<?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 Haritha K J on Medium]]></title>
        <description><![CDATA[Stories by Haritha K J on Medium]]></description>
        <link>https://medium.com/@harithaharry001?source=rss-a8bf23b2641e------2</link>
        <image>
            <url>https://cdn-images-1.medium.com/fit/c/150/150/1*DkZxpG5LL92ccKJm3ip15w.jpeg</url>
            <title>Stories by Haritha K J on Medium</title>
            <link>https://medium.com/@harithaharry001?source=rss-a8bf23b2641e------2</link>
        </image>
        <generator>Medium</generator>
        <lastBuildDate>Fri, 20 Nov 2026 04:58:36 GMT</lastBuildDate>
        <atom:link href="https://medium.com/@harithaharry001/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[From Talend to AWS Glue: A Modernization Win]]></title>
            <link>https://hai2.ai/from-talend-to-aws-glue-a-modernization-win-1f92741fa36a?source=rss-a8bf23b2641e------2</link>
            <guid isPermaLink="false">https://medium.com/p/1f92741fa36a</guid>
            <category><![CDATA[aws]]></category>
            <category><![CDATA[talend]]></category>
            <category><![CDATA[llm]]></category>
            <category><![CDATA[agentic-ai]]></category>
            <category><![CDATA[legacy-modernization]]></category>
            <dc:creator><![CDATA[Haritha K J]]></dc:creator>
            <pubDate>Tue, 11 Nov 2025 14:00:18 GMT</pubDate>
            <atom:updated>2025-11-11T14:18:27.194Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*oDDnUWVOJVe6xcOkutQtpQ.png" /><figcaption>HAI Modernize</figcaption></figure><p>At <a href="https://www.presidio.com/solutions/hai-modernize/"><strong>HAI Modernize</strong></a>, our mission is simple — to redefine how organizations modernize their legacy systems. But every now and then, a project comes along that pushes our limits, proving just how powerful and adaptable our modernization framework truly is.</p><p>Recently, we partnered with a Fortune 500 company facing a complex modernization challenge — one that had already consumed significant time, resources, and effort before we stepped in.</p><h3>The Challenge</h3><p>The client had been using <strong>Talend</strong> for their data integration and ETL workloads for several years. Over time, the environment had grown large and complex, with hundreds of jobs, dependencies, and custom components built across different teams.</p><p>They had been trying to <strong>manually convert </strong>their Talend jobs to <strong>AWS Glue</strong> for almost a year. Despite the effort, the manual process was slow, difficult to manage, and not producing results at the scale they needed.</p><p>With a <strong>multi-million-dollar Talend license renewal</strong> approaching in just three months, the pressure to complete the migration quickly was high.</p><p>At that point, the client<strong> reached out to Presidio</strong> for help. We worked closely with their team to explore multiple options, including specialized tools designed for Talend job conversion. However, none of those tools delivered the accuracy or scalability required to complete the project successfully.</p><p>That’s when we introduced them to <strong>HAI Modernize</strong> — a flexible, AI-driven modernization framework capable of analyzing, transforming, and optimizing complex workloads.</p><h3>The HAI Modernize Approach</h3><p>Although <strong>HAI Modernize</strong> was not originally built specifically for Talend-to-Glue conversions, its AI-driven Modernize Agents were designed to adapt to complex modernization challenges. These agents can analyze, understand, and transform legacy workloads intelligently, regardless of the platform or language.</p><p>The migration included Talend jobs of varying complexity, from medium to highly complex. Many of these jobs contained thousands of lines of code and multiple interconnected files. The available tools in the market, even those specifically designed for Talend job conversion, were not able to perform effectively at this scale.</p><p>With HAI Modernize, we followed a well-defined process tailored to handle complex workloads. The framework uses a network of specialized agents, each responsible for a specific part of the modernization process:</p><ul><li><strong>Discovery Agents</strong> analyze and document what each Talend job is doing, including dependencies, components, and transformation logic.</li><li><strong>Conversion Agents</strong> reconstruct that logic inside AWS Glue, ensuring a precise, one-to-one translation of the original Talend jobs.</li><li><strong>Validation Agents</strong> verify that the converted jobs behave exactly like the originals, preserving accuracy and data integrity.