Key Moments
Why Every Agent needs Open Source Cloud Sandboxes
Latent Space PodcastKey Moments
E2B provides open-source cloud sandboxes for AI agents, enabling complex tasks like data analysis and code execution, and is evolving into a comprehensive AI cloud platform.
Key Insights
E2B offers open-source cloud sandboxes that act as a runtime environment for AI agents, enabling them to execute code, analyze data, and perform complex tasks.
The sandboxes have evolved from a simple code interpreter to a general-purpose 'dev box' for AI, supporting various use cases like data analysis, code generation, and research.
E2B experienced significant growth, moving from 40,000 sandboxes in March 2024 to 50 million in March 2025, indicating a strong market pull for AI infrastructure.
The company emphasizes a strong developer experience (DX) and aims to be the 'Kubernetes for agents' with a focus on ease of use and generality.
E2B is expanding its offerings beyond basic Linux sandboxes to include features like GPU support and potentially Windows environments for broader AI applications.
The platform supports LLM training and reinforcement learning, as demonstrated by its use in Hugging Face's Open R1 project, offering a cost-effective alternative to GPU clusters.
E2B is exploring more advanced features like sandboxing for reinforcement learning, computer vision tasks, and potentially forking and checkpointing to support complex agentic workflows.
FROM DEV TOOLS TO AI SANDBOXES
Vasek Mlejnsky, co-founder of E2B, shares the company's origin story, stemming from an interest in developer tools and interactive documentation. Their initial product, DevBook, featured interactive playgrounds built on a sandbox infrastructure. This technology, though unscalable, became the foundation for E2B. A pivot occurred in early 2023 with the advent of improved LLMs like GPT-3.5, leading them to focus on sandboxes as essential environments for AI agents to run code and automate tasks.
THE EXPLOSION OF AI AGENT USE CASES
E2B's sandboxes have rapidly evolved to support a wide array of AI agent capabilities. Initially envisioned for code interpretation and data analysis, they now facilitate deep research, code generation, and even more complex tasks like reinforcement learning and computer vision. The platform's versatility allows for a generalized runtime environment, akin to a 'dev box' for AI, enabling agents to perform tasks that mirror human actions on a computer, such as creating files, analyzing data, or writing applications.
MASSIVE GROWTH AND SCALABILITY CHALLENGES
The demand for E2B's sandboxes has been astronomical, with usage soaring from 40,000 sandboxes in March 2024 to 15 million in March 2025. This exponential growth highlights a critical infrastructure gap in the AI ecosystem. E2B positions itself as an 'LLM OS' company, providing the underlying infrastructure that agents leverage. The company emphasizes a superior developer experience (DX), aiming to be the intuitive and accessible 'Kubernetes for agents'.
BEYOND CODE INTERPRETATION: A GENERAL RUNTIME
E2B's sandboxes are more than just code interpreters; they function as a general runtime for LLMs and agents. This means they can handle diverse tasks, from simple data transformation and analysis to complex operations like creating spreadsheets or running entirely new applications. The platform treats the sandbox as a general-purpose machine, capable of running various programming languages and even enabling agents to start servers accessible from the internet, thus offering comprehensive utility similar to any developer's workstation.
SUPPORTING AI DEVELOPMENT AND TRAINING
E2B plays a crucial role in the AI development lifecycle, from model training to agent evaluation. Hugging Face utilizes E2B sandboxes for the reinforcement learning phase of their Open R1 model, enabling massive parallelization of training steps and offering a cost-effective alternative to expensive GPU clusters. The sandboxes provide isolated environments, mitigating risks associated with LLMs altering cluster permissions and ensuring secure, efficient training operations.
ADVANCED FEATURES AND FUTURE HORIZONS
Looking ahead, E2B is focused on expanding its capabilities to include GPU support for more demanding AI tasks, and potentially Windows environments for broader compatibility. The platform is also developing advanced features like forking and checkpointing to support complex agentic workflows, such as parallel problem-solving and tree searches. Ultimately, E2B envisions itself as the future AWS for LLMs, providing elastic and comprehensive cloud infrastructure for AI development and deployment.
NAVIGATING THE AI INFRASTRUCTURE LANDSCAPE
E2B positions itself as a generalist sandbox provider, catering to AI engineers rather than traditional infrastructure engineers. While competitors might focus on specific aspects like browser emulation or custom Python sandboxes, E2B offers broad utility. Educating the market on the potential of these sandboxes has been key, moving from specific use cases like code interpretation to demonstrating the full spectrum of possibilities for AI agents, fostering innovation and adoption.
THE COMPLEXITY OF PRICING AND BILLING
The transition to usage-based billing, common in infrastructure, presents E2B with challenges, particularly in accurately pricing diverse workloads. While starting with simple pricing models is appealing, the reality of scaling involves handling unpredictable usage patterns and resource demands. E2B is evaluating third-party billing providers like Orb and Metronome to manage this complexity, aiming to accurately account for compute, storage, and networking in their evolving pricing structures.
THE EVOLVING ROLE OF LLM FRAMEWORKS AND PROTOCOLS
The AI landscape is seeing the rise of various frameworks and protocols, like Langchain and MCP (Machine Computation Protocol). E2B integrates with these to enhance agent capabilities and interoperability. The company believes in providing foundational tools rather than opinionated frameworks, allowing developers flexibility. The MCP, while promising, still faces questions regarding its integration, particularly for remote servers and higher-order applications, with E2B exploring how to best support these emerging standards.
