First Rays Venture Partners

𝗧𝗵𝗲 𝗡𝗲𝘄 𝗣𝗿𝗼𝘁𝗼𝗰𝗼𝗹 𝗪𝗮𝗿𝘀: 𝗪𝗵𝘆 𝗠𝗖𝗣 𝗮𝗻𝗱 𝗔𝟮𝗔 𝗠𝗮𝘁𝘁𝗲𝗿 𝗡𝗼𝘄 We're at a strategic inflection point in AI infrastructure. Two protocols—𝗠𝗖𝗣 and 𝗔𝟮𝗔—are quietly redrawing the map of how tools, data, and agents interact. If you’re building, investing, or betting on where the AI stack goes next, pay attention. Why? Because just like the protocol wars of the late '90s, what looks like low-level plumbing today will shape who owns the user, the interface, and the economic power tomorrow. 𝗣𝗿𝗶𝗺𝗲𝗿: 𝗠𝗖𝗣 𝘃𝘀 𝗔𝟮𝗔 (📌 𝗦𝗮𝘃𝗲 𝘁𝗵𝗶𝘀) 𝗠𝗖𝗣 (𝗠𝗼𝗱𝗲𝗹 𝗖𝗼𝗻𝘁𝗲𝘅𝘁 𝗣𝗿𝗼𝘁𝗼𝗰𝗼𝗹) Think: “HTTP for AI agents” Designed for: Agent ↔️ Tool interaction Format: Structured, stateful function calls Use case: Plugging AI into existing apps, data, APIs Analogy: Like Zapier for AI, but smarter and runtime-aware 𝗔𝟮𝗔 (𝗔𝗴𝗲𝗻𝘁-𝘁𝗼-𝗔𝗴𝗲𝗻𝘁) Think: “TCP/IP for multi-agent systems” Designed for: Agent ↔️ Agent communication Format: Natural language prompts + shared context Use case: Choreographing workflows across domains Analogy: Microservices, but everyone speaks English They’re not rivals. They’re layered. And together, they’re creating a stack where agents can call tools (MCP) or collaborate with other agents (A2A)—a foundational shift in how work gets done. Google just joined the A2A bandwagon. OpenAI, Anthropic, and others have embraced MCP. We're watching consolidation around de facto standards. That alone is enough to raise eyebrows. 𝗙𝗶𝗻𝗮𝗹 𝗧𝗵𝗼𝘂𝗴𝗵𝘁: 𝗔𝗣𝗜𝘀 𝗔𝗹𝗹 𝘁𝗵𝗲 𝗪𝗮𝘆 𝗗𝗼𝘄𝗻 “𝘉𝘦 𝘤𝘢𝘳𝘦𝘧𝘶𝘭 𝘸𝘩𝘢𝘵 𝘺𝘰𝘶 𝘦𝘹𝘱𝘰𝘴𝘦—𝘢𝘯𝘥 𝘸𝘩𝘢𝘵 𝘺𝘰𝘶 𝘥𝘰𝘯’𝘵.” — Uri Maoz, WireMock If your app has an API, it’s already on the menu for agents. The decision is whether you serve it up intentionally—or let someone else mediate that access and own the experience. This is bigger than features. This is about control.