</li></ul><p>The Talend jobs were originally defined in XML and other file formats and often included custom Java components and embedded SQL queries. These agents worked together to break down the jobs, understand their logic, and convert them accurately into AWS Glue equivalents.</p><p>This multi-agent approach enabled the team to handle jobs of varying complexity — low, medium, and high — with consistency, precision, and speed, resulting in a smooth and reliable migration to AWS Glue.</p><h3>The Outcome</h3><p>Within a week, we were able to build custom agents, understand the Talend source, and successfully convert jobs across all levels of complexity. This effort established a solid foundation for the broader migration, enabling the team to confidently move forward and convert more than 900 remaining jobs.</p><p>What had taken nearly a year of manual effort was achieved in a matter of weeks, thanks to the intelligence and automation built into <strong>HAI Modernize</strong>.</p><h3>Key Takeaways</h3><ul><li><strong>Adaptability:</strong> Even though <strong>HAI Modernize</strong> wasn’t originally built for Talend-to-Glue conversions, its AI-driven framework adapted seamlessly to the challenge.</li><li><strong>Accuracy and Reliability:</strong> The custom agents ensured precise, one-to-one conversions without any refactoring or functional changes.</li><li><strong>Scalability:</strong> The approach established a solid base for the team to continue and scale the migration across 900 additional jobs.</li><li><strong>Business Impact:</strong> The project saved the client millions in potential licensing costs and accelerated their overall cloud modernization journey.</li></ul><h3>Final Thoughts</h3><p>Modernization is not just about moving from one platform to another. It’s about understanding legacy systems, preserving the business value they contain, and enabling agility for the future.</p><p>This project demonstrated that with the right combination of AI, automation, and engineering expertise, even the most complex modernization challenges can be solved efficiently and reliably.</p><p>At <strong>HAI Modernize,</strong> we don’t just migrate. We modernize with clarity, intelligence, and impact.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/388/1*-9-wqc_yQa5EsRPzwrrQQg.png" /></figure><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=1f92741fa36a" width="1" height="1" alt=""><hr><p><a href="https://hai2.ai/from-talend-to-aws-glue-a-modernization-win-1f92741fa36a">From Talend to AWS Glue: A Modernization Win</a> was originally published in <a href="https://hai2.ai/">Presidio HAI</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[I Gave Google’s Agent Payments Protocol (AP2) a Try: Here’s What I Learned About the Future of…]]></title>
            <link>https://medium.com/@harithaharry001/i-gave-googles-agent-payments-protocol-ap2-a-try-here-s-what-i-learned-about-the-future-of-fcbe009d1b86?source=rss-a8bf23b2641e------2</link>
            <guid isPermaLink="false">https://medium.com/p/fcbe009d1b86</guid>
            <category><![CDATA[google-adk]]></category>
            <category><![CDATA[google]]></category>
            <category><![CDATA[agentic-ai]]></category>
            <category><![CDATA[aws]]></category>
            <category><![CDATA[ap2]]></category>
            <dc:creator><![CDATA[Haritha K J]]></dc:creator>
            <pubDate>Thu, 23 Oct 2025 17:48:58 GMT</pubDate>
            <atom:updated>2025-10-24T08:31:05.951Z</atom:updated>
            <content:encoded><![CDATA[<h3>I Gave Google’s Agent Payments Protocol (AP2) a Try: Here’s What I Learned About the Future of Autonomous Commerce</h3><p><em>How I proved AP2 really works by implementing it from scratch — and why this protocol matters for everyone</em></p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*kZfYdtgq0raXA6L8.png" /></figure><p><strong>Imagine this scenario:</strong></p><p>Your parents’ favorite band from their youth just announced a reunion tour. Tickets go on sale at midnight tonight. You know from experience they’ll sell out in minutes — and you want to surprise your parents with this gift.</p><p>But you have a crucial work deadline tomorrow. You can’t stay up until midnight fighting ticket bots and refreshing the page. You need sleep.</p><p>What if you could simply tell your AI agent:</p><blockquote>“Buy 2 concert tickets as soon as they go on sale. Seated together, under $200 each, as close to the stage as possible”</blockquote><p>— and then just go to sleep?</p><p>The agent monitors the ticketing site. At midnight, it navigates the queue, finds two seats in Section 104 Row F for $185 each — meeting all your criteria. It completes the purchase automatically while you sleep.