<div id="readability-page-1" class="page"><div><figure><div data-video-poster-cover="true" aria-hidden="false"><img alt="" draggable="false" loading="lazy" width="220" height="60" decoding="async" data-nimg="1" src="https://e2b.dev/images/use-cases/logos/rogo.svg?dpl=dpl_3hVnKMvy6wnJW731vZkYhwrafznw"></div></figure>
<h2 id="state-of-the-art-orchestration-on-top-kernel-level-isolation-and-10000-to-15000-concurrent-sandboxes-underneath">State-of-the-art orchestration on top, kernel-level isolation, and 10,000 to 15,000 concurrent sandboxes underneath</h2>
Rogo is an AI platform for financial institutions, helping bankers and investors execute complex research, analysis, and deal workflows. Their customers include hundreds of leading firms globally, whose teams send tens of thousands of queries to AI agents every day.
Two constraints shape everything Rogo builds. The work is heterogeneous: a question might need a narrow specialist or a broad generalist, and routing it wrong degrades the answer. And the customers are financial institutions, so any code an agent generates has to run in a sandbox with full auditability, traceability, and per-session isolation. Those two constraints map onto two layers: the agent harness, and the execution layer underneath it.
<h2 id="two-layers-claude-managed-agents-and-e2b">Two layers: Claude Managed Agents and E2B&nbsp;</h2>
Claude Managed Agents allow Rogo to configure specialist and generalist agents. Rogo was drawn to Managed Agents for its orchestration and post-training capabilities. Because Anthropic develops both the harness and the underlying models, the platform can benefit from improvements to Claude's capabilities without requiring teams to maintain every aspect of orchestration themselves.
Rogo built its own agent harness for finance-specific workflows. Evaluating it alongside Claude Managed Agents highlighted how the two approaches could complement one another, with a natural division between finance-specific logic and general-purpose orchestration. E2B is the execution layer. When an agent generates code, it runs inside an isolated E2B sandbox.
The flow is straightforward. A user asks an agent for a workflow to be done. Rogo routes it to a generalist or a specialist. Claude Managed Agents is designed to handle the agent flow. E2B provides an isolated container for any code execution. Artifacts and the final answer stream back to the user.
The result is managed-agent convenience without giving up isolation, control, or speed.
<blockquote>“With Claude Managed Agents, we easily configured state-of-the-art model orchestration, and with E2B, we got the trust and isolation our clients require.”</blockquote>
- Alex Wang, Applied AI, Rogo.
<h2 id="the-harness-specialists-and-generalists-without-the-plumbing">The harness: specialists and generalists without the plumbing</h2>
Rogo uses both specialist and generalist agents — specialists for highly scoped tasks with a clear deliverable, and generalists for open-ended work. A single agent might carry both general Q&amp;A and the ability to build a full information memorandum, the roughly 90-page document that covers everything about a company. Bundling unrelated capabilities into a single agent, however, deteriorates overall performance.
With Claude Managed Agents, Rogo can easily configure dedicated agents that each know only what their task requires. For example, the CIM (Confidential Information Memorandum) builder — the builder for the roughly 90-page document covering everything about a company for potential buyers — needs only the context to build that document. Scoping data sources, tools, and abilities to a single task is straightforward out of the box, as is setting system instructions per agent — and Rogo plans to maintain both approaches for the foreseeable future.
Scoping agents to a single task also makes parallelization possible, which can be particularly useful for complex workflows. For example, creating a CIM is a high-stakes process with multiple sections, each requiring different types of analysis. Rather than forcing all of that context into a single agent, the work can be split into focused subtasks, such as separate agents for the market narrative and financial analysis. This helps manage complexity and can improve the quality of the final output.
<blockquote>"Claude Managed Agents allows you to abstract away the harness, so you get all the benefits of a state-of-the-art harness for free."&nbsp;</blockquote>
- Alex Wang, Applied AI, Rogo.
‍
Offloading context management, error recovery, and graceful failure lets Rogo's team spend its time on the parts specific to finance, not on the orchestration plumbing that every agent product has to rebuild.
<h2 id="the-execution-layer-isolation-scale-and-rapid-startup">The execution layer: isolation, scale, and rapid startup</h2>
The harness decides what runs. E2B decides where it runs, and for regulated industries, that is the part that has to be right.
Per-session isolation, by default. Any code an agent generates runs in its own E2B sandbox with full auditability and traceability. E2B runs each sandbox as an isolated MicroVM, so isolation is enforced at the kernel level rather than by a shared container. Teams with stricter perimeters can run the same setup inside their own cloud through BYOC; Rogo chose E2B's managed infrastructure to avoid operating it themselves.
<blockquote>"We work with clients from hundreds of financial firms, so per-session isolation is a hard requirement."</blockquote>
- Alex Wang, Applied AI, Rogo.
‍
Scale that holds in production. With tens of thousands of users firing concurrent queries, Rogo needs to spin up sandboxes on the order of 10,000 to 15,000 at once. Doing that without investing in their own infrastructure was the deciding factor.
<blockquote>"E2B offers the flexibility to scale in production that is tough to match."</blockquote>
- Alex Wang, Applied AI, Rogo.
‍
Networking and streaming that other providers didn't have. E2B gave Rogo native support for domain-based network allowlists and pre-signed URL streaming. The allowlists keep agent network access scoped to approved domains; pre-signed URL streaming returns artifacts to the user without routing data through extra hops.
Startup time, because the product is latency-sensitive. Rogo pre-configures E2B templates with the scripts and Python packages that its agents need, so a sandbox comes up ready to run instead of installing dependencies on every run.
<blockquote>"On a latency-sensitive product, that startup time makes all the difference. It helps craft a magical experience for our end users, who go from problems to solutions in no time."</blockquote>
- Alex Wang, Applied AI, Rogo.
‍
<h2 id="security-is-the-through-line">Security is the through line</h2>
Rogo's customers are in highly regulated industries, and they delegate the cutting edge of AI to Rogo on the condition that it stays secure. That is the bar the pairing is built to clear: Claude Managed Agents simplifies how Rogo configures specialist and generalist agents, and E2B supplies the isolation for secure code execution.
<blockquote>"At Rogo, top-of-the-line AI infra built with security by design is paramount. E2B gives us just that."</blockquote>
- Alex Wang, Applied AI, Rogo.</div></div>

E2B Blog & Customer Stories

Vaclav Mlejnsky

State-of-the-art orchestration on top, kernel-level isolation, and 10,000 to 15,000 concurrent sandboxes underneath

Rogo Delivers AI for Financial Institutions with Claude Managed Agents & E2B

댓글