<?xml version="1.0" encoding="UTF-8"?><rss xmlns:dc="http://purl.org/dc/elements/1.1/" xmlns:content="http://purl.org/rss/1.0/modules/content/" xmlns:atom="http://www.w3.org/2005/Atom" version="2.0" xmlns:cc="http://cyber.law.harvard.edu/rss/creativeCommonsRssModule.html">
    <channel>
        <title><![CDATA[Stories by Shashwat Tiwari on Medium]]></title>
        <description><![CDATA[Stories by Shashwat Tiwari on Medium]]></description>
        <link>https://medium.com/@shashverse?source=rss-9a5497c44b4e------2</link>
        <image>
            <url>https://cdn-images-1.medium.com/fit/c/150/150/1*tCwXclEHvrH5QaBU-xJDGw.png</url>
            <title>Stories by Shashwat Tiwari on Medium</title>
            <link>https://medium.com/@shashverse?source=rss-9a5497c44b4e------2</link>
        </image>
        <generator>Medium</generator>
        <lastBuildDate>Fri, 20 Nov 2026 09:47:07 GMT</lastBuildDate>
        <atom:link href="https://medium.com/@shashverse/feed" rel="self" type="application/rss+xml"/>
        <webMaster><![CDATA[yourfriends@medium.com]]></webMaster>
        <atom:link href="http://medium.superfeedr.com/" rel="hub"/>
        <item>
            <title><![CDATA[The Emerging Internet of AI Agents (MCP vs A2A vs Nanda vs Agntcy)]]></title>
            <link>https://medium.com/design-bootcamp/the-emerging-internet-of-ai-agents-mcp-vs-a2a-vs-nanda-vs-agntcy-60f7f9963509?source=rss-9a5497c44b4e------2</link>
            <guid isPermaLink="false">https://medium.com/p/60f7f9963509</guid>
            <category><![CDATA[ai-agent]]></category>
            <category><![CDATA[claude]]></category>
            <category><![CDATA[mcp-server]]></category>
            <category><![CDATA[ai]]></category>
            <category><![CDATA[agentic-ai]]></category>
            <dc:creator><![CDATA[Shashwat Tiwari]]></dc:creator>
            <pubDate>Sun, 20 Apr 2025 06:25:32 GMT</pubDate>
            <atom:updated>2025-04-20T09:58:57.395Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*01L07rn51iQU9VQCmWSrAw.png" /><figcaption>MCP vs A2A vs NANDA vs AGNTCY</figcaption></figure><p>All the buzz around AI often centers on the capabilities of individual models, the next GPT-X or Claude-Y iteration. While these advancements are undoubtedly impressive, a tectonic shift is quietly happening with the emergence of the <strong>Internet of AI Agents (IoA).</strong></p><blockquote>The future isn’t just about bigger, more powerful standalone AI, it’s about a network of interconnected intelligences where the real leverage comes from their ability to collaborate.</blockquote><figure><img alt="" src="https://cdn-images-1.medium.com/max/524/1*9hzlH9WjzredbGjcHoKnZA.png" /></figure><p>We’re transitioning from a world of <em>isolated, task-specific AI Agents</em> towards a future where they can seamlessly connect, collaborate, and collectively tackle intricate challenges.</p><p>Imagine a digital ecosystem where diverse software components, autonomous systems, and specialized AI tools operate as a highly coordinated team, each contributing its unique skills.</p><p>This move from <em>siloed functionalities</em> to <em>interconnected systems</em> promises to unlock unprecedented levels of intelligence and capability, marking a shift where the <strong>exponential AI advancements are driven not just by the scaling of individual models and more by the powerful network effects of collaborative intelligence.</strong></p><p>Think back to the early days of the internet, before the development of the TCP/IP. Various networks existed, but their ability to communicate was very very limited. <strong>It was the standardization of these protocols that paved the way for the globally interconnected internet we rely on today.</strong></p><blockquote>AI Agents today face a similar challenge of fragmentation.</blockquote><p><strong>Brilliant agentic systems have been developed, but they often function like specialists in separate rooms</strong>, unable to easily communicate with each other. They mostly operate in silos, requiring custom integrations for any form of interaction. This creates significant friction and limits their potential.</p><p><strong>Interoperability becomes a major stumbling block when agents operate with different communication rules and data formats.