</p><p>You wake up at 7 AM to a notification:</p><blockquote>“Concert tickets secured. Section 104, Row F, Seats 12–13. $370 total. Tickets in your email.”</blockquote><p>You get to surprise your parents with an unforgettable gift. You didn’t lose sleep or fight with bots.</p><p><strong>This isn’t some distant future. This is happening now.</strong></p><p>This is what the <strong>Agent Payments Protocol (AP2)</strong> enables. And I just built a working implementation to prove it actually works.</p><h3>What Is AP2? (And Why Should You Care)</h3><p>In September 2025, Google announced something that could fundamentally reshape how we buy things online: the <strong>Agent Payments Protocol</strong>.</p><p>Developed in collaboration with <strong>over 60 major organizations</strong> — including Mastercard, PayPal, American Express, Coinbase, Adyen, and Salesforce — AP2 is an open protocol designed to solve a critical problem:</p><blockquote><strong><em>How do we trust AI agents to make purchases on our behalf?</em></strong></blockquote><p>Think about it.</p><p>Today’s payment systems assume one fundamental thing: <strong>a human is directly clicking “buy” on a trusted website</strong>. Everything from fraud detection to liability rules is built on that assumption.</p><p>But what happens when an AI agent initiates the payment instead?</p><h4>Three Critical Questions Emerge:</h4><p><strong>Authorization</strong><br>How do we verify that you actually gave an agent permission to make a specific purchase?</p><p><strong>Authenticity</strong><br>How can merchants be sure the agent’s request truly reflects your intent and isn’t an AI hallucination or error?</p><p><strong>Accountability</strong><br>If something goes wrong — a fraudulent transaction or a mistake — who’s responsible? You? The AI developer? The merchant? The payment processor?</p><p>Without solving these questions, we’d end up with a fragmented mess: every company building its own proprietary agent payment system. Confusing for users, expensive for merchants, impossible for regulators to oversee.</p><p><strong>AP2 solves this by creating a common language</strong> — an open, interoperable protocol that any AI agent can use to transact securely with any merchant, anywhere.</p><p>AP2 isn’t limited to traditional payment methods. It also supports cryptocurrency and stablecoin payments through the <strong>A2A x402 extension</strong>, developed in collaboration with Coinbase, Ethereum Foundation, and MetaMask.</p><h4>The Core Innovation: Verifiable Credentials and Mandate Chains</h4><p>At the heart of AP2 is a brilliantly simple idea: <strong>Mandates</strong>.</p><p>A mandate is a cryptographically signed digital contract that provides tamper-proof evidence of what you authorized and when. These mandates are expressed as <strong>W3C Verifiable Credentials</strong>, ensuring tamper resistance, portability, and interoperability across the entire payments ecosystem.</p><p>Think of it like a digital receipt that can’t be forged, can’t be altered, and can prove exactly what happened at every step.</p><h3>AP2 Defines Three Types of Mandates:</h3><h4>1. Intent Mandate — Captures Your Initial Request</h4><p>This is the foundation of every AP2 transaction. Whether you’re shopping in real-time or delegating a future purchase, your initial request is always captured in an Intent Mandate.</p><p><strong>When you’re actively shopping (Human-Present):</strong></p><p>You ask your agent: <em>“Find me Airpods Pro under $250”</em></p><ul><li>An Intent Mandate is created capturing this general request</li><li>This provides auditable context for the entire interaction</li><li>You’ll review options and approve the final cart</li></ul><p><strong>When you’re delegating a task (Human-NOT-Present):</strong></p><p>You tell your agent: <em>“Buy AirPods Pro if the price drops below $180 AND 2-day delivery is available”</em></p><ul><li>An Intent Mandate is created with detailed conditions</li><li>You cryptographically sign it upfront, giving the agent authority to act autonomously</li><li>The agent monitors continuously and acts when your conditions are met</li></ul><p><strong>The key difference:</strong></p><ul><li><strong>Human-Present Intent:</strong> General request, you stay involved to approve the final cart</li><li><strong>Human-NOT-Present Intent:</strong> Detailed conditions, agent acts autonomously within constraints</li></ul><p>The Intent Mandate provides the auditable foundation — verifiable proof of what you originally asked for.</p><h4>2. Cart Mandate — Locks In The Final Purchase</h4><p>The Cart Mandate captures the exact items, quantities, and prices for a transaction. This is the critical moment that ensures “what you see is what you pay for.”