<?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 Ahmed Shamy on Medium]]></title>
        <description><![CDATA[Stories by Ahmed Shamy on Medium]]></description>
        <link>https://medium.com/@ashamy?source=rss-8ad754b471e3------2</link>
        <image>
            <url>https://cdn-images-1.medium.com/fit/c/150/150/1*vsNBWmLX13z6egBCx3iN3Q.jpeg</url>
            <title>Stories by Ahmed Shamy on Medium</title>
            <link>https://medium.com/@ashamy?source=rss-8ad754b471e3------2</link>
        </image>
        <generator>Medium</generator>
        <lastBuildDate>Sun, 22 Nov 2026 05:17:40 GMT</lastBuildDate>
        <atom:link href="https://medium.com/@ashamy/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[Architecting the Agentic Enterprise: A CIO’s Roadmap to Autonomous Action at Scale]]></title>
            <link>https://medium.com/@ashamy/architecting-the-agentic-enterprise-a-cios-roadmap-to-autonomous-action-at-scale-8edd0b7ad67d?source=rss-8ad754b471e3------2</link>
            <guid isPermaLink="false">https://medium.com/p/8edd0b7ad67d</guid>
            <category><![CDATA[google-cloud-platform]]></category>
            <category><![CDATA[enterprise-architecture]]></category>
            <category><![CDATA[agentic-ai]]></category>
            <category><![CDATA[strategy]]></category>
            <dc:creator><![CDATA[Ahmed Shamy]]></dc:creator>
            <pubDate>Thu, 30 Apr 2026 16:14:31 GMT</pubDate>
            <atom:updated>2026-04-30T19:34:12.727Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*J0bg6V2YtaaFIQbyjzj3RQ.png" /></figure><p>When I sit with executives in boardrooms, the question I keep getting asked is: “What do we need to do to be Agentic-ready?”</p><p>That question taps into the thrill of exploring the possibilities and effects of becoming an Agentic-driven organization. In order to answer it, we first need to come to an agreement about what “being Agentic” is. The terms have changed quickly: It used to be nothing more than Agent chatbots that would talk to data sources, and now we have a new era where Agents are able to act, even process transactions on behalf of end-users.</p><p>In preparation for this “<em>Agentic Action</em>” era, it is critical that organizations adequately digitally identify and secure agents, while ensuring they have access to the right data in the relevant context (regardless of where this data is). This foundation is then built, and Agents must be designed within an <em>Engine</em> where they are controlled (orchestrated) and evaluated within a production-grade framework.</p><p>It is imperative that these foundational elements and the engine are vertically stacked. This ensures organizations remain in full control of their strategic vision, long-term costs, and — more importantly — the security and governance of the full stack.</p><p>In this article, I’ll pull you, as a CXO leader, toward the three building blocks needed for Agent readiness — Identity, Security, and Data — addressing both the “why” and the “how” to lead your organization in the embrace of<em> Agentic action</em> at scale. Next, we will delve into the Engine of Autonomy.</p><p><strong>Pillar 1: Identity — The Anchor Of Accountability</strong></p><p>Agents’ identity is a critical element of building the foundation of an Agentic-driven organization, especially in regulated industries such as finance and health care, where each autonomous action must be cryptographically signed to satisfy compliance and accountability requirements. This is where agents will be treated as distinct digital identities that will go through an authentication and authorization before they can do things or access data.</p><p>Traditionally, humans were identified through the use of <em>usernames or passwords</em>, whereas machines and services were given <em>service accounts</em> designed to define their scope. Agents have found themselves in the “gray area” between the two — regularly borrowing human tokens or leveraging over-privileged service accounts not designed for autonomous logic.</p><p>First, for Authentication (AuthN), Agents will have to be assigned a unique and verifiable ID before all work can proceed. It is more than a username, it is a digital certificate applied for the “Who” to define the agent; the key stipulation to which the person as an agent is subject.</p><p>Second for Authorization (AuthZ): clearly stating what an agent can do. With protocols and mechanisms like OAuth, an agent can act as an authorized citizen of a human user using delegated authority and maintain a consistent audit trail (immutable, clear logs) of its delegated authority.