</strong></p><p>Just as language barriers hinder human collaboration, or incompatible APIs prevent software integration, a lack of standardized agent protocols creates friction and slows down progress. Establishing common ground through standardized protocols is essential, making collaboration between AI agents simpler and more efficient.</p><p>Several new initiatives are emerging to build this crucial infrastructure for networked intelligence. These are the ones to watch closely</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*g_qWWPbgh6g4-QgJ3YRjig.png" /><figcaption>Internet of AI Agents</figcaption></figure><h3>Emerging Players:</h3><h4><strong>Anthropic’s Model Context Protocol (MCP)</strong></h4><ul><li>MCP acts like a universal plug, allowing any AI assistant to connect to any data source or service without the need for custom wiring for each integration.</li><li>It’s designed as an open standard that securely connects AI assistants (hosts) to data repositories and tools (servers) through intermediary clients. This enables secure, two-way communication for both retrieving information and performing actions.</li><li>The core focus is on equipping AI models with the necessary context and tools to function effectively. Think of it as the universal adapter for AI.</li></ul><h4><strong>Google’s Agent2Agent (A2A) Protocol</strong></h4><ul><li>A2A focuses on how agents can discover each other, negotiate interactions, exchange information, and coordinate to complete tasks together, regardless of the vendor or underlying framework.</li><li>The aim is to create a common language for agentic AI. This open protocol employs a client-remote agent model. Key features include “Agent Cards” (which advertises an agent’s capabilities), support for long-running tasks, and the ability to handle various data types (text, audio, video).</li></ul><h4><strong>MIT’s NANDA (Networked Agents And Decentralized AI)</strong></h4><ul><li>NANDA’s vision is to build a decentralized “Internet of AI Agents,” leveraging and expanding upon Claude’s MCP.</li><li>Their goal is to establish the infrastructure for distributed agent intelligence at scale, with a strong emphasis on discovery, distributed knowledge search, secure authentication, and verifiable accountability for agent-to-agent exchanges. It features a layered architecture encompassing the protocol itself, developer tools, and an infrastructure registry.</li><li>Ultimately, NANDA aims to create an inclusive, responsible, and open landscape where agents can collaborate trustlessly across organizational boundaries.</li></ul><h4><strong>Cisco Outshift’s AGNTCY</strong></h4><ul><li>AGNTCY is focused on building the foundational, open-source infrastructure for the entire IoA lifecycle. This aims to enable seamless, end-to-end agent collaboration, allowing developers to easily find, connect, run, and manage any agent, irrespective of its underlying framework.</li><li>Through the AGNTCY collective, they are concentrating on four key capabilities: <br>- <strong>Discover</strong> (finding agents via registries and the Open Agent Schema Framework OASF)<br>- <strong>Compose</strong> (connecting agents into workflows)<br>- <strong>Deploy</strong> (running agents securely at scale), and <br>- <strong>Evaluate</strong> (monitoring agent performance). The Agent Connect Protocol (ACP) plays a crucial role in standardizing collaboration.</li></ul><p>These initiatives are tackling the fundamental challenges in building internet of agents in their own way:</p><h4><strong>Agent Discovery:</strong> How do agents find others with the capabilities they need?</h4><ul><li><strong>A2A:</strong> Utilizes Agent Cards to advertise skills and connection details.</li><li><strong>NANDA:</strong> Focuses on building discovery mechanisms within its decentralized network.</li><li><strong>AGNTCY:</strong> Employs the Open Agent Schema Framework (OASF) to standardize capability descriptions, with the Agent Directory acting as a central registry.</li></ul><h4><strong>Secure Communication:</strong> How can agents interact with each other in a trustworthy manner?