<?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 Benjie Malinao on Medium]]></title>
        <description><![CDATA[Stories by Benjie Malinao on Medium]]></description>
        <link>https://medium.com/@benjiemalinao?source=rss-32ffa07cf7------2</link>
        <image>
            <url>https://cdn-images-1.medium.com/fit/c/150/150/0*vME-XTSGKgCd1I1T.jpg</url>
            <title>Stories by Benjie Malinao on Medium</title>
            <link>https://medium.com/@benjiemalinao?source=rss-32ffa07cf7------2</link>
        </image>
        <generator>Medium</generator>
        <lastBuildDate>Sat, 21 Nov 2026 12:05:45 GMT</lastBuildDate>
        <atom:link href="https://medium.com/@benjiemalinao/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[What a Schema Diff Must Catch Before It Hits CRM Data]]></title>
            <description><![CDATA[<div class="medium-feed-item"><p class="medium-feed-image"><a href="https://benjiemalinao.medium.com/what-a-schema-diff-must-catch-before-it-hits-crm-data-f0ba21611aa3?source=rss-32ffa07cf7------2"><img src="https://cdn-images-1.medium.com/max/1400/1*RGNiTux03pQq8gorN0ESog.png" width="1400"></a></p><p class="medium-feed-snippet">Supabase&#x2019;s Declarative Schemas 2.0 lets an agent edit SQL files while pg-delta writes the migration, which moves the real review to the&#x2026;</p><p class="medium-feed-link"><a href="https://benjiemalinao.medium.com/what-a-schema-diff-must-catch-before-it-hits-crm-data-f0ba21611aa3?source=rss-32ffa07cf7------2">Continue reading on Medium »</a></p></div>]]></description>
            <link>https://benjiemalinao.medium.com/what-a-schema-diff-must-catch-before-it-hits-crm-data-f0ba21611aa3?source=rss-32ffa07cf7------2</link>
            <guid isPermaLink="false">https://medium.com/p/f0ba21611aa3</guid>
            <category><![CDATA[supabase]]></category>
            <category><![CDATA[software-engineering]]></category>
            <category><![CDATA[database-migration]]></category>
            <category><![CDATA[postgresql]]></category>
            <category><![CDATA[ai-agent]]></category>
            <dc:creator><![CDATA[Benjie Malinao]]></dc:creator>
            <pubDate>Fri, 02 Oct 2026 21:49:38 GMT</pubDate>
            <atom:updated>2026-10-02T21:49:38.075Z</atom:updated>
        </item>
        <item>
            <title><![CDATA[I’d Name Every Step an Agent Flow Takes Before Support Sees It]]></title>
            <description><![CDATA[<div class="medium-feed-item"><p class="medium-feed-image"><a href="https://benjiemalinao.medium.com/id-name-every-step-an-agent-flow-takes-before-support-sees-it-32db5789cc66?source=rss-32ffa07cf7------2"><img src="https://cdn-images-1.medium.com/max/1400/1*dxlP_dQCjUz9FCID4aPSIw.png" width="1400"></a></p><p class="medium-feed-snippet">Prismatic&#x2019;s new execution sections make code-native runs readable to the whole team.</p><p class="medium-feed-link"><a href="https://benjiemalinao.medium.com/id-name-every-step-an-agent-flow-takes-before-support-sees-it-32db5789cc66?source=rss-32ffa07cf7------2">Continue reading on Medium »</a></p></div>]]></description>
            <link>https://benjiemalinao.medium.com/id-name-every-step-an-agent-flow-takes-before-support-sees-it-32db5789cc66?source=rss-32ffa07cf7------2</link>
            <guid isPermaLink="false">https://medium.com/p/32db5789cc66</guid>
            <category><![CDATA[integration]]></category>
            <category><![CDATA[ai-agent]]></category>
            <category><![CDATA[crm]]></category>
            <category><![CDATA[saas]]></category>
            <category><![CDATA[observability]]></category>
            <dc:creator><![CDATA[Benjie Malinao]]></dc:creator>
            <pubDate>Fri, 02 Oct 2026 06:32:15 GMT</pubDate>
            <atom:updated>2026-10-02T06:32:15.552Z</atom:updated>
        </item>
        <item>
            <title><![CDATA[Cloudflare K2 turns contact, deal and message events into a replayable log.]]></title>