</p><p><strong>When you’re actively shopping (Human-Present):</strong></p><p>After your Intent Mandate captures your request (“<em>find me Airpods Pro under $250</em>”), the agent searches and presents options. You review them and pick one. <strong>The merchant signs the Cart Mandate first (guaranteeing fulfillment), then</strong> when you approve, your signature on the Cart Mandate creates undeniable proof that you authorized that exact cart — these specific airpods, at this exact price, from this merchant.</p><p><strong>When you’ve delegated a task (Human-NOT-Present):</strong></p><p>Your Intent Mandate specified conditions (“<em>buy AirPods Pro if price &lt; $180 with 2-day delivery</em>”). When the agent finds a match, it automatically generates a Cart Mandate on your behalf based on your pre-authorized Intent Mandate. <strong>The merchant signs this Cart Mandate, and the</strong> agent’s <strong>Cart Mandate</strong> references your original Intent, creating a verifiable chain of authorization.</p><p><strong>The Cart Mandate is:</strong></p><ul><li><strong>Merchant-signed</strong> (guaranteeing they’ll fulfill at that exact price)</li><li><strong>User-approved</strong> (cryptographically proving you authorized this specific purchase)</li><li><strong>Unchangeable once signed</strong> (creating a secure record for dispute resolution)</li></ul><h4>3. Payment Mandate — Signals Agent Involvement to Financial Networks</h4><p>This is what gets shared with payment networks and banks.</p><p>The Payment Mandate is a minimal credential derived from your Cart or Intent Mandate. It provides issuers and networks with critical visibility into:</p><ul><li><strong>Agent presence</strong> (was an AI agent involved in this transaction?)</li><li><strong>Transaction modality</strong> (human-present vs human-not-present)</li><li><strong>Verifiable link</strong> back to original user intent</li></ul><p>This gives financial institutions the context they need to assess risk properly and apply appropriate fraud detection rules — without exposing your complete mandate chain, maintaining your privacy while enabling proper risk management.</p><h4>The Mandate Chain:</h4><p>The power of AP2 lies in how these mandates link together to create an unbreakable chain of evidence:</p><pre><strong>Human-Present Flow (Real-Time Shopping):</strong><br><br>Step 1: User: &quot;Find me AirPods Pro under $250&quot;<br>↓<br>[Intent Mandate — Created by agent to capture user&#39;s request for audit context]<br>↓<br>Step 2: Shopping Agent delegates to Merchant Agent<br>↓<br>Step 3: Merchant builds cart and presents options<br>↓<br>Step 4: User reviews and approves: &quot;Yes, I want those for $245&quot;<br>↓<br>[Cart Mandate — USER signs this, confirming exact purchase]<br>↓<br>[Payment Mandate — Derived/created, sent to payment network with agent-presence signals]<br>↓<br>Step 5: Payment executes<br>↓<br>Transaction Complete ✓<br><br><br><strong>Human-not-present Flow(Delegated Shopping):</strong><br><br>Step 1: User: &quot;Buy AirPods Pro if price drops below $180 AND 2-day delivery available&quot;<br>↓<br>[Intent Mandate — Detailed conditions, USER signs this upfront]<br>↓<br>Step 2: Agent monitors continuously (you&#39;re asleep/at work/living your life)<br>↓<br>Step 3: Conditions met! AirPods Pro found: $175 with 2-day delivery<br>↓<br>[Cart Mandate — Agent auto-generates this, referencing your signed Intent Mandate]<br>↓<br>[Payment Mandate — Created and sent to payment network]<br>↓<br>Step 4: Payment executes autonomously<br>↓<br>Transaction Complete ✓<br><br>You wake up to: <strong><em>&quot;AirPods Pro purchased for $175. 2-day delivery confirmed.&quot;</em></strong></pre><p>Every step is cryptographically signed. You can trace the entire decision path from your original authorization to the final purchase.</p><p>Nothing can be hidden. Nothing can be faked.</p><p><strong>This complete sequence — from intent, to cart, to payment — creates a non-repudiable audit trail that answers the critical questions of authorization, authenticity, and accountability.</strong></p><h3>Why This Matters: Real-World Examples</h3><p>Let me show you what AP2 actually enables:</p><h4>Smarter Shopping: Never Lose a Sale</h4><p>A customer finds the perfect winter jacket online — but it’s sold out in green, their preferred color.</p><p>Normally, that’s the end of the story. The sale is lost.</p><p><strong>With AP2,</strong> they tell their agent:</p><blockquote>“I really want this jacket in green, size M. I’m willing to pay up to 20% more for it.”</blockquote><p>They sign an Intent Mandate and forget about it.