DigitalOcean

Last month, DigitalOcean’s Product Security team came together to work out what a security program looks like when the SDLC is no longer a line you can shift left on. I’ve been calling what replaced it the Jeremy Bearimy SDLC (yes, from The Good Place). I wrote up where we landed, including why mandates are fine as long as they apply to agents and not people, and how a sandbox escape in one of our AI products ended up as a routine bug fix instead of an incident. Take a look! https://lnkd.in/eP2DFUvd

Introducing DigitalOcean MicroVMs: isolated virtual machines that pause when idle and resume exactly where they left off. An AI agent works in bursts. It runs code, calls a model, waits on a tool or a person, then picks the work back up. Traditional infrastructure was never built for that shape of work, so teams building agents and agentic infrastructure end up building their own virtualization layer for fast startup, pause and resume, and per-tenant isolation. That is real engineering, and for most of them it is not the product. MicroVMs are the answer to that problem. We run the fleet, the isolation, and the lifecycle underneath, so you can build the part that is actually yours. MicroVMs already power DigitalOcean Managed Agents. Now we're making them directly available to teams building their own agent infrastructure, in private preview. If that's the layer you've been building yourself, stop building it and start using it. Read the full write-up: https://lnkd.in/gmSsqD8Q Request access to the private preview: https://lnkd.in/gvDBdUmw Real credit to the team that got us here. Mavis Franco Braden Bassingthwaite Kushal W. Omair Gillani Chris Browder Vinay Kumar Aparna Prabhu Vinay Reddy

I joined DigitalOcean as Product Director last week. Inference is now the center of gravity of every AI stack. DigitalOcean is building the AI-native cloud for production AI workload, from GPU infrastructure through inference to data. In my first week here, I’ve already seen the speed of execution, deep customer focus, and real urgency around building AI infrastructure at scale. I’m excited to be leading our Compute product charter to help power this shift toward an AI-native world. We’re hiring across Product and Engineering in AI and Compute. If this sounds like your kind of challenge, I’d love to hear from you.

Introducing DigitalOcean Insights: a new unified observability product for all your AI agents, infrastructure, and applications running on DigitalOcean, all in one place. If you know me professionally, you know how much I care about observability 🙂 It has only gotten harder as stacks grow more complex and layered. For example, an agent spans a runtime, a model, and a set of tools. Your applications run on Kubernetes or App Platform. When something slows down or breaks, the evidence is scattered across separate consoles, each with its own query language and its own bill. Answering a simple question about your own stack should not require standing up an observability pipeline first. Insights is integrated with the DigitalOcean products you already run. Metrics, logs, and traces flow in with nothing to install and no instrumentation. Since all of them share the same backend, an anomaly in one signal leads you straight to the cause in another. It is built on OpenTelemetry, so your data stays in an open, portable format rather than locked to a proprietary one. Insights already powers observability for DigitalOcean Managed Agents. Now we’re opening it across the rest of your stack, from inference and agents down to the infrastructure they run on. Detect, investigate, and monitor, in one place. Insights is available at no additional charge during public preview. Read the full blog post at: https://lnkd.in/gGq9sEDA All credits to the incredible team that built this.

Launching an agent is the easy part now. Getting it to improve with every run is where most teams stall. At the Open Intelligence Summit, we made one argument: DigitalOcean is the best place to run self-improving agents. Self-improvement is a loop, not a feature. A loop only closes if everything it touches sits in the same place. Here are the four pieces we shipped to make that happen. Think. Inference Services puts 85+ open and frontier models behind one endpoint. Inference Router sends each prompt to the model that clears your quality bar at the best cost. Nearly 90% of the tokens on our platform now run on open-weight models, and token volume is up 50x in five months. Act. Managed Agents gives every session its own microVM, ready in under a second. It runs Claude Code, Codex, OpenCode, Hermes and LangGraph as they are, and you pay for active CPU only. Action Gateway adds one endpoint to 20,000+ tools. The credentials never touch the model, the prompt or the sandbox. Know. Agent Data Services lets an agent ask for your business data, the web’s data (through Bright Data) and its own traces mid-run, in SQL or by meaning. No overnight ETL. It’s built on Apache Iceberg and Apache Gravitino, so your data stays yours. Improve. Agent Learning Services is where the loop closes. Signals finds the failures. Simulations turn them into test scenarios. Evaluations score the old agent against the new one. The fix ships only after it’s proven. Each of these can be bought separately. What’s hard to assemble from five vendors is one record that all five read and write. The runtime writes traces. Signals and Evaluations read them. Data Services holds them. That is the loop, and it’s why teams stop hand-tuning prompts and start compounding. One of our customers, Amplitude is already running this way. Spenser Skates, CEO of Amplitude - explained why Wave, their agentic product team, runs on Managed Agents. It studies product usage, writes the pull requests and measures the result. Agents are the new primitive for the cloud, the way the VM was for the last one. Cloud 1.0 was built for software that humans drive. This one is built for agents that think, act and learn. If every fix to your agent had to be proven before it shipped, how many of your current fixes would pass? Keynote replay: https://lnkd.in/gZrAUz_4 Managed Agents: https://lnkd.in/gr4aj8bi Learning Services: https://lnkd.in/gJFPGxzF Agent Data Services: https://lnkd.in/gxT-ZUZk #SelfImprovingAgents #OpenIntelligence #AIAgents

🚀 Today we're launching DigitalOcean Signals and Simulations in public preview. Your agent is up and latency is fine, but it picked the wrong tool, used stale context, or made a user give up. Dashboards miss it, and validating a fix usually means guesswork. Signals monitors traffic on Managed Agents and shows you where your AI agents are going wrong in production. Simulations lets you test the fix before it reaches a user. Signals keeps watching, so every fix sharpens the next one. The loop repeats 🚀 Link in the comments #AI #ManagedAgents #ArtificialIntelligence #GenerativeAI