Rain, Agents and Open Source: My Week at Web Summit Lisbon Cisco developer advocate and CAIPE contributor presented a Web Summit Lisbon masterclass, "Democratising AI Platform Engineering with Open Source," to 111 attendees in a room built for 100, pitching the CNCF Sandbox project CAIPE as an open-source alternative to building or buying proprietary AI agent infrastructure. CAIPE bundles authentication, authorization, MCP tools, RAG on Neo4j, and workflows behind a single interface deployable via kind, Helm, or Docker Compose, with swappable components such as Keycloak and the user's LLM of choice. Web Summit's second day was canceled as Storm Charles forced an emergency state in Lisbon, and a live demo at the preceding AI Builders event failed mid-session because of an OpenAI API outage, prompting the speaker's advice to record demos. Every November, Lisbon turns into the capital of tech with Web Summit https://websummit.com/ . The event covers everything from cybersecurity and energy to developers, policy, health, and of course AI. Students, journalists, investors, CEOs, and founders looking for funding all end up in the same hallways. I was there representing Cisco with a session called “Democratising AI Platform Engineering with Open Source”. My goal was to explore what AI platform engineering is, why it matters whatever you use your agents for, and how open source can get you there. AI Platform Engineering, the Open Source Way If you want AI agents to be a real part of your Monday, working side by side with your team, you need some solid foundations: authentication, authorization, observability, and plenty more, so your agents, tools, and skills run responsibly. You have three options: 1. Build it yourself or vibe code it , and own the plumbing and maintenance. 2. Pay for a black box you can’t change. 3. Adopt a mature open-source project with the plumbing ready and the batteries included. My session was all about option three. CAIPE https://caipe.io pronounced “cape” is a CNCF Sandbox project: an open-source AI platform that helps teams build, govern, and operate secure AI agents and workflows. It brings authentication, authorization, MCP tools, RAG, and workflows together under a single pane of glass WebUI, APIs, Slack, Webex, and more . You deploy it your way kind, Helm, Docker Compose and tinker at will. My favorite part is that every component is FOSS and swappable. Swap the default Keycloak for the Duo you already use, bring your LLM of choice, or index your knowledge from Confluence, Slack, or any URL into a RAG built on the included Neo4j. There is no silver bullet, of course. Every team has different needs and trade-offs. But a project like this gives you a head start, and contributions from people in the trenches make it better for everyone. Talking to AI Leaders The Monday before Web Summit, I was invited to the AI Engineering Night for Leaders in Tech https://luma.com/ai-leaders-in-tech by AI Builders. The audience was mostly founders, CEOs, and CTOs of startups and mid-size tech companies. As an engineer and Developer Advocate, I had to switch gears and talk their language: how open source can bring value to your team, starting with the bare minimum and scaling at the pace of your company. Enter Storm Charles Web Summit’s second day was canceled Storm Charles was heading straight for Lisbon, so an emergency state was declared and non-essential business Web Summit included shut down. Charles came and went, the city held up, and the next day was the big one for my masterclass https://websummit.com/sessions/lis26/5f4ac039-83e3-4b2b-9415-3efa64804ee1/democratising-ai-platform-engineering-with-open-source-hosted-by-cisco-systems/ . I had 45 minutes and 111 attendees in a room built for a hundred, and I loved every minute: the energy, the vibe, the follow-up conversations. The weeks of preparation were worth it. In Conclusion If you want to bring AI agents to your environment the right way, give open source a chance. CAIPE is a robust alternative to building and maintaining everything yourself, or relying on proprietary services. One honest caveat: open source isn’t free. You pay in a different currency. The tinkering is on you and the support comes from the community, but so does the roadmap, and ultimately the benefit is shared by everyone. Sidenote: Record Your Demos I’m a true believer in live demos. I love the adrenaline. But they can bite back. Midway through my demo at the AI Builders event, none of my LLM API requests were working. I didn’t even try to troubleshoot live: I accepted defeat, apologized, and moved on. During Q&A, someone told me that an OpenAI outage had caused it. What are the odds of that happening at that exact moment? The moral: record your demos, or run your LLMs locally. Whichever suits you best. Curious about AI platform engineering and CAIPE? Your starting point is the official CAIPE website https://caipe.io/ . Try the DevNet Learning Lab https://developer.cisco.com/learning/search/?contentType=lab&keyword=caipe right in your browser, for free And here’s a small, silly video https://www.youtube.com/shorts/PwotrMpXfHk about my Web Summit Lisbon 2026 experience.