Sam Partee (CTO & co-founder of Arcade.dev) and Nate Barbettini (Founding Engineer at Arcade.dev) sit down at the MCP Dev Summit to unpack what nobody wants to admit about the Model Context Protocol: the security model is still full of sharp edges. From tool poisoning and prompt injection to why OAuth got bolted onto the spec, this is a builder 's-eye view of where MCP breaks — and how to ship agents safely anyway.

The Dark Side of MCP Servers

Managing subagent personas across multiple AI coding platforms (like Claude Code, Gemini CLI, and Copilot) leads to fragmented configuration files and instruction drift. This article introduces `supagents`, a lightweight, idempotent Python CLI tool that lets developers maintain subagent definitions in a single markdown file with multi-target frontmatter. The tool automatically compiles and distributes these instructions to the correct paths, ensuring a consistent persona across all development environments.

Supagents: One Source to Rule Your Coding Subagents

Shahram Anver is the Co-Founder and CEO of Cleric, the autonomous AI SRE that investigates and root-causes production issues like an experienced teammate — often in under two minutes. Before Cleric, Shahram led MLOps, DevOps, and FinOps platform engineering at Gojek, Southeast Asia's super-app. In this conversation, he breaks down why production operations never kept pace with AI-accelerated development, and why the real unlock for an AI SRE isn't faster triage — it's an agent that *learns* and compounds operational memory across your whole org.

Sandboxing, Agent Harnesses, and Agent Teamwork

Retrieval-Augmented Generation and agentic AI are increasingly common in enterprise deployments, but real enterprise environments introduce challenges largely absent from academic treatments and consumer-facing APIs: multiple tenants with heterogeneous data, strict access-control requirements, regulatory compliance, and cost pressures that demand shared infrastructure. 

This paper identifies a fundamental problem underlying existing RAG architectures in these settings. Retrieval systems rank documents by relevance, not by authorization, so a query from one tenant can surface another tenant’s confidential data simply because it scores highest. The authors formalize this relevance-authorization gap alongside related shortcomings (tool-mediated disclosure, context accumulation across turns, client-side orchestration bypass) and introduce a layered isolation architecture combining policy-aware ingestion, retrieval-time gating, and shared inference, enforced through server-side orchestration. They validate it through an open-source implementation in OGX, a vendor-neutral OpenAI-compatible Responses API, showing empirically that ABAC gating eliminates cross-tenant leakage while introducing negligible overhead.

Securing the Agent: Vendor-Neutral, Multitenant Enterprise Retrieval and Tool Use

AI agents are only as good as the information they can find, retrieve, and remember.Join us for a special community discussion with the Qdrant team as we explore the latest advances in agentic memory, vector search, retrieval systems, and production AI architectures. As agents evolve beyond simple chatbots into autonomous systems capable of reasoning across large amounts of information, retrieval has become one of the most critical layers in the AI stack.Together, we'll dive into the practical challenges of building systems that can remember what matters, forget what doesn't, and consistently retrieve the right context at the right moment. <h3>What we'll explore</h3> 🧠 Agentic Memory Architectures<ol><li data-list="bullet">Why memory is becoming a core component of production AI systems</li><li data-list="bullet">Managing long-term memory without bloating context windows</li><li data-list="bullet">Memory lifecycle management: what agents should remember, and what they should forget</li><li data-list="bullet">Practical patterns for persistent memory using vector databases</li></ol>🔍 Late Chunking &amp; Production RAG<ol><li data-list="bullet">Why traditional chunking strategies often fail</li><li data-list="bullet">Preserving document-level context and meaning</li><li data-list="bullet">Improving retrieval quality and relevance in real-world applications</li></ol>📊 Evaluating Search Quality<ol><li data-list="bullet">Measuring retrieval performance beyond simple accuracy metrics</li><li data-list="bullet">Semantic search vs exact-match retrieval</li><li data-list="bullet">Practical frameworks for evaluating search systems in production</li></ol>🌐 Embeddings Beyond Text<ol><li data-list="bullet">Retrieval for code, images, tables, time-series, and multimodal data</li><li data-list="bullet">Emerging patterns for next-generation AI applications</li><li data-list="bullet">Real-world examples from production deployments</li></ol>You’ll hear from an exceptional lineup of experts from Qdrant, including Neil Kanungo, Head of Developer Relations, alongside Ewa Szyszka and Dylan Couzon, both Developer Relations Engineers, and Evgeniya Sukhodolskaya, Senior Developer Advocate. Together, they bring deep experience in vector search, retrieval systems, RAG architectures, and production AI applications, offering practical insights from helping developers and organizations build scalable AI systems powered by semantic search and agentic workflows. Hosted by Demetrios Brinkmann, Chief Happiness Engineer at MLOps Community, this session promises an engaging mix of technical expertise, real-world lessons, and interactive discussion.Expect practical insights, real-world examples, audience Q&amp;A, and plenty of opportunities to engage directly with the engineers helping shape the future of vector search and AI retrieval systems.Whether you're building RAG applications, AI agents, search systems, or knowledge platforms, you'll leave with actionable ideas you can apply immediately.Because when AI systems fail, it's often not because they can't think.It's because they can't find what they need to know.