𝗧𝗵𝗲 𝗡𝗲𝘄 𝗣𝗿𝗼𝘁𝗼𝗰𝗼𝗹 𝗪𝗮𝗿𝘀: 𝗪𝗵𝘆 𝗠𝗖𝗣 𝗮𝗻𝗱 𝗔𝟮𝗔 𝗠𝗮𝘁𝘁𝗲𝗿 𝗡𝗼𝘄 We're at a strategic inflection point in AI infrastructure. Two protocols—𝗠𝗖𝗣 and 𝗔𝟮𝗔—are quietly redrawing the map of how tools, data, and agents interact. If you’re building, investing, or betting on where the AI stack goes next, pay attention. Why? Because just like the protocol wars of the late '90s, what looks like low-level plumbing today will shape who owns the user, the interface, and the economic power tomorrow. 𝗣𝗿𝗶𝗺𝗲𝗿: 𝗠𝗖𝗣 𝘃𝘀 𝗔𝟮𝗔 (📌 𝗦𝗮𝘃𝗲 𝘁𝗵𝗶𝘀) 𝗠𝗖𝗣 (𝗠𝗼𝗱𝗲𝗹 𝗖𝗼𝗻𝘁𝗲𝘅𝘁 𝗣𝗿𝗼𝘁𝗼𝗰𝗼𝗹) Think: “HTTP for AI agents” Designed for: Agent ↔️ Tool interaction Format: Structured, stateful function calls Use case: Plugging AI into existing apps, data, APIs Analogy: Like Zapier for AI, but smarter and runtime-aware 𝗔𝟮𝗔 (𝗔𝗴𝗲𝗻𝘁-𝘁𝗼-𝗔𝗴𝗲𝗻𝘁) Think: “TCP/IP for multi-agent systems” Designed for: Agent ↔️ Agent communication Format: Natural language prompts + shared context Use case: Choreographing workflows across domains Analogy: Microservices, but everyone speaks English They’re not rivals. They’re layered. And together, they’re creating a stack where agents can call tools (MCP) or collaborate with other agents (A2A)—a foundational shift in how work gets done. Google just joined the A2A bandwagon. OpenAI, Anthropic, and others have embraced MCP. We're watching consolidation around de facto standards. That alone is enough to raise eyebrows. 𝗙𝗶𝗻𝗮𝗹 𝗧𝗵𝗼𝘂𝗴𝗵𝘁: 𝗔𝗣𝗜𝘀 𝗔𝗹𝗹 𝘁𝗵𝗲 𝗪𝗮𝘆 𝗗𝗼𝘄𝗻 “𝘉𝘦 𝘤𝘢𝘳𝘦𝘧𝘶𝘭 𝘸𝘩𝘢𝘵 𝘺𝘰𝘶 𝘦𝘹𝘱𝘰𝘴𝘦—𝘢𝘯𝘥 𝘸𝘩𝘢𝘵 𝘺𝘰𝘶 𝘥𝘰𝘯’𝘵.” — Uri Maoz, WireMock If your app has an API, it’s already on the menu for agents. The decision is whether you serve it up intentionally—or let someone else mediate that access and own the experience. This is bigger than features. This is about control.