</p><p>The agent monitors availability across multiple merchants continuously. The moment that specific variant appears anywhere — whether it’s back in stock at the original store or available at another retailer — the agent executes the purchase automatically with cryptographic proof of authorization.</p><p><strong>Result:</strong> Customer gets exactly what they wanted. Merchant captures a sale that would have been completely lost. Everyone wins.</p><p><strong>This isn’t possible today.</strong> Current payment systems can’t handle this level of autonomous, cross-merchant monitoring and purchasing securely.</p><h3>My Challenge: I Wanted to Give AP2 a Try</h3><p>When I read about AP2, I was fascinated by the concept.</p><p>An open protocol for autonomous AI commerce that works across any platform? That sounded incredible — but <strong>did it actually work?</strong></p><p>I’m a developer who loves experimenting with new technologies. The <strong>AWS AI Global Hackathon</strong> was the perfect opportunity to dive in and give AP2 a real try.</p><p><strong>My goal was simple:</strong> Build a working implementation and see if AP2 delivers on its promise of interoperability.</p><p>Could I implement this protocol using completely different tools than the examples I’d seen? Would the mandate chains actually work? Could I get autonomous purchasing working end-to-end?</p><p>There was only one way to find out: <strong>build it and see.</strong></p><h3>My Goal: Prove Interoperability</h3><p>Prove that AP2 achieves its promise of interoperability by building a production-quality implementation that:</p><ul><li>Uses a completely different AI framework</li><li>Uses a completely different LLM provider</li><li>Deploys on entirely different infrastructure</li><li>But implements the exact same AP2 protocol</li></ul><p><strong>If it works, AP2 is truly open. If it doesn’t, we’ve learned something important about the protocol’s limitations.</strong></p><h3>What I Built: A Full Autonomous Shopping Platform</h3><p>I created <strong>Strands AP2 Payment Agent</strong> — a complete, working implementation of the Agent Payments Protocol that supports both shopping modes:</p><h4>Mode 1: Human-Present Shopping</h4><p>Conversational, real-time shopping where you’re actively engaged:</p><ul><li>Natural language product search</li><li><strong>Intent Mandate captures your initial search request</strong></li><li>Agent presents real-time results</li><li>You review and approve specific items</li><li>Your approval creates a signed Cart Mandate</li><li>Complete purchase with full cryptographic audit trail</li></ul><h4>Mode 2: Human-NOT-Present Shopping</h4><p><strong>This is the revolutionary part</strong> — fully autonomous shopping:</p><ul><li>You set detailed constraints once: <em>“Buy X if price &lt; Y and delivery ≤ Z”</em></li><li>You sign an Intent Mandate upfront (your authorization)</li><li>You walk away completely (sleep, work, live your life)</li><li>Agent monitors continuously (every [X] duration)</li><li>When conditions match exactly, agent autonomously generates Cart Mandate</li><li>Agent purchases on your behalf</li><li>You wake up to a notification with complete cryptographic proof</li></ul><h4>The Architecture: How It All Works Together</h4><p>I built a multi-agent system where specialized agents handle different tasks:</p><ul><li><strong>Supervisor Agent</strong> — Routes user requests to the right specialist</li><li><strong>Shopping Agent</strong> — Handles immediate purchases with Intent → Cart flow</li><li><strong>Delegate Agent</strong> — Sets up autonomous monitoring jobs with detailed Intent Mandates</li><li><strong>Payment Agent</strong> — Validates mandate chains and processes payments (completely reusable!)</li></ul><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*-AFSfyvSsrFzd8EAcj5B5A.jpeg" /></figure><h4>Key Implementation Details:</h4><ul><li>Complete AP2 mandate chain (Intent → Cart → Payment → Transaction)</li><li>HMAC-SHA256 cryptographic signatures (I used HMAC-SHA256 for this proof-of-concept; the AP2 specification supports multiple W3C Verifiable Credential signature schemes including ECDSA, EdDSA, and RSA for production use)</li><li>W3C Verifiable Credentials format (ensuring tamper resistance, portability, and interoperability across the entire payments ecosystem)</li><li>Persistent monitoring jobs that survive server restarts</li><li>Real-time streaming of agent decisions</li><li>Mock merchant/payment processor APIs (since AP2 is so new, no real merchants support it yet)</li></ul><h4>AP2’s Role-Based Architecture</h4><p>The protocol defines specific roles to ensure security and PCI