</p><p>Last but not least, unlike the classical service accounts, which can last for many years, Agent Identities should in part be short-lived — defined to serve a definite purpose and retired after job completion to reduce the size of the firm’s attack surface.</p><p>Using <em>Google Gemini Enterprise Agent Platform, </em>you can assign each Agent a Unique Cryptographic ID using <em>Agent Identity service</em>. This way every action is linked back to an agent signature, not just one generic cloud ID or access key.</p><p>Organization can use <em>Google Cloud Agent Registry</em> which acts like the “Yellow Pages” — a single source of truth where every approved agent is indexed with its identity, tools and assigned permissions. And with the<em> Google Cloud Agent Gateway,</em> you can use Agent Identity to track every single interaction. When an agent tries to invoke a network API that isn’t authorized for it, <em>Agent Gateway </em>blocks that API at the network level based on identity profile — as some people call it “air traffic control.”</p><p>In this way, Google Cloud is equipping enterprises with an Identity Stack that allows enterprises to move beyond simple user-based security. By integrating Verified Agent IDs, a centralized Agent Registry, and the Agent Gateway, organizations are moving from asking, “Who is the user?” to the more critical question: “Which agent is acting on behalf of which user, and is this action within its specific mandate?”</p><p><strong>Pillar 2 — Security: The Perimeter of Purpose</strong></p><p>While identity helps us understand who these agents are and what they are allowed to do, security is the boundary that prevents not only traditional network intrusions but also protects against various types of attacks and logic manipulation.</p><p>In an agentic system, various different types of attacks need to be considered, such as an injection of commands to bypass internal command instructions of the agent. This leads us to the introduction of a<em> logic filter</em> that scans every input for injection attacks and every output for sensitive data leaks (PII) or other harmful content. We also have to deal with <em>Data Exfiltration</em>: preventing agents from summarizing and/or exporting sensitive information to unauthorized destinations.</p><p>There should also be <em>Tool-Call Validation</em>, in which when an agent calls a tool to an API or data source, the call can be intercepted for validation and verification with parameters considered safe and also trust where the call is being made.</p><p><em>Cloud Armor for Agents</em> sits at the center of your Google Cloud defense system and acts as the first line of defense. It exploits global DDoS and WAF (Web Application Firewall) capabilities to validate that only legitimate, secure traffic connects to your agents. With the <em>Google Cloud Adaptive Protection service</em>, it is intelligent enough to detect and block advanced, bot-enabled attacks that your agent can’t handle for processing capabilities.</p><p>Organizations also use<em> Google Cloud Model Armor </em>with the <em>Gemini Enterprise platform</em>. It scans every prompt for “jailbreak” attempts and every response for exfiltration. There is also the <em>Google Cloud Agent Sandbox </em>that lets agents run code or deal with files in secure containers, a “bubble” environment, which is able to thwart any attempts to breach your infrastructure.</p><p><strong>Pillar 3 — Data: Your Basis for Action.</strong></p><p>There is no question that data is the most essential part of the puzzle. Organizations have their data spread across all kinds of clouds and back-ends. For agents to be able to take action, and to become autonomous, data must move from a passive “<em>System of Record</em>” (where data is being passively reported) to an active “<em>System of Action</em>” for autonomous reasoning.</p><p>This can only be achieved by having agents access all the data they need no matter where it resides. And, importantly, we need to build a context layer on top of this data itself to map the user intent to a business and technical context of the data they are trying to access.</p><p>This guarantees that the agent can pull and query the relevant information from all sources without experiencing hallucinations. This layer of translation is known as the <em>Semantic Layer.</em></p><p>Google Cloud introduced <em>Agentic Data Cloud</em> based on the <em>Open Lakehouse</em> structure where all the data — regardless of where it resides — is in the <em>Iceberg </em>format. This way, the data is represented to the semantic layer as one logical data source, even though the data lives out in the different clouds or sources.