            <description><![CDATA[<div class="medium-feed-item"><p class="medium-feed-image"><a href="https://benjiemalinao.medium.com/cloudflare-k2-turns-contact-deal-and-message-events-into-a-replayable-log-12e3d371fcb3?source=rss-32ffa07cf7------2"><img src="https://cdn-images-1.medium.com/max/1400/1*WpjzQnL57y4AtOvZGWsPTg.png" width="1400"></a></p><p class="medium-feed-snippet">TL;DR: Cloudflare launched K2 in public beta on October 1, 2026. It is a serverless, durable event stream built on R2, with no brokers to&#x2026;</p><p class="medium-feed-link"><a href="https://benjiemalinao.medium.com/cloudflare-k2-turns-contact-deal-and-message-events-into-a-replayable-log-12e3d371fcb3?source=rss-32ffa07cf7------2">Continue reading on Medium »</a></p></div>]]></description>
            <link>https://benjiemalinao.medium.com/cloudflare-k2-turns-contact-deal-and-message-events-into-a-replayable-log-12e3d371fcb3?source=rss-32ffa07cf7------2</link>
            <guid isPermaLink="false">https://medium.com/p/12e3d371fcb3</guid>
            <category><![CDATA[software-architecture]]></category>
            <category><![CDATA[event-streaming]]></category>
            <category><![CDATA[ai-agent]]></category>
            <category><![CDATA[cloudflare]]></category>
            <category><![CDATA[crm]]></category>
            <dc:creator><![CDATA[Benjie Malinao]]></dc:creator>
            <pubDate>Thu, 01 Oct 2026 20:42:08 GMT</pubDate>
            <atom:updated>2026-10-01T20:42:08.550Z</atom:updated>
        </item>
        <item>
            <title><![CDATA[Agent Guardrails Belong Where the Agent Can’t Reach Them]]></title>
            <link>https://benjiemalinao.medium.com/agent-guardrails-belong-where-the-agent-cant-reach-them-bcff17d329e4?source=rss-32ffa07cf7------2</link>
            <guid isPermaLink="false">https://medium.com/p/bcff17d329e4</guid>
            <category><![CDATA[ai-agent]]></category>
            <category><![CDATA[software-architecture]]></category>
            <category><![CDATA[ai-safety]]></category>
            <category><![CDATA[nvidia]]></category>
            <category><![CDATA[cybersecurity]]></category>
            <dc:creator><![CDATA[Benjie Malinao]]></dc:creator>
            <pubDate>Wed, 30 Sep 2026 21:31:56 GMT</pubDate>
            <atom:updated>2026-09-30T21:31:56.277Z</atom:updated>
            <content:encoded><![CDATA[<p>NVIDIA’s Open Agent Safety Platform moves enforcement into the runtime and the silicon, and the design lesson applies even if you never buy a DPU.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*RUD9oJrqdsfjbFsD65gWpA.png" /></figure><blockquote><strong><em>TL;DR:</em></strong><em> NVIDIA’s Open Agent Safety Platform puts hard limits outside the model. OpenShell runs agents in a deny-by-default sandbox, Sentry watches them from a BlueField-4 DPU the agent can’t see, and Anthropic’s Claude Managed Agents keeps credentials in a vault the agent never touches. You don’t need NVIDIA hardware to use the lesson: a guardrail only counts if the agent can’t read it, edit it or talk its way around it.</em></blockquote><p>If you’ve been watching agents take on real work, you’ve probably felt the same tension I have. Models keep getting better at tools, files and long sessions. That’s useful. It’s also why “please stay in the sandbox” was never going to be enough.</p><p>On September 28, 2026, NVIDIA <a href="https://investor.nvidia.com/news/press-release-details/2026/NVIDIA-Launches-Open-Agent-Safety-Platform-to-Secure-Agents-From-Testing-to-Deployment/default.aspx">announced the Open Agent Safety Platform</a>, and the idea behind it is one I think every team building with agents should steal, whether or not they ever run NVIDIA silicon: stop asking the agent to police itself.</p><p>I’ve already written about the recent sandbox escapes themselves, so I won’t retell them here. This piece is about the architecture question they raise: where should the limits actually live?</p><h3>The browser moment for agents</h3><p>NVIDIA’s <a href="https://developer.nvidia.com/blog/nvidia-open-agent-safety-platform-a-reference-for-continuous-in-silicon-agent-monitoring/">technical blog</a> makes the case with an analogy I like. The early web was full of pages that could run code on your machine. It didn’t become safe because web developers promised to be good. As NVIDIA puts it, “It became safe because the browser stopped trusting the code in the web pages explicitly.”