DISTINGUISHING HUMANS AND AGENTS ONLINE
As AI agents proliferate, the distinction between human and agent traffic online becomes critical. E2B notes the significant increase in agent-generated traffic versus human traffic, highlighting the need for websites to adapt. While efforts like LLM.txt aim to improve LLM legibility, E2B suggests a dualistic approach, envisioning separate human and agent-facing internet experiences to manage incentives and user experience effectively.
THE MIGRATION TO SAN FRANCISCO
Relocating to San Francisco was a strategic move for E2B, driven by the city's emergence as an AI hub. This proximity to users facilitated 'collision installations'—deeply embedding with early customers to rapidly iterate and refine the product based on direct feedback. While building a dev tool company from anywhere is possible, proximity to a concentrated user base accelerated E2B's understanding of user needs and product development, especially in the early, highly iterative stages.
RECRUITING TALENT & FUTURE GROWTH
E2B is actively hiring for roles in distributed systems, platform engineering, AI engineering, account management, and customer success. The company highlights the wealth of talent in Europe, particularly in the Czech Republic, but also emphasizes the strategic importance of its San Francisco presence for customer relations and sales. With a clear roadmap and strong market momentum, E2B aims to accelerate its growth, viewing its current position as a spark it wants to fan into a fire.
Mentioned in This Episode
●Software & Apps
●Companies
●People Referenced
E2B Sandbox Usage Growth
Data extracted from this episode
| Time Period | Number of Sandboxes |
|---|---|
| March 2024 | 40,000 |
| March 2025 | 15 million |
Common Questions
E2B originated from Vic and Thomas's experience with DevBook, an interactive documentation tool. After facing burnout and seeing the potential of GPT-3.5, they pivoted to explore AI agents and developed E2B, initially focusing on automating their own work with developer sandboxes.
Topics
Mentioned in this video
An earlier project by Vic and his co-founder, focused on interactive documentation for developers, which laid the groundwork for E2B's sandbox technology.
A database toolkit for which DevBook built an interactive playground, showcasing early sandbox technology.
An AI model with a large context window, used in an early demo that later showed decreased capability when re-tested.
Another usage-based billing provider that E2B is considering.
The AI model released at the time that inspired Vic and Thomas to pivot their focus and explore AI agent development.
A programming language that models were initially very good at generating code for, making it a strong candidate for code interpreting use cases.
A usage-based billing provider that E2B is discussing using.
A company focused on providing cloud sandboxes for AI agents.
A company that offers custom Python sandboxes, mentioned as a point of comparison in the LLM OS landscape.
A promising new agent framework that is TypeScript-first.
A model developed by Hugging Face that utilizes E2B for codegen reinforcement learning.
A protocol supported by E2B's desktop SDK, enabling human-in-the-loop interaction with sandboxes.
A service whose representative is noted as the only one implementing a 'bring your own key' solution for LLMs.
Amazon Web Services, a major cloud provider whose complex pricing models are contrasted with E2B's initial goals.
An orchestration system used as an analogy for E2B aiming to be the 'Kubernetes of the world for agents' but with better DX.
A popular agent framework with millions of downloads, despite some developers questioning its popularity.
University from which emerged the AI App generation comparison work using E2B.
Google Cloud Platform, another major cloud provider with complex pricing.
A tool mentioned as being used by individuals who might have otherwise used E2B for building projects.
The continent from which E2B originated, with the potential to build a dev tool company successfully.
The specific Linux distribution used in E2B sandboxes, customizable and Debian-based.
A platform that launched a marketplace credit system, which the speaker found exciting.
A company whose CEO, Matthew Prince, was interviewed about agent traffic and website statistics.
The operating system environment provided by E2B sandboxes, with possibilities for expanding to Windows.
A programming language that Mastra, a promising agent framework, is built with, aligning with E2B's focus on Product Developers.
Significant usage for E2B SDK downloads, particularly for building web applications.
A company offering tools for agent website navigation, with its Stagehand feature being highlighted as elegant.
Payment processing company whose usage-based billing features E2B has used, finding them sometimes 'rough around the edges'.
The company whose crawl-to-referral ratio is significantly higher than traditional search engines, indicating heavy API usage.
A company whose insights on cloud infrastructure pricing and principles are referenced.
A company that transitioned from tier-based to usage-based billing, illustrating the challenges and importance of accurate infrastructure pricing.
A company using E2B sandboxes for their Open R1 model's codegen reinforcement learning, as well as for evals.
Company where Allesio is Partner and CTO.
Company that launched a computer vision model, which prompted E2B to realize the potential of their desktop sandbox.
Company associated with Henry, who discussed MCP concepts.
The city where E2B relocated its team to be closer to the AI hub and its users.
The country where E2B initially recruited talent, and where they have re-opened an office.
Mentioned as a country where Vic and Thomas worked from initially.
The city where Vic and Thomas initially worked, and where E2B has re-established an office, attracting talent.
The country where E2B's team relocated, emphasizing the importance of a US base for sales and customers in the B2B space.
More from Latent Space
View all 85 summaries
86 minNVIDIA's AI Engineers: Brev, Dynamo and Agent Inference at Planetary Scale and "Speed of Light"
72 minCursor's Third Era: Cloud Agents — ft. Sam Whitmore, Jonas Nelle, Cursor
77 minWhy Every Agent Needs a Box — Aaron Levie, Box
42 min⚡️ Polsia: Solo Founder Tiny Team from 0 to 1m ARR in 1 month & the future of Self-Running Companies
Found this useful? Build your knowledge library
Get AI-powered summaries of any YouTube video, podcast, or article in seconds. Save them to your personal pods and access them anytime.
Try Summify free