The Current State of Agentic Retrieval

Join us for Session 16 of the Coding Agents Lunch &amp; Learn series as we take a closer look at how developers, researchers, and engineering teams are integrating coding agents into their day-to-day workflows. While the capabilities of coding agents continue to evolve rapidly, understanding how to effectively incorporate them into real engineering processes remains one of the most important challenges facing teams today. In this session, we'll explore practical approaches to working with coding agents across the software development lifecycle. From planning and research to implementation, testing, debugging, and deployment, participants will have the opportunity to discuss the workflows that are delivering real value and the lessons learned from using these tools in production environments. We'll cover a range of experiences, from individual developers experimenting with new tools to teams building collaborative agent-driven workflows. As always, this Lunch &amp; Learn is designed to be highly interactive and community-driven. Attendees are encouraged to share their own experiences, challenges, successes, and discoveries. Whether you're just beginning your journey with coding agents or already operating sophisticated agent-assisted workflows, you'll leave with new ideas, practical insights, and a better understanding of how others are approaching this rapidly changing space. Bring your questions, your curiosity, and anything interesting you've been building.

Coding Agents Lunch & Learn Session 16: Real-World Coding Agent Workflows

As coding agents become increasingly capable, the conversation is shifting from simply using agents to understanding how to build reliable, repeatable, and scalable agent-powered systems. Session 17 focuses on the engineering practices, evaluation techniques, and operational considerations that help transform promising agent workflows into dependable tools that teams can trust. Throughout this session, we'll discuss how practitioners are approaching agent development, including workflow design, prompt engineering, skill development, evaluation methodologies, observability, testing, and operational best practices. We'll explore common challenges teams encounter when working with agents and share lessons learned from both successful implementations and failures. Participants will gain insight into how experienced builders think about reliability, quality, and performance when deploying agent-assisted workflows at scale. In addition to practical discussions, we'll highlight emerging tools, frameworks, and techniques that are helping developers create more effective agent systems. Whether you're experimenting with coding agents for personal projects or deploying them across larger engineering organizations, this session aims to provide actionable ideas and valuable perspectives from people actively building in the space. Join us for an engaging conversation about what it takes to build, evaluate, and operate coding agents effectively in today's rapidly evolving ecosystem.

Coding Agents Lunch & Learn Session 17: Building, Evaluating & Operating Agents

Session 18 is dedicated to one of the most valuable aspects of the Coding Agents Lunch &amp; Learn series: learning directly from the community. This special Community Show &amp; Tell session provides an opportunity for attendees to share the projects, workflows, tools, experiments, and discoveries they've been working on while exploring the world of coding agents and agentic engineering. The pace of innovation in this space continues to accelerate, and many of the most interesting breakthroughs come from practitioners experimenting with new ideas, combining tools in unexpected ways, and pushing existing workflows beyond their intended use cases. During this session, participants are encouraged to showcase what they've built, discuss challenges they've encountered, demonstrate useful techniques, and exchange lessons learned with fellow builders. Whether you're working on coding agents, multi-agent systems, evaluation frameworks, developer tooling, infrastructure, open-source projects, or entirely new approaches to software development, this is a chance to share your experiences and learn from others doing the same. No formal presentation is required—community discussion, live demonstrations, workflow walkthroughs, and informal knowledge sharing are all welcome. If you're looking for inspiration, feedback, or simply a chance to connect with other practitioners exploring the future of software development, this session is for you. Bring your projects, your questions, and your ideas as we close out another chapter of the Coding Agents Lunch &amp; Learn series together.