⚡️𝗗𝗶𝗰𝗵𝗼𝘁𝗼𝗺𝘆 𝗼𝗳 #𝗲𝗰𝗼𝗺𝗺𝗲𝗿𝗰𝗲 𝘃𝘀 𝗽𝗵𝘆𝘀𝗶𝗰𝗮𝗹 𝗿𝗲𝘁𝗮𝗶𝗹  7 out of 10 customer visits convert in physical retail 7 out of 10 online visits don’t convert online. 💡𝘚𝘰𝘭𝘷𝘪𝘯𝘨 𝘵𝘩𝘦 𝘢𝘵𝘵𝘳𝘪𝘵𝘪𝘰𝘯 𝘱𝘳𝘰𝘣𝘭𝘦𝘮 —> What does physical retail do better that #ecommerce can emulate? 𝗔𝘀𝗸 𝘀𝗺𝗮𝗿𝘁 𝗾𝘂𝗲𝘀𝘁𝗶𝗼𝗻𝘀- what kind of photographer are you? Not just what type of camera do you want? 𝗛𝗮𝘃𝗲 𝗘𝗺𝗽𝗮𝘁𝗵𝘆 - buy a disposable battery for your camera vs rechargeable battery (depending on the situation). Eg. 20% of retail customers have a tech and a language literacy issue. They cannot express themselves succinctly in a few keywords in English in a search box. ** 𝗧𝗼𝗱𝗮𝘆, Blotout 𝗶𝘀 𝗮𝗻𝗻𝗼𝘂𝗻𝗰𝗶𝗻𝗴 𝘁𝗵𝗲𝗶𝗿 𝗻𝗲𝘅𝘁 𝗲𝘃𝗼𝗹𝘂𝘁𝗶𝗼𝗻 𝗳𝗿𝗼𝗺 𝗮 𝗽𝗿𝗼𝗱𝘂𝗰𝘁 𝘁𝗼 𝗮 𝗽𝗹𝗮𝘁𝗳𝗼𝗿𝗺 Brands/merchants face many challenges that prevent them from maturing their AI/ML stack and deploying models into production.  With Blotout ShopBOT, they can quickly deploy solutions that move the needle in terms of conversion. For brands/merchants, AI can supercharge personalization. AI creates tailored shopping journeys that remember past purchases, browsing habits, and even predict future wants and needs. This personalized approach enhances the overall shopping experience, making it more engaging and enjoyable. AI algorithms analyze 1st party customer data (powered by Blotout 𝗘𝗱𝗴𝗲𝗧𝗮𝗴 𝗶𝗻𝗳𝗿𝗮𝘀𝘁𝗿𝘂𝗰𝘁𝘂𝗿𝗲) to provide recommendations that precisely align with individual preferences and purchase history, increasing the likelihood of discovering products they actually want. ShoptBOT enables better understanding of customer queries, improving search experiences even with conversational or voice-based input (this breaks through the language/tech literacy barriers). AI tools can process images for visual search, allowing customers to find similar products based on photos rather than text descriptions alone. The combination of keyword and vector search technologies dramatically improves accuracy in matching customer intent with available products. These AI-powered search and navigation capabilities make it substantially easier for shoppers to find exactly what they want quickly and efficiently, reducing frustration and creating more satisfying shopping journeys. 🔗 𝗟𝗮𝘂𝗻𝗰𝗵 𝗣𝗼𝘀𝘁 -> https://lnkd.in/gtbgyEEU 🔗 𝗦𝗶𝗴𝗻 𝘂𝗽 𝗻𝗼𝘄 -> https://lnkd.in/gwW8QaxB 🤝 𝗠𝗲𝗲𝘁 𝗮𝘁 @𝗦𝗵𝗼𝗽𝗧𝗮𝗹𝗸 -> https://lnkd.in/gBYU9uCw ** Source @Omar Hijazi, Americas Industry Leader in Retail, Microsoft 

In a world where big vendors often want to carve markets up into walled gardens, it's great to see a useful standard being universally adopted. MCP arrived at just the right moment for WireMock. We were considering how to integrate AI-driven mocking features into the product and had originally planned to embed these directly in our app. But after some quick spiking with MCP it was clearly the better way, both for us during product development and for our end users. In no particular order: - MCP enables users to attach your product to *their* tools, meaning their choice of tool, model, IDE, prompting rules etc. are all maintained. - In enterprise environments any security/compliance tooling in place such as LLM proxies also work as-is. This means we won't need to go through an additional security/compliance journey to get customer traction, or build a big hairy “turn off AI” switch for AI-hesitant customers. - Being integrated into a user’s existing coding tool enables composite tasks to be executed (vs. prompts only being able to operate on our product’s resources). - From a product development perspective, it turns out that prototyping workflows using other products’ agents is far more productive than attempting to build your own from scratch. Connecting WireMock’s MCP prototype to Cursor allowed me to very quickly explore what could be possible without building any agent logic of my own. - And finally, when it’s the customer’s tools that are talking to the LLM instead of your product, the AI bills are not yours to worry about 🙂 Our marketing department hasn't allowed me to post all the details just yet, but we'll be doing a big reveal this Wednesday so stay tuned...

WireMock now has an MCP server, available via the new CLI. This lets you generate and enhance mock APIs directly in WireMock Cloud via your AI coding tool, either using existing sources of information like code or documentation or prompting to create new API features that don't exist yet. 👇 Used here with Cursor to generate a stateful mock API for a to-do list app feature while also generating the feature's UI and integration code. #mcp #ai

We're #hiring a new Human Resources Manager in Hyderabad, Telangana. Apply today or share this post with your network.

