Anthropic and Modal Unveil Next-Gen Managed Agents with Enhanced Sandboxes
- Gate
- Machine Learning
- Departs
- Crew
- John Rodriguez
- Passengers
- 773
The collaboration between Anthropic and Modal introduces Claude Managed Agents, offering developers scalable, cost-effective AI workflows with integrated sandbox environments.
Last month, Anthropic introduced Claude Managed Agents, a suite of composable APIs designed for constructing and deploying cloud-hosted agents at scale. This setup notably separates the agent loop from the code execution environment, enabling a more flexible architecture.
Today, Modal revealed its partnership with Anthropic, launching a robust integration that combines Modal Sandboxes with Claude Managed Agents. This integration allows users to execute all tool calls within a self-hosted Modal Sandbox, taking advantage of Modal's rapid cold-start capabilities, customizable images, various snapshotting techniques, and cost-efficient bursting mechanisms while retaining the convenience of Anthropic's agent loop management.
Architectural Separation for Enhanced Control
At Modal, the architectural approach of separating the agent loop from execution has been well-received. The Managed Agents operate as a fully hosted REST API, which means Anthropic handles the agent loop while managing session states, letting applications send events and stream results back seamlessly.
This clear distinction has started gaining traction among users, with many embracing the separation that Managed Agents enforce. The prevailing belief is that this setup enhances security boundaries and provides teams greater control over observability, failure management, and scalability.
Utilizing Modal Sandboxes with Managed Agents
Anthropic's managed sandboxes serve well for standard deployments, while Modal’s cloud platform is optimized for AI workloads, equipped with a custom container runtime and orchestration layer tailored for fast, flexible, and isolated computations. Here are some standout features:
- Fast startups: Modal Sandboxes allow for quick initialization, even on custom images, enabling developers to iterate faster by incorporating their complete dependency trees.
- Persistence options: Options such as directory snapshots, filesystem snapshots, and memory snapshots cater to diverse requirements for preserving context across multiple runs.
- Cost-efficiency: The burst pricing model means users only pay for actual compute usage, eliminating minimums and idle costs between tasks.
- Network security: Modal includes tools such as short-lived connect tokens, ensuring secured, authenticated access to running sandbox services without exposing them to public networks.
- Scalability: Users can operate upwards of 100,000 concurrent sandboxes, backed by generous resource configurations tailored for computational power, including CPUs, memory, and GPUs.
These features collectively enhance the infrastructure, allowing businesses to deploy and scale AI solutions more effectively.
Real-World Applications and Benefits
Several organizations are already harnessing the combined power of Claude Managed Agents and Modal Sandboxes. For example, DoorDash is developing an enhanced intelligence system with multiple AI agents across various channels, including chat, voice, and mobile apps. The aim is to simplify operations for their merchants. DoorDash is leveraging Modal as part of its AI infrastructure and is actively exploring Managed Agents to expand their intelligent system's capabilities.
“As we scale agentic commerce for local businesses, we need a highly efficient path to production with full harness control, scale, and reliability. We’re excited to evaluate Claude Managed Agents for this next step, building on our AI infrastructure with Modal.”
Similarly, Blend, an innovator in digital origination for banks, is utilizing the integration to drive agent-assisted triage. Given the complex needs of their banking clients, a simple support ticket can encompass intricate interactions across code, configurations, and user experiences. The integration of Claude Managed Agents and Modal Sandboxes simplifies the access to fully configured environments while maintaining security, allowing engineers to resolve issues more efficiently.
“Having a fully configured environment with the right tooling, context, and credentials already in place is what makes it possible for agents to help engineers reason across all of that faster.”
Starting with Claude Managed Agents and Modal Sandboxes
In combining Claude Managed Agents with Modal Sandboxes, businesses gain substantial advantages in scale, flexibility, and security for their production deployments. Developers eager to get started can access two reference examples — a command-line interface (CLI) agent and a Slack bot — specifically designed to operate on the Modal and Managed Agents framework. By cloning the examples and integrating their images and tools, they will be well on their way to establishing effective production deployments. Additionally, Modal offers $30 of free compute monthly, making it accessible for teams to explore their capabilities.
As organizations look to implement AI-focused workflows, the integration of Claude Managed Agents with Modal Sandboxes offers a powerful, secure, and efficient infrastructure, paving the way for advanced AI operations across various sectors.
