Principal Software Engineer - Security

DigitalOcean (NYSE: DOCN) has launched Agent Learning Services and Agent Data Services, which let companies build AI agents that improve themselves. The cloud provider announced the products ahead of its Open Intelligence Summit in San Francisco. Learning Services record what an agent does, flag where it fails, and test fixes before customers see them. Data Services give agents one managed place to read context and write back what they learn. The enhanced Inference Router already applies this loop to models, picking the best-scoring one for each task. DigitalOcean plans to extend it to agent workflows. Learning Services are in public preview. Data Services and the Inference Router optimizer are in private preview.

Digitalocean introduces Agent Droplets: everything an ai agent needs. Published on 10/01/2026 at 02:21 pm EDT S&P Capital IQ DigitalOcean Holdings, Inc. announced Agent Droplets giving companies a simple way to start their agentic journey on DigitalOcean and scale with predictable spending controls. DigitalOcean Managed Agents, launched in public preview on September 22, gave builders a managed runtime, governed access to 16,000+ tools, Serverless Inference, and persistent agent memory and storage on one cloud. Agent Droplets bundle those capabilities into a single monthly subscription simplifying adoption of emerging agent technologies without the guesswork of assembling a stack. Agent Droplets come in two sizes, Pro at $50 a month and Team at $200 a month, with discounts of 15% and 20%, respectively, across all included resources. AI coding subscriptions often cap usage and charge per seat. DIY agent stacks require multiple vendors, multiple invoices, dozens of pricing units, and a pile of glue code to hold them together. Both approaches are complex, and their costs are hard to predict. Agent Droplets bring essential services together in one plan, simplifying adoption and making spend easier to predict. An AI builder on Pro or a startup team on Team pays one monthly subscription and runs as many agents as the work needs, with resource usage covered by the plan?s allowance. Customers choose whether spending stops at the plan price or continues at list prices once the allowance is used. This gives teams a defined starting budget and an explicit choice about additional spend as workloads grow. As companies move from a first agent to production workloads, DigitalOcean provides elastic scale across the agentic primitives their applications depend on: isolated execution, Serverless Inference, persistent memory, session storage, and governed tool access. Agent Droplets simplify how teams adopt and pay for these services, while the underlying platform lets resource consumption grow with workload demand, within their chosen spending controls. Companies can start small and scale on the same cloud. Developers choose an Agent Droplet size to access a bundle of essential agent services, reducing the work of selecting, connecting, and managing each technology separately. The relevant discount applies automatically to the included Managed Agents resources their agents use: Harness Runtime: dedicated microVM sandboxes that start in about a second, resume from a pause in about 300 milliseconds (in DigitalOcean testing), and pause automatically when idle, with memory and session storage included in the discount. Use them to scale agent execution as workloads grow or run arbitrary code generated by an agent. Inference Engine: DigitalOcean-hosted open models including Kimi K3 and GLM 5.3. Frontier models such as Claude and GPT remain available on a pay-as-you-go basis at list price. Action Gateway: governed access to more than 16,000 tools, including code interpreter, browser automation, web search, and web fetch, with credentials brokered outside the agent. Also included: unlimited agents, no seat charges, observability, and enterprise security (single sign-on, multi-factor authentication, role-based access, audit logs, cloud firewalls, and DDoS protection) on every size. New customers can start with a free trial: $5 in credit, no card required, enough to take an idea from zero to running this afternoon. To scale, Agent Droplet Pro and Team apply a bigger discount to every hour, token, and gigabyte included in the plan, 15% off on Pro and 20% off on Team. DigitalOcean also announced Inference and Agents Balance to make it easier for customers to prepay for their agents. Inference and Agents Balance is a prepaid balance, funded in amounts from $5 to $500, that only Serverless Inference, Action Gateway, and Managed Agents can use, while the rest of the cloud keeps billing as usual. Fund your Inference and Agents Balance directly, and your agents draw on it automatically once your Agent Droplet allowance is used up. Agent Droplets and Inference and Agents Balance are available in all regions where DigitalOcean Managed Agents is offered. Developers can create an Agent Droplet or add to their Inference and Agents Balance in the DigitalOcean Cloud Console, and agents already running on Managed Agents move onto an Agent Droplet without changes. (C) S&P Capital IQ - 2026