compliance:</p><ul><li><strong>User</strong> — Originates intent and authorizes purchases</li><li><strong>Agent</strong> — Executes tasks on behalf of the user</li><li><strong>Credential Provider</strong> — Secures payment methods and manages authentication</li><li><strong>Merchant Endpoint/Processor</strong> — Receives mandates and executes settlement</li><li><strong>Issuer/Network</strong> — Authorizes transactions using standard risk models</li></ul><p><strong>This role separation is critical:</strong> it confines PCI-sensitive data to credential providers, limits merchant and agent exposure to payment credentials, and enforces clear accountability across the transaction lifecycle. My implementation respects these boundaries.</p><h4>The Tech Stack I Used:</h4><p>Instead of Google’s stack, I used:</p><ul><li><strong>AWS Bedrock with Claude Sonnet 4.5</strong> (instead of Gemini)</li><li><strong>Strands SDK</strong> for agent orchestration (instead of ADK)</li><li><strong>AWS infrastructure</strong> for deployment</li><li><strong>Python FastAPI</strong> backend</li><li><strong>React frontend</strong> with real-time mandate visualizations</li></ul><p>Built for the AWS AI Global Hackathon, but the principles apply anywhere.</p><p><strong>This proves AP2’s interoperability promise:</strong> completely different tools, same protocol, fully functional.</p><h3>The Real Challenges: What Nobody Tells You About Building With AP2</h3><h4>Challenge 1: AP2 Is Brand New</h4><p>AP2 v0.1 was announced in September 2025. That means:</p><ul><li>No real merchant integrations yet (Amazon, Shopify don’t support it)</li><li>No payment processor SDKs (Stripe, PayPal have no AP2 libraries)</li><li>No Stack Overflow answers</li><li>No production examples beyond Google’s reference</li></ul><p><strong>I had to:</strong></p><ul><li>Interpret the official specification from scratch</li><li>Build complete mock merchant APIs</li><li>Make architectural decisions without real-world examples</li><li>Design for future compatibility when real integrations arrive</li></ul><p><strong>Learning:</strong> The AP2 specification (https://ap2-protocol.org/specification/) is actually quite well-documented. The official docs at ap2-protocol.org provided enough detail to implement the core concepts correctly. The hardest part was imagining how real-world integrations would work.</p><h4>Challenge 2: Getting Multi-Agent Coordination Right</h4><p>Having multiple specialized agents pass context between each other while maintaining conversation flow was complex.</p><p><strong>Example problem:</strong> User says “yes” or “the first one” but the agent receiving that message doesn’t have the previous conversation context.</p><p><strong>Solution:</strong> Proper session management and contextual translation. The Supervisor Agent reformulates ambiguous inputs into concrete instructions that specialist agents can understand.</p><h4>Challenge 3: Mandate Chain Validation</h4><p>The Payment Agent had to validate complete mandate chains with proper Intent → Cart relationships:</p><p><strong>Core Validations (Both Flows):</strong></p><ul><li>Cryptographic signature integrity (are signatures valid?)</li><li>Timestamp expiration (mandates can have TTLs)</li><li>Parent-child relationships (does Cart properly reference Intent?)</li><li>Constraint satisfaction (does the cart match the intent conditions?)</li></ul><p><strong>Human-Present Specific:</strong></p><ul><li>Verify Intent Mandate captures the original user request for audit context</li><li>Confirm user explicitly approved the Cart (user signature present)</li></ul><p><strong>Human-NOT-Present Specific:</strong></p><ul><li>Verify Intent Mandate contains detailed, specific conditions signed by user</li><li>Validate all conditions in Intent were actually met before Cart was generated</li><li>Ensure agent had authority to act autonomously</li></ul><p><strong>Learning:</strong> Building robust validation logic with clear error codes was crucial. Every failure mode needs a specific error code (as defined in AP2 spec) so the system can handle edge cases gracefully. The Intent → Cart → Payment chain must be unbroken and verifiable at every step.</p><h3>What I Learned: Five Key Insights</h3><h4>1. AP2 Is Genuinely Interoperable</h4><p>This is the big one. <strong>My implementation works.</strong></p><p>I used completely different tools than Google’s reference implementation, and the protocol still functions correctly. This validates AP2’s core promise: it’s truly open and vendor-agnostic.</p><p>Any developer can implement AP2 with their preferred stack. You’re not locked into Google’s ecosystem. This is exactly what an open protocol should be.