</h4><ul><li><strong>MCP:</strong> Provides secure connections between agents and their associated tools and data.</li><li><strong>A2A:</strong> Architecture facilitates secure information exchange between agents.</li><li><strong>NANDA:</strong> Emphasizes the implementation of secure verification protocols.</li><li><strong>AGNTCY:</strong> The Agent Connect Protocol (ACP) standardizes message passing, state management, and context sharing to ensure secure interactions.</li></ul><h4><strong>Task Coordination:</strong> How can multiple agents work together effectively to achieve complex goals?</h4><ul><li><strong>A2A:</strong> Facilitates the coordination of actions across various agents, tools, and services.</li><li><strong>AGNTCY:</strong> Provides tools and protocols, such as ACP, to compose agents into efficient workflows.</li><li><strong>NANDA:</strong> Introduces mechanisms for dynamic teaming and the management of conversation flow between agents.</li></ul><h4><strong>Integration:</strong> How do agents gain access to the necessary data and tools to perform their functions?</h4><ul><li><strong>MCP:</strong> Provides the standardized means for AI systems to connect to external data sources and tools — the “context” they need.</li><li><strong>AGNTCY:</strong> Offers components like the API Bridge Agent to enable agents to connect to virtually any API.</li></ul><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*01L07rn51iQU9VQCmWSrAw.png" /></figure><h3>What Does It Mean for Businesses?</h3><p>This networked intelligence opens the door to entirely new and innovative service offerings and business models that were previously unattainable with isolated AI systems. Imagine the possibilities of automating complex workflows that currently require significant human intervention and are impossible to solve today.</p><p>The coordinated efforts of agents can deliver hyper-personalized and proactive customer experiences, while the collective intelligence derived from analyzing vast amounts of information allows for sharper, more data-driven decision-making. Furthermore, initiatives like Cisco’s Agntcy aim to break down vendor silos, fostering greater cross-collaboration and offering customers more diverse and integrated solutions.</p><p><strong>For builders and product managers, this represents a fundamental shift in approach.</strong></p><ul><li>The focus might move from designing standalone features to architecting products as interactive nodes within a larger agent ecosystem.</li><li>Increasingly, success will depend on seamlessly leveraging capabilities from external agents and carefully considering the <strong>“Agent Experience” (AX)</strong>: how easily can other agents discover, understand, and utilize your product or service?</li></ul><p>Final Thoughts:</p><p><strong>Standardized protocols like MCP and A2A</strong>, alongside <strong>comprehensive infrastructure initiatives like NANDA and AGNTCY</strong>, are the crucial enablers laying the groundwork for this collaborative future. By addressing the fundamental challenges of discovery, communication, coordination, and integration, they unlock the potential for AI agents to work together seamlessly. This promises to transform industries, drive innovation, and automate complexity on an unprecedented scale.</p><p>I write about Enterprise Agentic AI, More at: <a href="http://www.shashverse.com/">www.shashverse.com</a> <br>Written in collaboration with <a href="https://medium.com/u/82d3834a9510">Rohan Balkondekar</a>, <a href="https://medium.com/u/58234f6275d3">Divyansh Dwivedi</a>.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=60f7f9963509" width="1" height="1" alt=""><hr><p><a href="https://medium.com/design-bootcamp/the-emerging-internet-of-ai-agents-mcp-vs-a2a-vs-nanda-vs-agntcy-60f7f9963509">The Emerging Internet of AI Agents (MCP vs A2A vs Nanda vs Agntcy)</a> was originally published in <a href="https://medium.com/design-bootcamp">Bootcamp</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[How Apple Brainstorms AI Use Cases]]></title>
            <link>https://medium.com/design-bootcamp/how-apple-brainstorms-ai-use-cases-dc7a80bb1551?source=rss-9a5497c44b4e------2</link>
            <guid isPermaLink="false">https://medium.com/p/dc7a80bb1551</guid>
