Docker, Inc

AI workflows are moving faster than the systems teams use to review, secure, and ship them. This issue of the Docker Navigator newsletter looks at what breaks and how teams are adapting: - Hardening images without breaking developer workflows - Isolating workloads beyond containers - Responding to supply chain attacks in real time - Moving from blocked deployments to production-ready systems If you’re working through these challenges today, this issue brings together how teams are actually solving them. Read more →

Not all isolation models are built for agents. This post from Docker Captain Siri Varma Vegiraju walks through what happens when you try to run agents across different environments, chroot, systemd-nspawn, containers, VMs, gVisor, and microVMs, and where each one holds up or falls apart. The takeaway is pretty clear: once agents can install, mutate, and execute freely, most traditional assumptions around isolation start to break. If you’re thinking about how to run agents safely, this is a practical comparison worth reading. Read →

AI can help you ship faster, but what happens when you don’t fully trust the code it writes? In this episode of Ship Happens, Per Ploug Krogslund sits down with Ivar Conradi Østhus (CTO and creator of Unleash) to explore how teams are adapting as speed increases, but control becomes harder to maintain. They get into: - Why deploying daily is now table stakes for modern teams - How AI is accelerating development and increasing risk - The importance of using feature flags act as a safety net for production - How “feature ops” helps teams learn faster, not just ship faster One key idea: every change is an experiment, but production is where you learn what actually works. Watch the full episode → https://lnkd.in/dUS6qsvu

Letting agents run freely sounds great… until they start changing everything. In this episode of Data Science Dojo's ‘Future of Data and AI’, hosted by Raja Iqbal, Docker COO Mark Cavage explains why containers alone aren’t enough once agents start installing packages, mutating environments, and executing tasks end to end. The core idea: agents need freedom to work, but infrastructure has to define the boundary. Watch the full conversation to see why sandboxing is becoming a requirement for real agent workflows → https://bit.ly/4uDFPtl

Introducing Docker AI Governance. Every team is racing to deploy agents. Engineering, marketing, finance, all moving faster than ever. But agents don’t just assist. They access private repos, customer data, production APIs, and in most cases, there’s no visibility into what they’re doing. To run agents at scale, you need a policy layer underneath everything. Sandbox, network, and MCP controls, defined once, enforced everywhere, and fully auditable. Docker AI Governance is now available, so teams can move fast without losing control. Learn more →

🎙️What happens when you let an AI agent run in YOLO mode? Mark Cavage has a pretty interesting take on the current wave of Agentic AI.. Mark helped build early AWS, co-founded Oracle Cloud Infrastructure, and now runs Docker, Inc as President & COO. He's spent two decades at the infrastructure layer — so when he talks about how agents actually run, it's worth paying attention. In this episode, Raja Iqbal sits down with Mark Cavage to explore: 🔹 Why cloud infrastructure is being rebuilt for the agentic era 🔹 What happens when AI writes, runs, and deploys its own code 🔹 1000x productivity. 1000x risk. Who's managing the gap? This one's for the engineers building with agents, the security leads wondering what's slipping through, the CTOs figuring out what "agentic" means for their stack, the founders deciding where to build, and anyone who's ever wondered what actually happens when you give an AI too much autonomy. Episode drops tomorrow | May 12 at 12 PM PDT. YouTube · Spotify · Apple Podcasts

Containers became the default path to production. That also made the container supply chain one of the biggest security risks teams now have to manage. At Devoxx UK, Docker Captain Matthias Haeussler gave a great talk on Docker Hardened Images - what you need to know, and the level of interest in the room was a clear signal. Developers are actively looking for better ways to handle container security, not as an afterthought, but as part of how software gets built and shipped. A strong indication of where the ecosystem is heading.

There’s a point where using hosted AI tools starts to feel limiting. This guide walks through setting up your own local image generation stack with Docker Model Runner and Open WebUI, running entirely on your machine. You get a chat interface, API access, and full control over models and data without relying on external services. What stands out is how simple the setup is, and how easily it fits into real workflows. Worth exploring if you’re building anything multimodal or want to keep things local. Read →

Running multiple AI agents at once? That’s where the real problems start, and you’ve probably already hit the edge of it: shared context, conflicting changes, resource contention, and unclear boundaries between agents. In the recent issue of the Docker Navigator, we break down what that looks like in practice: • Isolated environments for agents • Local models in real workflows • Security and reliability at scale This is where you find out if your setup actually works - and what it takes to make it reliable: https://bit.ly/3QHZcTA