</p><p>The agents’ unified cross-cloud semantic layer, powered predominantly by <em>Looker</em>, sits on top of the <em>Agentic Data Cloud</em> so that you have the correct information for every agent, with zero hallucinations. One of the major benefits for CIOs, in this setup, is that Google ensures no data lock-in in the open Iceberg standard. This way your previous investments get fueled up and enhanced; Google actually builds from where you are. By using a <em>Unified Semantic Layer </em>together with <em>cross-cloud openness</em>, you provide your agents with “intelligence” that enables them to act without the technical capital of legacy data silos.</p><p><strong>The Engine of Autonomy</strong></p><p>So far, we have introduced the building blocks to prepare a production-ready Agentic system as laid out in our initial blueprint. The missing piece is the Engine of Autonomy that makes it possible to now operationally establish and create production-ready agents that can be operated. Those agents have access to data in all its different locations, with its correctly specified context for being used.</p><p><em>Gemini Enterprise App </em>is the enterprise front door to Agents and AI. This is where all the organizations’ business and technology users can build agents. The key differentiator that <em>Google</em> brings is the capability to design and create agents from basic chat agents to very complex autonomously orchestrating agents. Agent workflows can be built with the improved <em>Agent Designer</em> using natural language or a visual interface. In addition, you can easily connect to complex agents built by your organization’s engineers and developers on the <em>Gemini Enterprise Agent Platform</em>.</p><p>All these agents are firmly located and linked only to your data, limited to your organization’s scope and powered by Google’s state-of-the-art Gemini models. Using the new <em>Inbox command center</em>, leaders and employees can watch, direct, and manage agent actions and activities in real-time — getting full visibility into what tasks are performed and where a “human-in-the-loop” approval is needed.</p><p>If <em>Gemini Enterprise App </em>is the cockpit, <em>Gemini Enterprise Agent Platform</em> is the engine room. This single management plane serves as a lifecycle controller for the entire agent’s lifecycle (everything from initial logic through global orchestration and evaluation). By centralizing these operations, you can ensure your agentic strategy is not just a batch of scripts; it’s an enterprise-level solidity built in.</p><p><em>Gemini Enterprise Agent Platform</em> uses a hybrid approach to build agents to keep IT from becoming a bottleneck on its own. This platform offers options from no-code for rapid prototyping, to the Agent Development Kit (ADK) for deep process, programmatic customization. This flexibility lets your business grow at two different levels: respond quickly and get you where you need to be, for the business, but at the same time maintain the technical depth you need to get through complex integrations. This is also where true autonomy happens through Agent-to-Agent (A2A) workflows</p><p><em>Gemini Enterprise Agent Platform </em>coordinates multiple specialized agents to solve complex, multi-day problems, managing the context and hand-offs between them so your digital workforce operates as a cohesive unit. This orchestration ensures that even the most complex business processes are handled with the same consistency and logic you’ve defined in your foundation pillars.</p><p>Last, you can’t scale what you can’t trust and so the platform’s <em>Evaluation and Simulation </em>suite is the “litmus test” for production. Before any agent is deployed, it must be tested against synthetic workloads to confirm that it meets strong business KPIs and safety standards. Upon going live, <em>Gemini Enterprise Agent Platform </em>offers full observability — leaders can check an agent’s reasoning pathway in real-time and manage its performance as your business changes.</p><p><strong>From Strategy to Autonomy</strong></p><p>Moving to an Agentic Enterprise isn’t just one tech upgrade. It is a complete change in our way of thinking about the interaction between people, data, and digital actions. When you have aligned your roadmap on the three fundamentals of Identity, Security, and Data, you have resolved the “Trust Gap” that prevents the vast majority of AI projects from being trapped in the pilot phase.