Amazing showcase of Fused platform’s power to fuse sensor data across different domains and help in urban planning. Maxar Technologies’s imagery is at its best when combined with partners providing third party imagery and open source datasets like Vancouver’s Open Data Portal. Analytics today requires bringing all these different sources of data together: from raw pixels to contextual intelligence — in real-time, seamlessly, and at scale. This is an iterative process, each dataset being in its own format & hosted on different cloud providers, updated independently and on different schedules. Fused is built by data scientists, engineers & open source contributors who know the reality of building pipelines with breaking changes in libraries, new file formats that require rethinking how data is ingested and business requirements that constantly change. We firmly believe we need tools that help us bring as many datasets together as possible, especially in a world where LLMs and AI can increasingly talk directly to data. Learn more at https://lnkd.in/gHmQkfiP

My favorite part of working with well built products is you don't actually do work — you're given the opportunity to play. For the past couple weeks, I've had the honor of embracing creativity and curiosity in preparation for Peter Wilczynski's #TED2025 talk today advocating for a new era of modernized mapmaking. We played, asked questions, had new ideas, and kept rapidly iterating with each new seemingly disparate data set from Vancouver's open data program, our own imagery from Maxar Technologies, Sentinel, and more into my Fused workbench. As I quickly discovered relationships between tree cover, recent heat waves, and subsidized housing, Fused melted into the background ... as some of the best apps do when you're in the zone. Beautifully, it became effortless to ask the next question, build/tune a user defined function (UDF), find patterns, add in more data sets, and keep the cycle going. Huge props to Sina Kashuk and his whole team who collaborated alongside me and Joe Svejkosky for bringing some of my crazy ideas to life. Maxar's imagery APIs fluidly fed 15cm HD false color IR to Fused, which aligned perfectly to the 181,000+ annotated trees from Vancouver to show that the city is becoming greener and while there's still work to do, the neighborhoods who need the cover the most are beginning to see the benefit of a an audacious vision of creating a dense urban canopy. (Good work Vancouver!) What I love more than all of these discoveries is the approachability of being able to ask these questions of data I didn't even know existed. I'm not a data scientist, I've got a studio art degree and am just now learning Python after mostly focusing on front end languages. So my question to you is ... what Living Globe will you discover? How will you answer today's mystery to leave the world better tomorrow? How can we support you on that journey to make Peter's TED Conferences talk a reality for the thing you care about? Can't wait to see what we build together as a community!

WireMock now has an MCP server, available via the new CLI. This lets you generate and enhance mock APIs directly in WireMock Cloud via your AI coding tool, either using existing sources of information like code or documentation or prompting to create new API features that don't exist yet. 👇 Used here with Cursor to generate a stateful mock API for a to-do list app feature while also generating the feature's UI and integration code. #mcp #ai

In a world where big vendors often want to carve markets up into walled gardens, it's great to see a useful standard being universally adopted. MCP arrived at just the right moment for WireMock. We were considering how to integrate AI-driven mocking features into the product and had originally planned to embed these directly in our app. But after some quick spiking with MCP it was clearly the better way, both for us during product development and for our end users. In no particular order: - MCP enables users to attach your product to *their* tools, meaning their choice of tool, model, IDE, prompting rules etc. are all maintained. - In enterprise environments any security/compliance tooling in place such as LLM proxies also work as-is. This means we won't need to go through an additional security/compliance journey to get customer traction, or build a big hairy “turn off AI” switch for AI-hesitant customers. - Being integrated into a user’s existing coding tool enables composite tasks to be executed (vs. prompts only being able to operate on our product’s resources). - From a product development perspective, it turns out that prototyping workflows using other products’ agents is far more productive than attempting to build your own from scratch. Connecting WireMock’s MCP prototype to Cursor allowed me to very quickly explore what could be possible without building any agent logic of my own. - And finally, when it’s the customer’s tools that are talking to the LLM instead of your product, the AI bills are not yours to worry about 🙂 Our marketing department hasn't allowed me to post all the details just yet, but we'll be doing a big reveal this Wednesday so stay tuned...