How Effective AI Runs Secure, Multi-Agent Insurance Workflows at Scale

Effective AI helps insurance organizations collectively writing more than $180 billion in direct written premiums launch and maintain insurance products. Its platform uses specialized agents for market benchmarking, rate-plan design, actuarial analysis, regulatory filings, and compliance workflows.

As those agents moved beyond document analysis and began producing the actual models, spreadsheets, and files insurance teams rely on, Effective AI needed a secure execution layer that could support complex, customer-specific workflows at scale. With E2B, the company supports more than 10,000 user sessions each month.

‍

“E2B gives Effective AI a secure execution layer at large scale, turning complex filing, pricing, and compliance work into faster product launches.”

— Arijit Banerjee, Co-Founder, Effective AI

The challenge: moving from insurance intelligence to finished work

Turning analysis into usable outputs

Early versions of Effective helped insurance teams reason over filings, manuals, and regulatory documents. As customers brought more of the insurance product lifecycle onto the platform, they needed agents that could produce the artifacts required to act on those decisions.

That meant building actuarial models, generating spreadsheets, constructing rating engines, transforming files, and running customer-defined workflows. Effective AI needed an environment where agents could work with files and dependencies, connect to approved systems, and maintain state across complex workflows.

Deploying AI at scale in a highly-regulated industry

Effective serves insurance organizations with strict compliance requirements. Workflows may handle sensitive policyholder information, proprietary pricing models, claims histories, and loss and exposure data. Each agent workload must therefore run in isolation and receive only the files, credentials, tools, and systems it is authorized to access. 

Scaling for variable, multi-agent workloads

Effective AI’s workloads vary dramatically. A table-extraction task may finish in seconds, while a more complicated workflow like building a rating plan, comparing competitor filings, or running a portfolio analysis, may take up to 16 hours. Individual rater-building workflows may even require hundreds of agents running across tens of sandboxes, making low-latency handoffs between agents essential. Other use cases include long-lived sessions that wake up daily or weekly to perform recurring tasks, such as monitoring for new filings that match specific criteria.

“Startup time matters because our workflows are not one agent doing one task. A single customer workflow can fan out across hundreds of agents, and even small delays compound quickly.”

 - Arijit Banerjee, Co-Founder, Effective AI

‍

‍

Staying focused on insurance

Building an internal sandbox platform would have required Effective AI to solve isolation, provisioning, lifecycle management, observability, and scaling, pulling engineers away from building the features that differentiate the product.

‍

The solution: a secure execution layer built on E2B

Effective built its execution layer on E2B from the start. Clear documentation helped the team get up and running quickly, while fast sandbox startup times and the option to self-host gave Effective additional confidence in E2B.

‍

"We got started and never looked back."

— Arijit Banerjee, Co-Founder, Effective AI

‍

From a request to an executable workflow

A user submits a request in natural language on Effective - say, build a rating plan, compare competitor filings, or run a portfolio analysis. Their orchestration layer then decomposes it into specialized tasks handled by parallel agents: filing research, table extraction, rater construction, and actuarial calculations.

Each session gets an E2B sandbox booted from a pre-warmed execution environment. Effective uses E2B templates to define a base execution environment with pre-installed libraries and common insurance agent dependencies. A periodic build job layers each customer’s specific tools, code, and context onto that base template, then saves the configured environment as a snapshot. New sessions boot from the customer’s snapshot and retrieve only the code changes made since the snapshot was created.

Agents then execute code inside the sandbox and coordinate through Effective’s custom multi-agent runtime. When agents need to work from the same files, dependencies, or in-progress outputs, they share a sandbox; otherwise, the runtime uses a Jupyter PreforkProvisioner to launch isolated agent kernels from a preloaded daemon with copy-on-write memory sharing, reducing per-notebook overhead and allowing hundreds of agentic functions to run efficiently on the same machine.

E2B’s pause-and-resume feature lets Effective preserve a workflow’s state without keeping its sandbox running continuously, supporting thelonger-running and recurring sessions.

‍

“A generic sandbox is useful, but our customers need agents that work inside their world. Templates let us start each workflow with the customer’s tools, dependencies, and context already in place.”

 - Arijit Banerjee, Co-Founder, Effective AI

‍

Extending security controls into agent execution

Each workload runs in its own hardware-isolated E2B MicroVM, creating a strong boundary between tenants and preventing customer code, data, and agent activity from crossing into another customer’s environment. Sandboxes are treated as ephemeral execution environments: only the files and data needed for a workflow are loaded into the sandbox, and the entire environment is destroyed when the workflow is complete, with persistent files stored in Effective’s platform. 

‍

Keeping engineers focused on the product

By building on E2B, Effective avoided creating and operating its own sandbox infrastructure- saving the equivalent of one to two full-time engineering hires.

Instead, the team can focus on embedding insurance expertise into the platform and improving how customers research markets, develop rating plans, validate filings, and operationalize their underwriting and actuarial workflows.

The results

Today, Effective AI:

• Scales beyond 1,000 concurrent E2B sandboxes supporting complex workflows across hundreds of agents.
• Gracefully handles spiky workloads from workflows that wake up daily or weekly 
• Saved an estimated 14% of engineering capacity who would otherwise have been dedicated to building and operating a secure sandboxing platform.

Looking ahead

Effective AI is expanding into even more specialized workflows across underwriting, forms generation, generalized linear modeling, regulatory filing preparation, and submission support.

As its agents take on more of the technical work behind insurance product development, E2B provides the secure compute foundation needed to turn customer data, code, and institutional knowledge into finished work that insurance teams can review and use that will grow with Effective AI’s platform.

‍