</p><p>You have created the ecosystem where everything can be audited, all boundaries are enforced, and every decision is based on real-world business context. With the <em>Gemini Enterprise App and Gemini Enteprise Agent Platform</em> as your <em>Engine of Autonomy,</em> you are no longer managing a series of high-potential prototypes. You are running a scalable, secure, and context-aware Agents that ultimately can drive meaningful ROI across your organization.</p><p>The era of AI “talking” is now becoming the era of AI “doing.” The boardroom isn’t questioning whether we should trust this technology anymore — they are awaiting the architecture for that trust. What you have is a blueprint, pillars, and engine. Now it’s time to go beyond the MVP and guide your organization towards the agentic future at scale.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=8edd0b7ad67d" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Revolutionizing the Drive-Thru with AI: A New Era for QSRs]]></title>
            <link>https://medium.com/@ashamy/revolutionizing-the-drive-thru-with-ai-a-new-era-for-qsrs-e6648ff613f3?source=rss-8ad754b471e3------2</link>
            <guid isPermaLink="false">https://medium.com/p/e6648ff613f3</guid>
            <category><![CDATA[drive-thru]]></category>
            <category><![CDATA[google-cloud-platform]]></category>
            <category><![CDATA[ai-transformation]]></category>
            <category><![CDATA[quick-service-restaurant]]></category>
            <category><![CDATA[generative-ai-use-cases]]></category>
            <dc:creator><![CDATA[Ahmed Shamy]]></dc:creator>
            <pubDate>Wed, 11 Dec 2024 14:02:03 GMT</pubDate>
            <atom:updated>2024-12-11T14:02:03.396Z</atom:updated>
            <content:encoded><![CDATA[<p>This article is based on my firsthand experience working with leading Quick Service Restaurants (QSRs) on automating their drive-thrus, and with the Google Product teams in introducing the Google AI Food Ordering Assistant, a cutting-edge AI solution designed to revolutionize the industry.</p><p>Drive-thrus have long been a critical revenue stream for QSRs, but their importance has skyrocketed in recent years. Some reports estimate that drive-thrus now generate nearly 70% of revenue for North American QSRs, a trend accelerated by the pandemic and shifting consumer preferences. This surge in reliance on drive-thrus has exposed operational challenges and unveiled exciting opportunities to leverage technology for increased efficiency and revenue generation. In this article, we’ll explore how innovative AI-driven solutions are revolutionizing the drive-thru experience, addressing key challenges, and unlocking new possibilities for QSRs.</p><h3>The Drive-Thru: A Critical Revenue Stream</h3><p>The drive-thru experience is a critical aspect of the QSR business, impacting revenue, operations, and customer satisfaction. For QSR corporations and franchises, a primary focus is optimizing operational costs. This involves improving the efficiency of their restaurant team members by minimizing the time spent on order taking, allowing them to concentrate on back-of-house tasks like meal preparation and serving dine-in customers. Generating more revenue is another key priority, primarily through upselling.</p><p>Currently, many QSRs rely on upsell scripts for their crew, but this approach can be uncomfortable for team members and difficult to track effectively. This often results in missed revenue opportunities. Furthermore, long queues at drive-thrus can also lead to lost revenue as impatient customers may choose to go elsewhere. Ensuring a smooth and quick ordering process not only improves customer satisfaction but also reduces “bailout” rates and encourages repeat business.</p><h3>The Rise of AI-Powered Drive-Thrus</h3><p>Technology vendors recognized these challenges and sought to address them by automating the drive-thru ordering process. The idea was to support human team members with automated agents capable of understanding guest orders, submitting them to the restaurant’s point of sale (POS) system, and sending them to the kitchen for preparation.</p><p>However, the complexity of guest orders, with millions of possible permutations for even a single burger, presented a significant obstacle. This challenge remained unconquerable until the advent of Large Language Models (LLMs) and generative AI (GenAI).