BEGIN:VCALENDAR VERSION:2.0 PRODID:-//HasGeek//NONSGML Funnel//EN DESCRIPTION:Real systems. Real engineers. Real lessons. X-WR-CALDESC:Real systems. Real engineers. Real lessons. NAME:Platform Engineering meet-up - September 5 X-WR-CALNAME:Platform Engineering meet-up - September 5 REFRESH-INTERVAL;VALUE=DURATION:PT12H SUMMARY:Platform Engineering meet-up - September 5 TIMEZONE-ID:Asia/Kolkata X-PUBLISHED-TTL:PT12H X-WR-TIMEZONE:Asia/Kolkata BEGIN:VEVENT SUMMARY:Check-in DTSTART:20260905T091500Z DTEND:20260905T093000Z DTSTAMP:20260909T011940Z UID:session/Lv482tfR3TNvpjCqLYyRsq@hasgeek.com SEQUENCE:1 CREATED:20260901T071423Z LAST-MODIFIED:20260901T071427Z LOCATION:Bangalore & Hybrid ORGANIZER;CN=Rootconf:MAILTO:no-reply@hasgeek.com BEGIN:VALARM ACTION:display DESCRIPTION:Check-in in 5 minutes TRIGGER:-PT5M END:VALARM END:VEVENT BEGIN:VEVENT SUMMARY:Introduction to the meet-up DTSTART:20260905T093000Z DTEND:20260905T094000Z DTSTAMP:20260909T011940Z UID:session/PxVduVuAFLviDQPncqHYax@hasgeek.com SEQUENCE:2 CREATED:20260901T071458Z GEO:12.9027111;77.6010828 LAST-MODIFIED:20260901T071504Z LOCATION:Board Room - Blinkit\, 7th floor\, Vaishnavi BVS Senate\nBengalur u\,\nIN ORGANIZER;CN=Rootconf:MAILTO:no-reply@hasgeek.com BEGIN:VALARM ACTION:display DESCRIPTION:Introduction to the meet-up in Board Room in 5 minutes TRIGGER:-PT5M END:VALARM END:VEVENT BEGIN:VEVENT SUMMARY:Beyond AI sandboxing: decoupling AI from execution DTSTART:20260905T094000Z DTEND:20260905T101500Z DTSTAMP:20260909T011940Z UID:session/Qi5Dgq5nSdM5Wk538c4DGB@hasgeek.com SEQUENCE:4 CATEGORIES:Talk (30 mins),Bangalore meet-ups CREATED:20260901T071513Z DESCRIPTION:**Beyond AI Sandboxing: Decoupling AI from Execution**\n\n---\ n\nAI coding assistants are transforming software development\, but they a lso introduce a new security challenge. Today's AI agents typically execut e directly on developer workstations or require developers to expose secre ts\, cloud credentials\, and production systems to hosted platforms. This creates unnecessary trust assumptions and significantly expands the attack surface\, especially when AI-generated applications depend on hundreds of open-source packages that may themselves be compromised. The question org anizations should be asking is no longer *"Which AI model should we use?"* but rather *"Where should AI agents execute\, and what should they be all owed to access?"*\n\nThis session introduces a different architectural app roach using the open-source [**Agentry**](https://agentry.run/) platform. Rather than sandboxing the AI agent itself\, Agentry allows any MCP-compat ible AI coding assistant—such as Claude Code\, Cursor\, Cline\, or OpenC ode—to remotely operate a secure development environment running on infr astructure you already control. Developers continue using their preferred AI tools and can seamlessly switch between different models and coding ass istants throughout the development lifecycle without changing the executio n environment. Through architecture deep-dives and live demonstrations\, a ttendees will learn how remote sandboxing\, zero-trust connectivity\, isol ated execution\, and containerized deployments enable secure AI-assisted s oftware development without disrupting existing developer workflows.\n\n-- -\n\n# Takeaways\n\n- **Understand a new security architecture for AI-assi sted development** where AI agents interact with remote sandboxed environm ents over MCP instead of executing directly on developer machines or hoste d cloud platforms.\n\n- **Learn how remote sandboxing preserves developer productivity** by enabling teams to continue using their preferred AI codi ng assistants while securely building\, testing\, and deploying applicatio ns on infrastructure they already own.\n\n---\n\n# Who will benefit from t his session?