            <category><![CDATA[ai]]></category>
            <category><![CDATA[product-management]]></category>
            <category><![CDATA[technology]]></category>
            <category><![CDATA[apple]]></category>
            <category><![CDATA[generative-ai-use-cases]]></category>
            <dc:creator><![CDATA[Shashwat Tiwari]]></dc:creator>
            <pubDate>Sat, 14 Dec 2024 18:31:56 GMT</pubDate>
            <atom:updated>2024-12-17T18:04:48.123Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*7fn17eyc8x35pmhCGpGKIQ.png" /></figure><p>While tech giants raced to showcase their generative AI capabilities with flashy demos &amp; rapid releases, Apple took a characteristically different path.</p><p>Their measured approach to AI might seem conservative. But looking deeper, it reveals a sophisticated framework for evaluating AI initiatives that any product leader can learn from.</p><p><strong>Instead of being first, Apple aims to be the best.</strong> This deliberate approach reflects its core principles:</p><ol><li>Always prioritize consumer impact</li><li>Respect data privacy</li><li>Play the long-term game.</li></ol><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*5uge1Cg-wVpkmuIpVnlc9A.png" /></figure><p><strong>Apple views AI as a multi-decade game, It’s about delivering meaningful, user-centric products</strong> that align with Apple’s legacy of excellence.</p><p>As a Product Manager, I was curious to dig deeper into Apple PMs’ approach to deciding which AI ideas to work on. <strong><br>How do they decide what’s worth pursuing? <br>What process helps them focus on ideas that make sense and avoid wasting resources on AI without a clear purpose?</strong></p><p>Here is what I found:</p><p>Apple’s product teams begin by examining the fundamental driver behind any AI initiative</p><p>Before writing a single line of code, they ask themselves a brutally honest question: <strong>“Why are we really doing this?” </strong><em>If you can’t answer that clearly, you’re already lost.</em></p><p><strong>Why are we considering this initiative?</strong></p><ul><li><em>Is it a matter of business continuity, where not adopting AI would make us obsolete as competitors set the new standard?</em></li><li><em>Is it becoming table stakes for staying relevant in the industry?</em></li><li><em>Or does it represent a clear opportunity to boost profitability, improve productivity, or unlock entirely new user value that wouldn’t be possible without AI?</em></li></ul><p><strong>This clarity distinguishes strategic priorities from FOMO-driven pursuits.</strong></p><p>Once an AI initiative passes the “Why” test, <strong>Apple PMs evaluate it against four critical dimensions:</strong></p><ol><li><strong>Critical vs. Complementary</strong></li></ol><ul><li><em>Critical AI features are foundational. Without them, the product doesn’t work. Example: Face ID, where AI-driven facial recognition is essential.</em></li><li><em>Complementary AI features enhance the product but aren’t mandatory. Example: Gmail’s Smart Compose — helpful but not critical; Quick-type in Keyboard is complementary, not critical.</em></li><li><em>The more critical AI is, the higher the quality bar. Users tolerate fewer mistakes in core functionalities.</em></li></ul><p>2. <strong>Reactive vs. Proactive</strong></p><ul><li><em>Reactive AI features respond to user actions, like Siri answering a query.</em></li><li><em>Proactive AI features anticipate needs, like Maps suggesting alternate routes based on traffic or Siri suggesting a shortcut based on your routine.</em></li><li><em>Proactive features carry a higher risk of being perceived as intrusive. They must exceed user expectations to feel delightful rather than annoying.</em></li></ul><p>3. <strong>Dynamic vs. Static</strong></p><ul><li><em>Dynamic AI features adapt in real-time, personalizing based on user feedback (e.g., Face ID adjusting to physical changes).</em></li><li><em>Static AI features rely on occasional updates to shared models (e.g., object recognition in Photos).</em></li><li><em>Dynamic features require ongoing resource investment for personalization, while static features offer scalability.</em></li></ul><p>4.