</p><h4>2. The Mandate Chain Architecture Is Brilliant</h4><p>Having cryptographic proof at every step isn’t just security theater — it fundamentally solves the trust problem.</p><p>When you can trace every decision from intent to transaction with mathematical certainty, accountability becomes clear.</p><p>This is what makes autonomous commerce possible. Without it, you couldn’t trust agents to act on your behalf.</p><h4>3. Real-Time Transparency Builds Trust</h4><p>Streaming every agent decision in real-time (what it’s thinking, which tools it’s calling, why it made a choice) creates transparency that users need to trust autonomous systems.</p><p>People are much more comfortable with automation when they can see how decisions are made.</p><h4>4. We’re At The Very Beginning</h4><p>AP2 v0.1 is just the foundation. The protocol specification includes a clear roadmap:</p><p><strong>V1.x (Coming):</strong></p><ul><li>“Push” payment support (UPI, PIX, real-time bank transfers)</li><li>Recurring payments and subscriptions</li><li>Enhanced human-not-present flows</li><li>Better MCP integration</li></ul><p><strong>Long-term Vision:</strong></p><ul><li>Complex multi-merchant transactions</li><li>Real-time agent-to-agent negotiations</li><li>Full B2A (Business-to-Agent) commerce ecosystem</li></ul><p>We’re building the foundation of something much larger.</p><h4>5. The Ecosystem Needs To Catch Up</h4><p>The protocol is solid, but the ecosystem isn’t there yet:</p><ul><li>Merchants need to implement AP2 endpoints</li><li>Payment processors need AP2 SDKs and libraries</li><li>Risk engines need to understand agent-presence signals</li><li>Regulations need to evolve for autonomous transactions</li></ul><p>This will take time. But having an open protocol means everyone can move in the same direction.</p><h3>My Accomplishments: What I’m Proud Of</h3><h4>Proving Interoperability</h4><p>I successfully demonstrated that AP2 works beyond Google’s reference implementation. This is crucial for preventing vendor lock-in and ensuring the protocol serves the entire industry, not just one company.</p><h4>Building Truly Autonomous Purchases</h4><p>I didn’t just build an “AI shopping assistant” that helps you find things. I built a system where agents genuinely make purchase decisions autonomously while you’re completely absent — with full cryptographic accountability.</p><h4>Creating a Reusable Payment Agent</h4><p>The Payment Agent I built is completely domain-independent. It can be extracted and used in any e-commerce project that needs AP2 compliance. It’s a building block others can use.</p><h4>Contributing to AP2 Adoption</h4><p>As one of the earliest implementations outside Google, I’m helping validate the specification for real-world use. My code demonstrates how the protocol handles both synchronous and autonomous payment scenarios with different agent frameworks.</p><h4>Open-Sourcing Everything</h4><p>The entire project is on GitHub: <a href="https://github.com/harithaharry001/strands-ap2-payment-agent">https://github.com/harithaharry001/strands-ap2-payment-agent</a></p><p>Anyone can see exactly how AP2 works, learn from the implementation, and build their own agents.</p><h3>The Bigger Picture: Why Open Protocols Matter</h3><p>Imagine if every website used a different, proprietary version of HTTP. The web wouldn’t exist as we know it.</p><p><strong>Open protocols create interoperability, which creates network effects, which creates value for everyone.</strong></p><p>AP2 is trying to do for agent commerce what HTTP did for the web: provide a common foundation that everyone can build on.</p><p>If we get this right, the benefits are enormous:</p><h4>For Users</h4><ul><li>Consistent experience across all merchants</li><li>Your agent works everywhere</li><li>Clear accountability when things go wrong</li><li>True automation that saves time</li></ul><h4>For Merchants</h4><ul><li>One integration works with all agents</li><li>Lower development costs</li><li>Access to agent-driven customers</li><li>Future-proof architecture</li></ul><h4>For Developers</h4><ul><li>Build once, work everywhere</li><li>No vendor lock-in</li><li>Clear specifications</li><li>Growing ecosystem of tools</li></ul><h4>For The Economy</h4><ul><li>Reduced friction in commerce</li><li>New business models become possible</li><li>Increased efficiency</li><li>Global standards prevent fragmentation</li></ul><h3>What’s Next: The Road Ahead</h3><h4>For the Protocol</h4><p>AP2 will continue evolving through an open, collaborative process. Google has committed to working with standards bodies and the 60+ partner organizations to refine and extend the protocol based on real-world feedback.