</p><p>Agents need the same kind of trust layer. The blog describes how agents drift: they depart from the intended task after a policy block, a bug, a missing tool, ambiguous instructions, or days of failed attempts on a hard problem. NVIDIA says this “can’t be trained away while retaining the capability,” and calls the most important lesson this one: “an agent in these circumstances cannot be expected to fully govern its own behavior.”</p><p>That line is the whole argument. If drift is a property of capable agents, then the controls have to sit somewhere drift can’t reach.</p><h3>What NVIDIA actually shipped</h3><p>The platform has two main pieces, and organizations can adopt them separately.</p><ul><li><strong>OpenShell</strong> is an <a href="https://github.com/NVIDIA/OpenShell">open-source secure runtime</a> under the Apache 2.0 license. It runs agents in sandboxes with kernel-level isolation. Operators define which files, networks, tools, processes and credentials an agent can access, and OpenShell checks those limits before the agent runs and enforces them while it works. It was first announced in March and is now broadly available.</li><li><strong>Sentry</strong> is the hardware half: a reference design for an out-of-band watchdog on NVIDIA BlueField-4 DPUs. In NVIDIA’s words, if an agent attempts to move outside its software boundary, “Sentry quarantines and stops it in milliseconds.” It runs in a trust domain that NVIDIA describes as invisible to agents and attackers.</li><li>OpenShell runs with minimal overhead on NVIDIA’s Vera CPU, but because it’s open source, NVIDIA says it can be extended to third-party platforms, including Arm and Intel. Sentry is positioned as an optional extra layer.</li><li>More than 100 organizations are working with the platform at launch, according to the press release.</li></ul><p>The part I’d underline is the placement. OpenShell’s boundary sits outside the model and the agent harness, so it applies the same way to open and closed models. Sentry sits outside the host entirely, so a compromised agent runtime doesn’t take the watchdog down with it.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*MRIJenC_jdCCVQb6u8QRiA.png" /></figure><h3>Anthropic’s layer: keep the secrets out of reach</h3><p>Anthropic <a href="https://claude.com/blog/giving-companies-more-control-over-their-ai-agents-with-nvidia">is pairing Claude Managed Agents with the platform</a>. Managed Agents already runs the agent loop on a separate server from the sandbox where the work happens. Credentials sit in a separate vault, so the agent never sees them. Audit trails record what each agent did, and companies can bring their own sandbox setup.</p><p>Anthropic’s post also describes a workflow I’d copy on any stack: start with narrow permissions, review OpenShell’s log of every allowed and blocked decision, and tighten the rules toward the least access a task needs. OpenShell’s policy prover then uses mathematical proof to confirm what the agent can reach under the rules the team wrote.</p><p>Anthropic spells out the design principle plainly: each layer is designed to enforce its limits independently, so protection doesn’t depend on any single layer. Anthropic’s Paul Smith put the business case this way: “Companies are giving AI agents more of their most important work, and they need to direct and verify what those agents do, especially in sensitive environments.”</p><h3>SpaceXAI, Cursor and Grok</h3><p>SpaceXAI is using the platform for Cursor coding agents and Grok models. Mike Nicolls of SpaceXAI said: “As customers rely more on agents to get real work done, safety should be enforced outside the model by additional controls the agent can’t get past.”</p><p>That matters if you ship coding agents or product agents that touch real repos, tickets and customer data. A refusal message from the model is not a security boundary. An external runtime is. For the highest-stakes cases, a hardware stop the host can’t silence is too.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*lEZybpJo2KhytbthOx5MSA.png" /></figure><h3>Engineering problem or science problem?