Coding Agents Lunch & Learn Session 18: Community Show & Tell

Join us for Session 15 of the Coding Agents Lunch &amp; Learn, our weekly community series exploring the tools, workflows, frameworks, and real-world applications shaping the future of AI coding agents.This week, we're exploring one of the most important challenges in agent engineering: how to build AI systems that can retain knowledge, maintain context, and scale without exploding inference costs.As agent workflows become more complex, teams are running into the limits of today's architectures. Agents repeatedly reread documents, lose important context, struggle with long-running tasks, and consume increasingly expensive amounts of tokens to maintain performance. While larger context windows have helped, they haven't solved the fundamental problem of how AI systems store, share, and build upon knowledge over time.For this session, we're joined by Devansh, Head of AI at Iqidis and creator of Stateful Swarms, an open-source architectural paradigm designed to make agent systems more intelligent, cost-efficient, and auditable.Instead of forcing agents to repeatedly recompute information and pack everything into ever-growing context windows, Stateful Swarms introduce a persistent, structured memory layer powered by a coordinated blackboard system. Specialized agents collaborate through a shared knowledge base, preserving important discoveries, reducing context loss, and enabling information to be reused across iterations. The result is a more scalable approach to agent engineering that improves reliability while dramatically reducing the cost of execution.During the session, Devansh will walk through the design principles behind Stateful Swarms, why persistent memory may be the missing layer in modern agent systems, and how this approach can dramatically improve both performance and cost efficiency. He'll also share benchmark results from Harvey AI's Legal Agent Benchmark, where Stateful Swarms achieved an 83.74% pooled criteria pass rate and a 17.75% strict all-pass rate at a fraction of the cost of previously published approaches.Through a live walkthrough and technical deep dive, we'll explore:• Why context windows and token costs are becoming major bottlenecks for AI agents• Persistent memory architectures and structured knowledge systems• Blackboard-based coordination for multi-agent workflows• Reducing recomputation and preventing context loss across complex tasks• Building more auditable and observable agent systems• Open-source tooling and implementation details behind Stateful Swarms• Lessons learned from benchmarking and real-world agent deploymentsWe'll also discuss the broader shift toward stateful agents, why memory and context management have become foundational challenges in agent engineering, and how the wider Agentic AI Foundation ecosystem is helping shape the next generation of agent infrastructure.Whether you're building coding agents, autonomous workflows, AI copilots, or next-generation agent infrastructure, this session will provide practical insights into how persistent memory and stateful architectures could reshape the future of agent engineering.

Coding Agents Lunch & Learn Session 15: Building Stateful Agents with Open Source Frameworks

The MLOps Community is where machine learning practitioners come together to define and implement MLOps.
Our global community is the default hub for MLOps practitioners to meet other MLOps industry professionals, share their real-world experience and challenges, learn skills and best practices, and collaborate on projects and employment opportunities. We are the world's largest community dedicated to addressing the unique technical and operational challenges of production machine learning systems.

MLOps Community

Home

Forum

Events

Content

Collections

People

Contributors

Albums

Clubs

Chat

Channels

Tools

Help

Your Privacy Choices

It looks like there is no account associated with the email you provided. The event is for members only. Please complete the application to join our community first.

Thank you for your interest in joining us

Thanks for your interest in joining the event. However, your account is still pending. Once it's approved, you'll be able to register for the event. Please contact us via help@mlops.community if you need support.

Pending Approval

Login to the MLOps Community community to connect with others, attend community events, and more!

Login or Sign Up for even more

To unlock all parts of the community and get the best experience, complete your profile.

Hey, complete your profile

It looks like there is no account associated with the email you provided. Please complete the application to join our community.

Thank you for your interest in joining MLOps Community.

Thanks for your interest in joining MLOps Community. We will review your request shortly and email you once it's approved. Meanwhile, please feel free to contact us via help@mlops.community for any questions.