</p><p>Before LLMs and GenAI, attempts to automate the drive-thru ordering process, such as through enforced learning bots and automated call centers, faced significant hurdles. Not only did these systems struggle to achieve satisfactory accuracy, but they were also complex and time-consuming to build and maintain. This often resulted in high development costs and made them financially impractical for widespread adoption. Furthermore, these systems often required frequent intervention from restaurant staff to correct errors or complete orders, adding to their workload rather than reducing it.</p><h3>The Power of LLMs and GenAI</h3><p>This is where the transformative power of Generative AI and Large Language Models (LLMs) comes into play. Unlike previous approaches that relied on rigid rule-based systems or limited machine learning models, LLMs possess an unprecedented ability to understand the nuances and complexities of human language. This is particularly crucial in the drive-thru context, where customer orders can be highly varied, include slang and abbreviations, and even be incomplete or grammatically incorrect.</p><p>While we are still in the early stages of building systems that solely rely on AI agents to handle orders, the results are very promising, with some QSRs further along in their journey than others.</p><p>Complementing AI Agents with humans in the loop (HIL) is one approach that some technology vendors are using today to improve accuracy. However, this raises questions about outsourcing labor and scalability, and it may be perceived as simply outsourcing restaurant team members.</p><p>Google stands out as a leader in introducing fully automated AI agents that can handle the entire ordering process. For Google’s automated drive-thru systems, human intervention by restaurant team members is only allowed when necessary or when the guest explicitly requests help from a team member. This ensures that while the market and customers are getting used to submitting their orders to AI agents, there is still an option for those who prefer to order their favorite meals in the traditional way. Later in this article, we will discuss the impact of this flexibility on automation rates.</p><h3>Beyond Automation: Unlocking New Possibilities</h3><p>Beyond simply automating orders, AI-powered drive-thrus open doors to revolutionary upselling strategies. Static upselling scripts can be replaced with dynamic systems that adapt to factors like restaurant location, popular items, time of day, and even individual customer preferences (where regulations allow for identifying the customer, either through their voice, plate number, or other means). This allows for real-time personalization and optimization of upsell offers, maximizing revenue potential.</p><p>Imagine a drive-thru experience where digital menu boards display personalized recommendations based on a customer’s past order history. This level of personalization, enabled by integrating AI with cameras and audio systems, can significantly enhance the customer experience and drive higher order values. In fact, according to the 2024 QSR magazine drive-thru report, 44% of customers desire more personalized drive-thru experiences, highlighting the potential of AI in catering to this growing demand.</p><p>Furthermore, AI facilitates seamless integration with loyalty programs, allowing for automatic redemption of offers and points, further incentivizing customer engagement. The QSR magazine report also indicates that 57% of customers have expressed interest in using loyalty programs integrated into the drive-thru experience. This presents a significant opportunity for QSRs to leverage AI for increased customer loyalty and revenue generation.</p><p>Suggestive selling through AI can also address critical questions related to dietary and allergy restrictions, ensuring a safe and personalized experience for every customer.</p><p>Another significant advantage of AI-powered drive-thrus using LLMs and GenAI is their inherent ability to support multiple languages. This opens doors to wider customer reach and inclusivity, enabling QSRs to serve diverse communities without relying on multilingual staff.</p><h3>Key Success Factors for AI-Powered Drive-Thrus</h3><p>QSRs prioritize several key factors when evaluating the success of AI-powered drive-thru solutions. These factors often vary in importance depending on the specific goals and priorities of each QSR.</p><ul><li><strong>AI Automation Rate:</strong> For QSRs focused on team member efficiency and operational cost optimization, the AI automation rate (the percentage of orders handled by AI agents) is a crucial metric. Some QSRs set ambitious benchmarks, aiming for automation rates exceeding 90%.</li><li><strong>Upsell Rates and Check Upsize:</strong> QSRs prioritizing revenue generation place significant emphasis on the AI’s upselling features and their potential to drive increased order values.