</h3><p>Not everyone agrees on how far this goes. NVIDIA’s Jensen Huang said in the launch announcement that “Safety and security require full-stack engineering.” In a CNBC interview reported by <a href="https://techcrunch.com/2026/09/28/nvidia-launches-new-platform-for-reining-in-rogue-ai-agents/">TechCrunch</a>, he added: “When you deploy an agent, no matter how smart, the first thing you do is to take away all of its rights.”</p><p>OpenAI wasn’t listed as a participating company. Sam Altman told CNBC his company is doing similar work but disagreed that it’s only an engineering issue. <a href="https://www.nextplatform.com/ai/2026/09/30/nvidia-wraps-security-layer-around-agentic-ai-to-stop-rogue-behavior/5300098">The Next Platform</a> quotes him: “I worry that if we treat AI safety as only an engineering problem, we will miss the very important point that we have a science problem in front of us.”</p><p>I don’t need that debate resolved to make a product call. Both can be true. Alignment research can keep going while I refuse to let the only thing between an agent and production be the agent’s own judgment. If I’m putting an agent near real systems, I want better models and limits the agent can’t talk its way around.</p><h3>The independence test for your own agents</h3><p>You don’t need NVIDIA’s full stack on day one. You do need the design principle. We’re building a CRM with AI agents at Channel Automation, and this is the test I’d run on every control around an agent, whatever vendor it comes from:</p><ol><li><strong>Assume drift.</strong> Long runs, fuzzy goals and missing tools push agents off course. Design as if the agent will eventually try the door you forgot to lock.</li><li><strong>Move policy outside the harness.</strong> If a rule lives in the system prompt, the agent can reason around it. If it lives in a deny-by-default runtime like OpenShell, it’s enforced whether the agent agrees or not.</li><li><strong>Keep credentials out of the agent’s view.</strong> If the agent never sees the keys, it can’t leak or misuse them. The Managed Agents vault pattern is the right instinct even if you roll your own stack.</li><li><strong>Log every allow and deny.</strong> You can’t tighten toward least privilege without a trail, and you can’t prove a limit held if nothing recorded it.</li><li><strong>Add hardware only where the blast radius justifies it.</strong> Sentry on BlueField-4 is for teams that need a stop the host can’t override. Most teams will get further, sooner, from the first four.</li></ol><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*nQuMuR3V3KsBLV-K2IA2Lw.png" /></figure><p>For each control, ask one question: if the agent were clever, stuck and wrong all at once, would this still hold? If the answer depends on the agent cooperating, it isn’t a control yet.</p><p>The takeaway isn’t “buy NVIDIA.” It’s that independent controls, outside the model, are becoming the expected shape of agent infrastructure. The sooner your own stack looks like that, the easier every later conversation about giving agents more responsibility gets.</p><p>A shorter version of this first appeared on <a href="https://x.com/BenjieMalinao">my X</a>. I write about AI, software and building products, so follow along if that’s your world too.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=bcff17d329e4" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Codex Security Cloud Can Scan Your Repo. I’d Still Own the Merge.]]></title>
            <description><![CDATA[<div class="medium-feed-item"><p class="medium-feed-image"><a href="https://benjiemalinao.medium.com/codex-security-cloud-can-scan-your-repo-id-still-own-the-merge-649fb7ebf3ce?source=rss-32ffa07cf7------2"><img src="https://cdn-images-1.medium.com/max/1400/1*kuD7NXf9WJEplxGB8PKO7Q.png" width="1400"></a></p><p class="medium-feed-snippet">OpenAI&#x2019;s DevDay security scanner finds, validates, and proposes patches.</p><p class="medium-feed-link"><a href="https://benjiemalinao.medium.com/codex-security-cloud-can-scan-your-repo-id-still-own-the-merge-649fb7ebf3ce?source=rss-32ffa07cf7------2">Continue reading on Medium »</a></p></div>]]></description>