\n\nThis session is ideal for:\n\n- Software Engineers and Ful l-Stack Developers using AI coding assistants\n- Platform Engineers and De vOps Engineers\n- Security Engineers and Application Security (AppSec) pro fessionals\n- Engineering Managers and Technical Architects evaluating ent erprise AI adoption\n- Open Source contributors and AI infrastructure enth usiasts\n\n---\n\n# About me\n[Gowtham Sadasivam](https://www.linkedin.com /in/gowthamsadasivam/) is a Senior Staff Engineer at [Acceldata](https://w ww.acceldata.io/) with 13+ years of experience designing and operating Lin ux\, Kubernetes\, and cloud-native application development. His current wo rk focuses on AI infrastructure\, Agentic AI\, self-hosted LLMs\, and secu re software engineering.\n\n--- GEO:12.9027111;77.6010828 LAST-MODIFIED:20260901T071619Z LOCATION:Board Room - Blinkit\, 7th floor\, Vaishnavi BVS Senate\nBengalur u\,\nIN ORGANIZER;CN=Rootconf:MAILTO:no-reply@hasgeek.com URL:https://hasgeek.com/rootconf/platform-engineering-meet-up-september-5/ schedule/beyond-ai-sandboxing-decoupling-ai-from-execution-Qi5Dgq5nSdM5Wk5 38c4DGB BEGIN:VALARM ACTION:display DESCRIPTION:Beyond AI sandboxing: decoupling AI from execution in Board Ro om in 5 minutes TRIGGER:-PT5M END:VALARM END:VEVENT BEGIN:VEVENT SUMMARY:Birds of Feather (BOF) session: The new age of building on object storage DTSTART:20260905T101500Z DTEND:20260905T110000Z DTSTAMP:20260909T011940Z UID:session/L7QVjTY5Cg8oSFYQpeE9ZH@hasgeek.com SEQUENCE:4 CREATED:20260901T071534Z DESCRIPTION:\nWhy are database and database-adjacent systems increasingly being built directly on S3? A growing list of systems are betting on objec t storage as the primary layer\, not just cold storage.\n\nOpen discussion :\n* Why build on S3 in the first place — what does it actually buy you? \n* Why not — where does this break down?\n* What kinds of systems does this pattern not make sense for?\n GEO:12.9027111;77.6010828 LAST-MODIFIED:20260901T072053Z LOCATION:Board Room - Blinkit\, 7th floor\, Vaishnavi BVS Senate\nBengalur u\,\nIN ORGANIZER;CN=Rootconf:MAILTO:no-reply@hasgeek.com BEGIN:VALARM ACTION:display DESCRIPTION:Birds of Feather (BOF) session: The new age of building on obj ect storage in Board Room in 5 minutes TRIGGER:-PT5M END:VALARM END:VEVENT BEGIN:VEVENT SUMMARY:Observing and governing agents you did not build DTSTART:20260905T110000Z DTEND:20260905T113500Z DTSTAMP:20260909T011940Z UID:session/U5Jv5V3uH9Yy5yLsjq7378@hasgeek.com SEQUENCE:4 CATEGORIES:Talk (30 mins),Bangalore meet-ups CREATED:20260901T071542Z DESCRIPTION:Uber burned its entire 2026 AI budget in four months\, and its COO said publicly that the link to customer value is not there yet. Repli t's agent deleted a production database during a code freeze that existed only as English inside a prompt. The tools do work: Microsoft's study of i ts early 2026 command-line agent rollout measured about 24 percent more me rged pull requests for adopters\, and named token spend as the governing c onstraint. What is missing is a control plane\, not intelligence.\n\n The claim that these agents emit nothing is wrong\, which makes it worse rath er than better. Claude Code's OpenTelemetry exporter redacts prompt\, resp onse and tool content by default. Cursor's export is Enterprise only and i ts own documentation says\, in those words\, that cost is not billing. Cop ilot's audit log excludes the prompts a user sends locally. Three agents\, three partial and incompatible feeds\, and still no answer to what the ag ent changed\, on whose instruction\, who approved it\, and what it cost.\n \n Coding agents already write down what they did\, they just never expos e it. So ClawMetry reads those session files directly with no cooperation from the vendor\, normalises every runtime onto one contract\, and lands i t in a local DuckDB file\, so cost per model\, per session and per tool is a SQL query you own rather than a dashboard someone rents you. A real ada pter is nine hundred lines when the parsing is fifty\, and the other eight hundred and fifty are what I will spend the middle of the talk on.