DigitalOcean debuts Managed Agents to streamline AI execution, inference, and tool access. Developing and scaling enterprise-grade AI agents has encountered a significant infrastructure bottleneck: the operational complexity of stitching together disparate compute sandboxes, credential management layers, and inference providers. While autonomous agents are expected to execute complex tasks such as diagnosing system errors, running simulations, or orchestrating multi-system workflows traditional cloud virtual machines and local development environments lack the native security isolation, low-latency persistence, and execution efficiency that agentic workloads demand. Addressing these developer friction points, cloud infrastructure provider DigitalOcean announced the release of Managed Agents. By bringing agent execution runtimes, governed tool integration, and serverless LLM inference onto a single integrated cloud platform, DigitalOcean eliminates the need for software engineering teams to manually build and maintain custom sandbox environments. The news: Hardware-Isolated sandboxes, governed tool gateways, and billing optimization. The core technological advance of DigitalOcean Managed Agents centers on providing purpose-built primitives specifically designed for agentic workflows rather than retrofitting general-purpose virtual infrastructure. Key capabilities introduced in the public preview launch include: Hardware-Isolated Harness Runtime: Executes agent sessions within dedicated Firecracker microVM sandboxes, providing hardware-level isolation for executing generated code, installing dependencies, and running background processes safely. Action Gateway Tool Governance: Provides agents with access to over 16,000 tools and Model Context Protocol (MCP) servers, brokering API credentials at execution time outside the model's context window while enforcing human-in-the-loop approval guardrails for sensitive actions. Sub-Second Pause and Resume Persistence: Features an auto-pause mechanism that snapshots session state (files, memory, and processes), resuming execution in roughly 300 milliseconds when new tasks are triggered. Active CPU Usage Pricing Model: Charges developers strictly for active compute cycles, pausing CPU and memory fees while agents wait for model inferences or external tool responses. Native Ecosystem Integration: Connects directly with DigitalOcean's Serverless Inference Engine - supporting over 75 models - as well as managed databases and object storage. Transforming the Enterprise Cloud & AI infrastructure industry. DigitalOcean's launch of Managed Agents marks a major structural shift across the Enterprise Cloud Infrastructure, AI Development Tools, and Cloud PaaS sectors. The Obsolescence of "Stitched-Together" Agent Sandboxes For the past two years, developers building agentic AI applications were forced to assemble custom stacks purchasing compute sandboxes from one vendor, credential brokers from another, and inference tokens from a third. This multi-vendor approach introduced severe operational friction, high latency, complex billing structures, and security vulnerabilities due to scattered credential handling. DigitalOcean's release accelerates the phase-out of piecemeal agent setups. The cloud platform industry is entering an integrated AI-native cloud era. Infrastructure providers are no longer evaluated simply on raw compute pricing or virtual machine provisioning speeds, but on whether they deliver unified execution sandboxes with native credential security and model access out-of-the-box. Setting New Benchmarks for Consumption-Based Compute Economics Traditional cloud instances charge for continuous compute uptime regardless of whether an agent is actively executing code, waiting for LLM tokens, or paused for human approval. By stopping compute charges when agents idle and offering sub-second resume capabilities, DigitalOcean establishes agent-aware consumption economics as an industry benchmark. Cloud vendors must adapt their billing architectures to accommodate non-linear, bursty agentic execution patterns, lowering total cost of ownership (TCO) for enterprises deploying thousands of parallel agents. Broad operational impact on businesses operating in the cloud & AI sector. For Chief Technology Officers (CTOs), engineering managers, and AI application developers operating across the cloud infrastructure ecosystem, adopting a unified agentic runtime provides direct strategic advantages: Lower Total Cost of Ownership (TCO): Stop-the-clock compute fee discounts on model inference latency can reduce monthly cloud infrastructure costs by up to 37. relative to standalone sandbox services. Improved Security and Zero Credential Leakage: (Action Gateway) Brokered API keys in transit at runtime "in the Action Gateway" ensures sensitive corporate tokens are protected from the LLM prompts or agent memory contexts. Faster Time-to-Market for Agentic Products: agency teams can skip weeks of building infrastructure and get to work by as cheap, deploy parallel coding agents, support triage automation, and workflow bots right away. Minimized vendor lock-in: enterprise applications can stay compatible across large software ecosystems as long as there's native support for open-weight models, generated container images (OCI), and open MCP protocols. By bringing execution environments, managed tool registry, and serverless inference under one cloud, DigitalOcean offers enterprise software teams the building blocks for securely scaling autonomous AI from isolated research projects to production- ready infrastructure.