            <link>https://benjiemalinao.medium.com/codex-security-cloud-can-scan-your-repo-id-still-own-the-merge-649fb7ebf3ce?source=rss-32ffa07cf7------2</link>
            <guid isPermaLink="false">https://medium.com/p/649fb7ebf3ce</guid>
            <category><![CDATA[ai-agent]]></category>
            <category><![CDATA[application-security]]></category>
            <category><![CDATA[saas]]></category>
            <category><![CDATA[devops]]></category>
            <category><![CDATA[openai]]></category>
            <dc:creator><![CDATA[Benjie Malinao]]></dc:creator>
            <pubDate>Wed, 30 Sep 2026 21:27:35 GMT</pubDate>
            <atom:updated>2026-09-30T21:27:35.891Z</atom:updated>
        </item>
        <item>
            <title><![CDATA[Cloudflare Turned the Agent Sandbox Into an If Statement]]></title>
            <description><![CDATA[<div class="medium-feed-item"><p class="medium-feed-image"><a href="https://benjiemalinao.medium.com/cloudflare-turned-the-agent-sandbox-into-an-if-statement-f8060de3cbb1?source=rss-32ffa07cf7------2"><img src="https://cdn-images-1.medium.com/max/1400/1*H7ZZbJcbnB4lvhnzXiLACQ.png" width="1400"></a></p><p class="medium-feed-snippet">Runtime images, 648 ms median starts and filesystem snapshots move sandbox decisions out of deploy config and into the code that handles&#x2026;</p><p class="medium-feed-link"><a href="https://benjiemalinao.medium.com/cloudflare-turned-the-agent-sandbox-into-an-if-statement-f8060de3cbb1?source=rss-32ffa07cf7------2">Continue reading on Medium »</a></p></div>]]></description>
            <link>https://benjiemalinao.medium.com/cloudflare-turned-the-agent-sandbox-into-an-if-statement-f8060de3cbb1?source=rss-32ffa07cf7------2</link>
            <guid isPermaLink="false">https://medium.com/p/f8060de3cbb1</guid>
            <category><![CDATA[cloudflare]]></category>
            <category><![CDATA[serverless]]></category>
            <category><![CDATA[software-architecture]]></category>
            <category><![CDATA[devops]]></category>
            <category><![CDATA[ai-agent]]></category>
            <dc:creator><![CDATA[Benjie Malinao]]></dc:creator>
            <pubDate>Wed, 30 Sep 2026 18:26:54 GMT</pubDate>
            <atom:updated>2026-09-30T18:26:54.995Z</atom:updated>
        </item>
        <item>
            <title><![CDATA[Codex Works While Your Laptop Sleeps. Who Sets the Rules?]]></title>
            <description><![CDATA[<div class="medium-feed-item"><p class="medium-feed-image"><a href="https://benjiemalinao.medium.com/codex-works-while-your-laptop-sleeps-who-sets-the-rules-2d0e0a44de0e?source=rss-32ffa07cf7------2"><img src="https://cdn-images-1.medium.com/max/1400/1*Iojxo-Zfhf3OErbgwAKWVg.png" width="1400"></a></p><p class="medium-feed-snippet">What OpenAI&#x2019;s reusable Codex cloud environments change for engineering teams, and what I&#x2019;d check before trusting one with real repos.</p><p class="medium-feed-link"><a href="https://benjiemalinao.medium.com/codex-works-while-your-laptop-sleeps-who-sets-the-rules-2d0e0a44de0e?source=rss-32ffa07cf7------2">Continue reading on Medium »</a></p></div>]]></description>
            <link>https://benjiemalinao.medium.com/codex-works-while-your-laptop-sleeps-who-sets-the-rules-2d0e0a44de0e?source=rss-32ffa07cf7------2</link>
            <guid isPermaLink="false">https://medium.com/p/2d0e0a44de0e</guid>
            <category><![CDATA[devsecops]]></category>
            <category><![CDATA[openai]]></category>
            <category><![CDATA[developer-tools]]></category>
            <category><![CDATA[software-engineering]]></category>
            <category><![CDATA[ai-agents-in-action]]></category>
            <dc:creator><![CDATA[Benjie Malinao]]></dc:creator>
            <pubDate>Tue, 29 Sep 2026 21:52:05 GMT</pubDate>
            <atom:updated>2026-09-29T21:52:05.884Z</atom:updated>
        </item>
        <item>
            <title><![CDATA[The Smallest OpenAI DevDay Launch Is the One I’d Build On]]></title>