\n\n G overning needs less than people expect. An agent is a process on your mach ine\, so stopping one needs its process tree\, not a vendor integration. T he harder question is what to stop\, and risk turns out to be a property o f the call rather than the tool: definition-level classification cannot te ll `ls` from `rm -rf /` when both arrive through the same shell tool on ev ery harness. Normalise the tool vocabulary once and one rule covers every runtime you monitor. The talk opens with an agent deleting a production da tabase and ends with that call being held live on stage\, then denied. All of it ships off by default\, which is the only honest setting for softwar e that can halt someone's work. MIT core.\n\n## Takeaways\n\n**1. Classify the call\, not the tool.** Nothing that grades tool names can tell `ls` f rom `rm -rf /`\, because both arrive through the same shell on every harne ss. Normalise the vocabulary once\, classify from the arguments\, and one rule covers every runtime instead of one per vendor. Write the reasons as plain sentences\, because a human reads them in an approval prompt at thre e in the morning.\n\n**2. A format you do not control changes without tell ing you.** Pin real captures as fixtures so a vendor's change fails your C I instead of silently zeroing someone's cost report. Then check on a sched ule: clone the source where the harness is open\, and where it is closed\, install their latest build every night\, drive one minimal turn\, and ass ert it still lands. Grade each runtime live-verified\, fixture-replayed or unverified\, and never let unverified render as green.\n\n## Who this is for\n\n- Platform and developer-experience engineers rolling out coding ag ents across teams and CI\n- SREs and observability engineers who own the t elemetry pipeline and will be handed this problem next\n- Anyone who has t o read a data format they do not control and cannot ask the vendor to chan ge\n- Engineering managers and architects who have to produce a cost and a udit story for agent usage\n- Anyone building on OpenTelemetry who needs t o know what the GenAI conventions do not yet carry\n\n## About me\n\nI am Vivek Chand\, AI Engineer at Booking.com\, creator and maintainer of [Claw Metry](https://github.com/vivekchand/clawmetry)\, an open-source observabi lity and governance layer for AI coding agents (MIT core). I spend my time at the unglamorous end of this problem: ingest adapters for agent runtime s that were never designed to be observed\, an embedded columnar store\, a nd an enforcement proxy. \n GEO:12.9027111;77.6010828 LAST-MODIFIED:20260901T072140Z LOCATION:Board Room - Blinkit\, 7th floor\, Vaishnavi BVS Senate\nBengalur u\,\nIN ORGANIZER;CN=Rootconf:MAILTO:no-reply@hasgeek.com URL:https://hasgeek.com/rootconf/platform-engineering-meet-up-september-5/ schedule/observing-and-governing-agents-you-did-not-build-U5Jv5V3uH9Yy5yLs jq7378 BEGIN:VALARM ACTION:display DESCRIPTION:Observing and governing agents you did not build in Board Room in 5 minutes TRIGGER:-PT5M END:VALARM END:VEVENT BEGIN:VEVENT SUMMARY:Feedback\, group photo DTSTART:20260905T113500Z DTEND:20260905T114500Z DTSTAMP:20260909T011940Z UID:session/BfpAn1WdXu4wP35bQ7K8b1@hasgeek.com SEQUENCE:2 CREATED:20260901T072151Z GEO:12.9027111;77.6010828 LAST-MODIFIED:20260901T072159Z LOCATION:Board Room - Blinkit\, 7th floor\, Vaishnavi BVS Senate\nBengalur u\,\nIN ORGANIZER;CN=Rootconf:MAILTO:no-reply@hasgeek.com BEGIN:VALARM ACTION:display DESCRIPTION:Feedback\, group photo in Board Room in 5 minutes TRIGGER:-PT5M END:VALARM END:VEVENT BEGIN:VEVENT SUMMARY:Networking and snacks DTSTART:20260905T114500Z DTEND:20260905T123000Z DTSTAMP:20260909T011940Z UID:session/LTN7YS7VHamzsEPNxkD6YL@hasgeek.com SEQUENCE:1 CREATED:20260901T072207Z GEO:12.9027111;77.6010828 LAST-MODIFIED:20260901T072212Z LOCATION:Board Room - Blinkit\, 7th floor\, Vaishnavi BVS Senate\nBengalur u\,\nIN ORGANIZER;CN=Rootconf:MAILTO:no-reply@hasgeek.com BEGIN:VALARM ACTION:display DESCRIPTION:Networking and snacks in Board Room in 5 minutes TRIGGER:-PT5M END:VALARM END:VEVENT END:VCALENDAR