</p><h4>Trust Model Evolution:</h4><p>AP2 uses cryptographically signed mandates and curated allow-lists. The long-term vision includes <strong>real-time, decentralized identity assurance</strong> using HTTPS, DNS ownership, and mTLS, enabling <strong>more automated trust establishment</strong> between parties without manual registry curation.</p><h4>For Real-World Adoption</h4><p>We need:</p><ul><li>Major merchants to implement AP2 endpoints (Amazon, Shopify, etc.)</li><li>Payment processors to release AP2 SDKs (Stripe, PayPal, etc.)</li><li>AI platforms to integrate AP2 into their agent frameworks</li><li>Regulatory clarity on autonomous transactions</li></ul><h4>For My Implementation</h4><p>I’m planning to add:</p><ul><li>Production-grade cryptography (ECDSA, EdDSA, or RSA per W3C VC standards, with hardware-backed keys)</li><li>Integration with real merchant APIs when available</li><li>Multi-payment method support</li><li>Enhanced autonomous behaviors (multi-merchant negotiation, automatic coupon application)</li><li>Predictive intent suggestions based on purchase history</li></ul><h4>Try It Yourself</h4><p>The entire project is open source on GitHub:<br> <a href="https://github.com/harithaharry001/strands-ap2-payment-agent"><strong>https://github.com/harithaharry001/strands-ap2-payment-agent</strong></a></p><p>You’ll find:</p><ul><li>Complete multi-agent implementation</li><li>Full AP2 mandate chain validation</li><li>Real-time streaming architecture</li><li>Frontend visualizations</li><li>Mock merchant/payment processor implementations</li><li>Setup instructions and documentation</li></ul><p>Whether you’re a developer curious about AI agents, a researcher studying payment protocols, or a business exploring autonomous commerce — dive in and see how it works.</p><h3>Final Thoughts: We’re Building The Future</h3><p>The technology for autonomous AI agents exists today. The Agent Payments Protocol provides the trust foundation they need to transact securely. The infrastructure is ready. The ecosystem is forming.</p><p>What’s needed now is adoption and experimentation. Developers building implementations. Merchants adding support. Payment processors creating tools. Regulators providing clarity.</p><p><strong>The future of commerce isn’t humans clicking “buy” buttons. It’s humans setting intentions and AI agents executing them — securely, transparently, and with complete accountability.</strong></p><p>I gave AP2 a try, and I can confirm: <strong>this actually works</strong>. The protocol is solid. The vision is achievable. The interoperability is real.</p><p>Now it’s time for the rest of the industry to build on this foundation.</p><blockquote><strong><em>The question isn’t whether autonomous AI commerce will happen. It’s how fast we can get there together.</em></strong></blockquote><h3>Let’s Connect</h3><p>Have questions about AP2 or my implementation? Curious about autonomous agents? Want to discuss the future of commerce?</p><p><strong>I’d love to hear your thoughts:</strong></p><ul><li>What use cases excite you most?</li><li>What concerns do you have about autonomous AI agents?</li><li>What would you want your agent to buy for you?</li></ul><p>Drop a comment or connect with me on <strong>LinkedIn</strong>. Let’s build the future of commerce together.</p><p>Check out the project on <strong>GitHub</strong> and let me know what you think!</p><h3>Useful Resources</h3><ul><li><strong>Official AP2 Documentation:</strong> <a href="https://ap2-protocol.org/">https://ap2-protocol.org/</a></li><li><strong>AP2 GitHub Repository:</strong> <a href="https://github.com/google-agentic-commerce/AP2">https://github.com/google-agentic-commerce/AP2</a></li><li><strong>My Implementation:</strong> <a href="https://github.com/harithaharry001/strands-ap2-payment-agent">https://github.com/harithaharry001/strands-ap2-payment-agent</a></li><li><strong>Google Cloud AP2 Announcement:</strong> <a href="https://cloud.google.com/blog/products/ai-machine-learning/announcing-agents-to-payments-ap2-protocol">https://cloud.google.com/blog/products/ai-machine-learning/announcing-agents-to-payments-ap2-protocol</a></li><li><strong>GitHub for A2A x402:</strong> <a href="https://github.com/google-agentic-commerce/a2a-x402">https://github.com/google-agentic-commerce/a2a-x402</a></li></ul><p><strong>Tags:</strong> #AP2 #AgentPaymentsProtocol #AI #AutonomousCommerce #AgenticAI #OpenProtocol #MachineLearning #Ecommerce #Innovation #Hackathon #TechBlog #SoftwareEngineering #Interoperability #CryptographicSecurity #VerifiableCredentials</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=fcbe009d1b86" width="1" height="1" alt="">]]></content:encoded>
        </item>
    </channel>
</rss>