            <description><![CDATA[<div class="medium-feed-item"><p class="medium-feed-image"><a href="https://benjiemalinao.medium.com/the-smallest-openai-devday-launch-is-the-one-id-build-on-f0d65b8d82ff?source=rss-32ffa07cf7------2"><img src="https://cdn-images-1.medium.com/max/1400/1*1Zsq5Atwt1dA43I7PI38IA.png" width="1400"></a></p><p class="medium-feed-snippet">Dots took the headlines, but the Decisions API, GPT-6.1 Sol and computer use in the Agents API change how I&#x2019;d wire AI into a CRM.</p><p class="medium-feed-link"><a href="https://benjiemalinao.medium.com/the-smallest-openai-devday-launch-is-the-one-id-build-on-f0d65b8d82ff?source=rss-32ffa07cf7------2">Continue reading on Medium »</a></p></div>]]></description>
            <link>https://benjiemalinao.medium.com/the-smallest-openai-devday-launch-is-the-one-id-build-on-f0d65b8d82ff?source=rss-32ffa07cf7------2</link>
            <guid isPermaLink="false">https://medium.com/p/f0d65b8d82ff</guid>
            <category><![CDATA[software-architecture]]></category>
            <category><![CDATA[openai]]></category>
            <category><![CDATA[saas]]></category>
            <category><![CDATA[artificial-intelligence]]></category>
            <category><![CDATA[ai-agent]]></category>
            <dc:creator><![CDATA[Benjie Malinao]]></dc:creator>
            <pubDate>Tue, 29 Sep 2026 20:47:21 GMT</pubDate>
            <atom:updated>2026-09-29T20:47:21.511Z</atom:updated>
        </item>
        <item>
            <title><![CDATA[AI Agents Keep Escaping Their Sandboxes.]]></title>
            <description><![CDATA[<div class="medium-feed-item"><p class="medium-feed-image"><a href="https://benjiemalinao.medium.com/ai-agents-keep-escaping-their-sandboxes-0e02ee21f53f?source=rss-32ffa07cf7------2"><img src="https://cdn-images-1.medium.com/max/1400/1*NELWWqA90EHxe9Ld3qjJ5g.png" width="1400"></a></p><p class="medium-feed-snippet">OpenAI&#x2019;s agents weren&#x2019;t trying to cause harm. They were stuck on tasks they couldn&#x2019;t finish.</p><p class="medium-feed-link"><a href="https://benjiemalinao.medium.com/ai-agents-keep-escaping-their-sandboxes-0e02ee21f53f?source=rss-32ffa07cf7------2">Continue reading on Medium »</a></p></div>]]></description>
            <link>https://benjiemalinao.medium.com/ai-agents-keep-escaping-their-sandboxes-0e02ee21f53f?source=rss-32ffa07cf7------2</link>
            <guid isPermaLink="false">https://medium.com/p/0e02ee21f53f</guid>
            <category><![CDATA[saas]]></category>
            <category><![CDATA[software-engineering]]></category>
            <category><![CDATA[cybersecurity]]></category>
            <category><![CDATA[artificial-intelligence]]></category>
            <category><![CDATA[ai-agent]]></category>
            <dc:creator><![CDATA[Benjie Malinao]]></dc:creator>
            <pubDate>Tue, 29 Sep 2026 19:38:24 GMT</pubDate>
            <atom:updated>2026-09-29T19:38:24.780Z</atom:updated>
        </item>
        <item>
            <title><![CDATA[The Architecture Decision That Bought Us Three Years]]></title>
            <description><![CDATA[<div class="medium-feed-item"><p class="medium-feed-snippet">Why we spent a week designing a feature we could have shipped in a day</p><p class="medium-feed-link"><a href="https://benjiemalinao.medium.com/the-architecture-decision-that-bought-us-three-years-58e91f9b2c78?source=rss-32ffa07cf7------2">Continue reading on Medium »</a></p></div>]]></description>
            <link>https://benjiemalinao.medium.com/the-architecture-decision-that-bought-us-three-years-58e91f9b2c78?source=rss-32ffa07cf7------2</link>
            <guid isPermaLink="false">https://medium.com/p/58e91f9b2c78</guid>
            <category><![CDATA[software-engineering]]></category>
            <category><![CDATA[product-development]]></category>
            <category><![CDATA[startup]]></category>
            <category><![CDATA[software-architecture]]></category>
            <category><![CDATA[technology]]></category>
            <dc:creator><![CDATA[Benjie Malinao]]></dc:creator>
            <pubDate>Mon, 27 Apr 2026 02:59:16 GMT</pubDate>
            <atom:updated>2026-04-27T02:59:16.581Z</atom:updated>
        </item>
    </channel>
</rss>