<strong> Human Involvement</strong></p><p>Role of humans and degree of involvement in AI applications:</p><ul><li><em>Supportive AI: Assists users without autonomy (e.g., predictive text)</em></li><li><em>Semi-Autonomous AI: Handles basic tasks but escalates complex issues (e.g., customer support routing)</em></li><li><em>Autonomous AI: Fully independent, with minimal human oversight (e.g., automated fraud detection)</em></li></ul><p>Even Microsoft suggests a similar <a href="https://www.microsoft.com/en-us/industry/blog/retail/2023/10/23/from-discussion-to-deployment-4-key-lessons-in-generative-ai/">“Crawl-Walk-Run” framework</a></p><p>Then Apple PMs ask themselves some questions about user impact and business value:</p><ul><li><em>How does this feature enhance lives? Does it solve a real problem or simply add complexity?</em></li><li><em>How much manual effort will this reduce?</em></li><li><em>Will this increase the speed of user actions or team processes?</em></li><li><em>Can this feature scale to millions of users seamlessly?</em></li></ul><p><strong>They decide minimum standards for release &amp; set measurable goals.</strong> <br><em>The difference between a fun demo and a polished product is miles apart.</em> <br>A quick proof of concept might showcase potential, but <strong>Apple doesn’t ship until it meets their exacting standards.</strong> They ask questions like</p><ul><li><em>How precise must the feature be?</em></li><li><em>How will you measure adoption and delight?</em></li><li><em>What ROI justifies the feature’s development?</em></li></ul><p><strong>Features aren’t just features — they’re commitments.</strong> They require ongoing care, PMs must account for long-term maintenance &amp; ask themselves</p><ul><li><em>How will this feature evolve as AI capabilities advance? The AI space moves at bullet-train speed.</em></li><li><em>Can this solution handle millions of users without degrading? At Apple’s scale, even minor inefficiencies get magnified into major costs.</em></li></ul><p>Recently, Tim Cook emphasized, “<a href="https://www.business-standard.com/technology/tech-news/we-re-not-late-to-the-game-tim-cook-defends-apple-s-ai-strategy-excerpts-124120500633_1.html">Apple refuses to accept a trade-off between great privacy and great intelligence.”</a> The underlying question is: <em>“How do we minimize trade-offs and deliver intelligence without compromising user trust?”</em> When you ask this, only then do you take steps to ensure it happens — like avoiding shortcuts that risk user trust and prioritizing on-device computations whenever possible.</p><blockquote><strong>These are just a few questions every innovator should ask themselves.</strong> Take the time to dive deep. Once you’ve spent time reflecting on whether the use case you’re pursuing is truly worth building, you’ll have clarity.</blockquote><p><strong>If it’s a “YES,” then it’s time to move quickly: Build fast, Ship fast, and go full-on </strong><a href="https://en.wikipedia.org/wiki/Effective_accelerationism"><strong>e/acc</strong></a></p><p>Some might dismiss this as <em>“unnecessary analysis paralysis”</em> or worry about being labeled a <em>“skeptical laggard.”</em> <strong>But slowing down to think doesn’t mean falling behind, it’s about ensuring you’re building something meaningful. Because in the end, it’s not just about building fast; it’s about building something people actually want.</strong></p><blockquote>Here’s the choice: do you want to create a fast and flashy TikTok video (a showy AI demo that everyone will forget)? Or do you want to create <em>Interstellar</em> movie — a useful AI feature that stands the test of time? Both approaches can work depending on your goals, but only one builds enduring value. You know which one.</blockquote><blockquote>Adopt the “Apple-tude” — their approach to discovering AI use cases.</blockquote><p><strong>In technology, as in chess, the best move is often thinking three steps ahead instead of rushing the next one.</strong> Apple knows they’ve got the resources, talent, and moat to win the game in the long term.</p><p>Apple doesn’t aim to be an “AI company.” It aims to use AI wisely.</p><p>So take a page from Apple’s playbook:</p><blockquote>Think deeply before building. Because in the end, it’s not just about being fast or flashy. It’s about building something useful.</blockquote><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=dc7a80bb1551" width="1" height="1" alt=""><hr><p><a href="https://medium.com/design-bootcamp/how-apple-brainstorms-ai-use-cases-dc7a80bb1551">How Apple Brainstorms AI Use Cases</a> was originally published in <a href="https://medium.com/design-bootcamp">Bootcamp</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Who will win Enterprise Agentic AI? Why is it not working (yet)]]></title>