</li><li><strong>Order Accuracy:</strong> Regardless of other priorities, order accuracy is paramount for all QSRs. High accuracy rates are essential to minimize the need for team member intervention and ensure a smooth and efficient ordering process.</li><li><strong>Order Time:</strong> Reducing the time each guest spends at the order station is another critical success factor, particularly for QSRs experiencing high “bailout” rates due to long wait times. By optimizing order time, QSRs can not only recover lost revenue but also create opportunities for additional upselling without negatively impacting overall throughput.</li></ul><h3>Addressing Adoption Challenges</h3><p>When evaluating AI-powered drive-thru systems, QSRs prioritize the ability to fully automate the order process without team member intervention, aiming for high AI automation rates. However, achieving this can be challenging due to concerns about culture change and job displacement among team members.</p><p>To address this, it’s crucial to emphasize that these systems are designed to enhance, not replace, human roles. Providing comprehensive training to crew members on how to effectively collaborate with AI agents is essential for building trust and ensuring smooth adoption.</p><p>Vendors must also equip QSRs with robust monitoring tools, including dashboards, reports, and alerts, to track intervention rates and identify areas for improvement. This data-driven approach empowers QSRs to address concerns, optimize crew performance, and foster confidence in the AI’s capabilities.</p><p>While maximizing automation is important, it’s equally crucial to prioritize guest satisfaction. Forcing team members to completely avoid intervention can negatively impact the customer experience. Guests should always have the option to request human assistance, especially during the initial stages of system adoption.</p><p>To ease the transition for guests, QSRs can implement strategies such as clearly communicating the option for team member help and, in the case of dual-lane drive-thrus, dedicating one lane to traditional ordering during the initial rollout phase.</p><p>Another challenge for QSRs adopting these systems is the upfront cost and complexity of integrating with their existing technology stack, which includes headsets, point-of-sale (POS) systems, and kitchen display systems. While AI-powered drive-thru systems can readily integrate with modern, API-driven digital systems, challenges arise when working with older, non-digital, or non-cloud-based infrastructure.</p><p>Integrating with POS systems presents a unique set of challenges related to menu updates and limited-time offers (LTOs). The accuracy and effectiveness of AI drive-thru systems depend heavily on the quality and timeliness of data from POS systems. If menu updates, including changes in item availability or the introduction of LTOs, are not reflected instantly in the AI system, it can lead to customer dissatisfaction and order errors.</p><p>It is critical for QSRs embarking on their AI-powered drive-thru journey to adopt a phased approach, starting with a limited number of pilot programs across different locations and geographies before progressing to extended pilots and, eventually, full-scale rollouts.</p><p>During the pilot phase, the importance of creating a comprehensive business case cannot be overstated. This business case should encompass not only metrics related to cost and operation optimization and revenue generation but also the impact on restaurant traffic and the readiness of both team members and guests to embrace these systems. Careful consideration of these factors will help ensure a successful transition and prevent potential negative impacts, such as customers choosing alternative QSRs that offer a more familiar drive-thru experience.</p><h3>The Future of AI-Powered Drive-Thrus</h3><p>The future of AI in the QSR industry is incredibly promising. As technology continues to advance, we can expect even greater levels of personalization, efficiency, and customer satisfaction in the drive-thru experience. AI has the potential to transform not only how orders are taken but also how QSRs manage their operations, interact with customers, and optimize their businesses.</p><p>If you’re interested in exploring how Google’s AI Food Ordering Assistant can revolutionize your drive-thru, please reach out to your Google Cloud account representative to learn more.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=e6648ff613f3" width="1" height="1" alt="">]]></content:encoded>
        </item>
    </channel>
</rss>