Agent platform handled the data layer: a pre-built eBay agent, Batch Extract API for parallel page fetching, and a custom Carvana agent built with Agent Builder. The same architecture powers pricing intelligence, competitive monitoring, and market research pipelines without rebuilding the data layer for each use case. Full walkthrough here 👉 https://lnkd.in/gAPrzemx #AIAgents #AISearch #DevTools

Production-grade AI runs on fresh, structured, reliable data. As AI systems move into real workflows, reliability depends on whether the data layer can keep pace with the live web. Nimble is turning live web data into production infrastructure. Nimble’s Web Search Agent platform browses the live web, extracts the relevant data, and returns structured outputs production AI systems can use directly. The team building this infrastructure is growing across engineering, product, solutions, customer success, and go-to-market. Explore open roles to be part of it 👉https://lnkd.in/g-JfCf-M #Hiring #AI #AIInfrastructure #EnterpriseAI

The other night at J Bespoke was exactly the room we hoped to bring together! Monte Carlo, CoPlane, and Nimble hosted the AI Agents Happy Hour during AI Agent Conference Week in NYC. Observe & Orchestrate was the right frame for a room full of people paying close attention to where agents are headed and building the infrastructure that makes them useful beyond the prototype. Thank you to everyone who joined us! #AIAgents #AIAgentConference

Dashboards can show the price change. But they rarely close the gap between signal and response. Retail pricing has always been competitive. What changed is the clock speed. Prices, promotions, and availability now change by the hour, while automated pricing systems do not pause for a weekly review. By the time many teams verify the signal and bring it into a decision meeting, the opportunity may already be gone. The operating model has to change with the market. Pricing has become an execution sport. That means SKU-level signals and store-aware context that reach the workflow early enough to guide the next pricing move. Julie Averill, former CIO at Lululemon and REI, breaks down why retail pricing is moving from dashboard review cycles to real-time response loops, and what brands need to keep up. The data behind those loops has to arrive at the speed of the shelf, not the speed of the vendor’s refresh cycle. Read the full piece here: https://lnkd.in/gy8Ki8Td #RetailAI #DigitalShelf #RetailPricing

In this example, one prompt returns used EV results as both a clean table for humans and structured JSON for the agent. Model, price, range, battery size, tax credit eligibility. Fields the workflow can validate, compare, filter, and pass downstream. For agents, the useful web starts where raw pages become structured fields. Nimble Skills let agents search public web sources and return structured outputs instead of raw pages or text snippets. Install Nimble Skills and start building with live web data → https://lnkd.in/gkaFtnXZ #AIAgents #AIInfrastructure #DevTools

Next week, during AI Agent Conference in NYC, Monte Carlo, CoPlane, and Nimble are bringing together founders, operators, data leaders, and engineers turning agent ideas into working systems. The focus: live context, reliable data sources, structured outputs, workflow ownership, and trust when decisions move beyond the prototype. If you are building agents, evaluating the stack around them, or trying to understand what separates a cool workflow from a useful one, join us in NYC. 📍 5:30 PM | J Bespoke | NYC 📌 Spots are limited. 👉 Request to join: https://lnkd.in/g6_SCGnd #AIAgents #AIAgentConference

Enterprises no longer need to be convinced that agents can do useful work. The harder part now is earning trust in production. I liked the thought experiment in Cisco’s recent agent trust piece: Imagine waking up tomorrow with 1,000 new expert teammates. Engineers, analysts, operators, and strategists. Available around the clock. Fast, capable, and ready to work. At first, that sounds like leverage. Then you start thinking about risk: What would you actually trust them to do? That trust gap is already visible in the numbers: 85% of enterprises are already piloting AI agents. Only 5% have trusted them enough for production. A prototype proves the agent can complete the task. Production asks whether the system can be trusted when the environment changes. This is when agents stop being a model story and become an infrastructure story. A smart agent with weak inputs is still a production risk. Trusted delegation starts with trusted inputs. The next production bottleneck may be less about model capability alone and more about governance, trusted inputs, and the data layer underneath the agent. What is your take? #AIInfrastructure #AIAgents

Next week, during AI Agent Conference in NYC, Monte Carlo, CoPlane, and Nimble are bringing together founders, operators, data leaders, and engineers turning agent ideas into working systems. The focus: live context, reliable data sources, structured outputs, workflow ownership, and trust when decisions move beyond the prototype. If you are building agents, evaluating the stack around them, or trying to understand what separates a cool workflow from a useful one, join us in NYC. 📍 5:30 PM | J Bespoke | NYC 📌 Spots are limited. 👉 Request to join: https://lnkd.in/g6_SCGnd #AIAgents #AIAgentConference

There are three structural reasons why AI search tools fail enterprise agent workflows. None of them are model-related. Enterprise AI is already in production, but reliability still breaks when agents are used to support real decisions. Take procurement. An agent may have access to contract terms, delivery history, and internal vendor records. Useful on paper. But supplier evaluation also depends on outside signals: recent disputes, reputation changes, market movement, regulatory updates. Whether the agent can use those signals depends on how the search layer was built. Cached data creates freshness gaps. Long blocks of text create discoverability problems because the agent still has to extract, structure, and pass the right answer into the next step. Opaque retrieval paths create trust gaps because teams cannot verify where the answer came from or how current it was. In each case, the model can still sound confident. The workflow just becomes harder to trust. Current, structured, and traceable context is what separates a confident answer from a reliable one. Uriel Knorovich goes deeper on this in his latest RTInsights piece: 👉 https://lnkd.in/gvPtpDxA #AISearch #EnterpriseAI #AIAgents #AIGovernance

What if TikTok became a live dataset? We gave #Claude Code a simple task: “Show me the trending TikTok posts on shoes posted yesterday. Include the engagement for each post.” Nimble Skills turned that request into live, structured output. Social platforms are full of useful data. But when that data lives inside dynamic pages, agents still can’t use it directly. That’s the difference between viewing social content and making it usable for agents. With Nimble’s Web Search Agents, pages on TikTok, Instagram, YouTube, and other social platforms can become structured outputs your agents and workflows can actually use. Try Nimble Skills here → https://lnkd.in/gYWp2cdj #AISearch #AIAgents #DevTools