            <link>https://medium.com/@shashverse/who-will-win-enterprise-agentic-ai-why-is-it-not-working-yet-7f1aed42fd99?source=rss-9a5497c44b4e------2</link>
            <guid isPermaLink="false">https://medium.com/p/7f1aed42fd99</guid>
            <category><![CDATA[generative-ai-tools]]></category>
            <category><![CDATA[enterprise-technology]]></category>
            <category><![CDATA[ai-startups]]></category>
            <category><![CDATA[ai-agent]]></category>
            <category><![CDATA[llm-agent]]></category>
            <dc:creator><![CDATA[Shashwat Tiwari]]></dc:creator>
            <pubDate>Thu, 05 Dec 2024 17:46:56 GMT</pubDate>
            <atom:updated>2024-12-05T17:46:56.084Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*89bmRwh5EsNiXbB0" /><figcaption>Agentic AI For Enterprise: Promising Startups</figcaption></figure><p>In the race to redefine work, businesses are obsessed with a single question: <strong>When will AI agents live up to their promise?</strong> The promise of autonomous systems that can navigate complex workflows, interact with multiple applications, and execute business processes with minimal human intervention.</p><p>Despite the billions flowing into this space, we’re still waiting for the breakthrough that will make Agentic AI truly enterprise-ready.</p><p>Every enterprise leader I speak with shares the same vision: AI agents orchestrating their business processes, by seamlessly integrating with legacy &amp; modern systems in dynamic environments &amp; delivering reliable outcomes. The potential efficiency gains are astronomical. But there’s a catch — <em>or rather, several catches.</em></p><p>The strength of LLMs is also their weakness. What makes them creative and generalizable across domains also causes them to hallucinate, as <strong>they are inherently probabilistic or stochastic in nature.</strong></p><p><strong>But enterprise operations demand deterministic outcomes.</strong> A system that works brilliantly 95% of the time is still fundamentally unusable for enterprises.</p><p>This isn’t just about accuracy — it’s about trust and control. <strong>Enterprise leaders need systems they can govern, audit, and trust with mission-critical operations.</strong> Current agent architectures don’t provide this level of reliability.</p><p>According to me, the winners in this space will need four things:</p><ol><li><strong>A radically different technical approach that solves the “determinism” and “interoperability” problem</strong></li><li><strong>A talented team with enterprise DNA combined with cutting-edge AI expertise</strong></li><li><strong>Substantial funding to survive the experimentation phase</strong></li><li><strong>A deep understanding of both developer and enterprise needs</strong></li></ol><p>Based on these criteria, I see few companies with potentially transformative approaches to this challenge:</p><p><a href="https://www.emergence.ai/"><strong>Emergence AI</strong></a></p><p>Founded by IBM Research veterans and backed by $100M, <strong>Emergence AI is tackling perhaps the hardest problem in enterprise AI: Cross-System Orchestration.</strong></p><p>Their differentiator is an ability to bridge any enterprise system, modern or legacy. Think of it as a meta-agent that can coordinate complex workflows across your entire technology stack — something even giants like Microsoft and Salesforce haven’t cracked (yet). Their secret sauce appears to be a hybrid approach combining deterministic API interactions with adaptive web navigation, making their agents both reliable and flexible. Emergence AI, with its focus on enterprise compatibility, has the best shot at winning enterprise trust.</p><p><a href="https://sdsa.ai/"><strong>Dev/agents/</strong></a></p><p><strong>When the team that built Android raises $56M to create an operating system for AI agents, you pay attention.</strong></p><p>Their thesis is compelling: just as Android provided a standard platform that transformed the smartphone era, the AI agent ecosystem needs a unified foundation to truly scale. Expected to launch in mid-2025, their platform promises to do for AI agents what Android did for mobile apps: provide a standard development environment, handle core infrastructure concerns, and create a sustainable economic model through transaction fees. <a href="https://www.theverge.com/2024/11/27/24307525/android-leaders-dev-agents-ai-agent-operating-system-startup">Read More</a></p><p><a href="https://www.ema.co/"><strong>Ema</strong></a></p><p>With $50M in funding and leadership from ex-CPO of Coinbase veteran Surojit Chatterjee, Ema is approaching the problem through the lens of business process automation and building universal AI employees.</p><p>Their <a href="https://www.ema.co/blog/agentic-ai/generative-workflow-engine-building-emas-brain">generative workflow engine</a> is a fundamental rethinking of how enterprises model and execute processes. <a href="https://www.ema.co/blog/agentic-ai/why-does-model-routing-not-work-how-emafusion-is-revolutionizing-the-art-and-science-of">Emafusion</a> combines the best of 100+ LLMs to deliver highest accuracy at lowest cost. They have won some big names as early adopters, and overall it gives me a feeling that they know what they’re doing. Ema, with its verticalized approach, may carve out lucrative niches while others battle for dominance.</p><h4><strong>The incumbents aren’t far behind.</strong></h4><p>Microsoft’s <a href="https://www.microsoft.com/en-us/research/articles/magentic-one-a-generalist-multi-agent-system-for-solving-complex-tasks/">Magentic-One architecture</a> looks promising, while Salesforce’s <a href="https://www.salesforce.com/in/agentforce/">Agentforce</a> has the biggest advantage: customer data. OpenAI’s <a href="https://github.com/openai/swarm">Swarm AI</a> shows potential with its concepts of “routines” and “handoffs.” Amazon has introduced multi-agent capabilities to Bedrock, and <a href="https://www.servicenow.com/now-platform/generative-ai.html">ServiceNow</a> has brought AI agents to workflows. There are several other promising Startups like <a href="https://www.lyzr.ai/enterprise/">Lyzr </a>that are constantly innovating.</p><p>Anthropic is actively helping solve enterprise adoption blockers. They recently open-sourced <a href="https://www.anthropic.com/news/model-context-protocol">Model Context Protocol</a>, a new standard for connecting AI assistants to systems where data lives, including content repositories, business tools, and development environments. Similar to how RAG transformed LLM applications, this protocol could be key to making agents enterprise-ready.</p><p><strong>Agentic AI is a marathon, not a sprint. The technical hurdles are immense, but so is the opportunity.</strong> Businesses are willing to pay millions for proof of concepts and GenAI strategy decks, <a href="https://www.thestack.technology/accenture-ai-bookings-2024/">Accenture booked $3B</a> in GenAI related projects, $1B in august alone — proof that the demand is real.</p><p>The next 6–12 months will be critical. We’ll see early enterprise deployments of agent systems, probably in narrow but high-value use cases first. <strong>The winners will emerge from those who can deliver “reliability” at scale.</strong></p><p>The opportunity is massive. Just as cloud computing transformed IT, agentic AI will transform how enterprises operate. <strong>The question isn’t if, but when — and who will unlock it first.</strong></p><p><strong>The smart money isn’t betting on who has the most impressive demo. It’s betting on who can make agents boring — reliable enough to be unremarkable. In enterprise tech, that’s when you know you’ve won.</strong></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=7f1aed42fd99" width="1" height="1" alt="">]]></content:encoded>
        </item>
    </channel>
</rss>