Open Source Startup Podcast

The latest Open Source Startup Podcast episode has our co-hosts Robby and Tim in conversation with Neal Swaelens and Oleks Yaremchuk, 2 of the Co-Founders of runtime agent security company Manifold Security.
Manifold recently released Manifest, their open-access, graph-based supply chain intelligence tool for users to scan skills and plugins to uncover any potential supply chain risks.
In this episode, Neal and Oleks explain why AI agents are reshaping cybersecurity - shifting the focus from guardrails to runtime security.
As tools like Claude Code and Codex spread rapidly, companies often have little visibility into the agents, plugins, skills, and external assets employees are using, creating major supply chain and runtime risks.
Drawing on their experience building LLMGuard and leading security teams at Protect AI and Palo Alto Networks, they argue that runtime detection and response is still a wide-open market opportunity.
They also discuss what it takes to build in the crowded AI security space, where buyers now expect real products instead of roadmap promises.
The conversation highlights lessons from open projects like LLMGuard and Manifest, why reducing noise and false positives matters, and how open ecosystems can help establish trust and industry standards for securing AI agents and assets.

The latest Open Source Startup Podcast episode has our co-hosts Robby and Tim in conversation with Batuhan Taskaya, the founding engineer and current Head of Engineering at generative media cloud Fal. Fal is a developer platform that allows builders to develop and fine-tune models with serverless GPUs and on-demand clusters.
This episode explores how a small, highly technical team carved out a unique position in the AI boom by focusing on generative media - images, video, and audio - while most of the industry rushed toward language models.
Early on, they recognized that image and video models operate very differently from LLMs. With no strong API-first players in image generation, they started there and doubled down on building reliable, high-performance infrastructure for running these models in the cloud, leveraging deep expertise in systems and performance engineering.
Their strategy of embracing open-source models, then fine-tuning and optimizing them for real-world use cases, helped them quickly gain traction - growing from zero to $400M of revenue by 2026 and scaling rapidly as demand for generative media surged.
The conversation also dives into how the company evolved into a full-stack generative media platform, expanding from images into video and audio as those markets matured, especially with video seeing explosive growth in 2024–2025.
A key differentiator has been their relentless focus on inference performance, custom kernel optimization, and cost efficiency, which has driven strong customer retention. Rather than betting on a single model, they embrace rapid model turnover and ecosystem fragmentation, ensuring flexibility for developers and enterprises alike.
Looking ahead, the biggest challenges lie in scaling video models and securing enough compute capacity in a supply-constrained GPU market. Throughout, the story highlights the power of small, focused teams with clear strategy and the ability to pivot quickly in a fast-moving AI landscape.

In our latest Open Source Startup Podcast episode, co-hosts Robby and Tim talk with Pedram Amini, the creator of open source platform Maestro which allows users to run fleets of AI coding agents autonomously for long periods of time. Their project has 3K stars on GitHub.
The episode explores how Maestro's multi-agent system overcame a key limitation in generative AI: context overload. After juggling many Claude sessions for different tasks, Pedram realized each problem needed its own isolated workflow. Maestro turns this into a system letting users run many agents and tabs in parallel, keeping tasks separate and avoiding context degradation during long or complex work.
Maestro is designed for scale, enabling dozens or even hundreds of agents to handle complex projects simultaneously. It’s flexible, model-agnostic, and especially useful for breaking big problems into independent units. The project has quickly grown into a community-driven effort, reflecting a broader shift: instead of buying a bunch of tools, developers can build highly customized AI systems themselves, pointing toward a future of large-scale agent orchestration.

In our latest Open Source Startup Podcast episode, co-hosts Robby and Tim talk with Magnus Müller, the Co-Founder & CEO of Browser Use - the platform that makes web agents come to life. Their open source, browser-use, has almost 80K stars on GitHub and is widely adopted.
This episode dives into the unexpected rise of an open-source browser automation project that took off during Y Combinator - while many similar projects before and after it never gained traction. The founder reflects on why: delivering a “magical moment” fast. Early demos showing AI controlling a browser, inspired by trends like OpenAI’s Operator, and immediately clicked with people. What began as a developer-only Python library evolved into a hosted product as non-technical users - from sales teams to startups - wanted access.
Along the way, the team leaned into controversial but compelling use cases, like AI applying for jobs on your behalf, which sparked conversation and accelerated growth. The core challenge they focused on solving was reliability: unlike deterministic automation scripts, AI agents can behave unpredictably, making trust and repeatability central problems to overcome.
The long-term vision goes beyond UI automation toward agents that can skip the browser entirely and interact directly with website servers through structured actions. But the conversation isn’t just about infrastructure. The founder admits that early growth came mostly from building and talking to users, while recent months have been dedicated to storytelling and marketing rather than coding. A personal through-line emerges as well: learning to replace defensiveness with curiosity - questioning assumptions, staying open to feedback, and continuously refining both the technology and the narrative around it.

In our latest Open Source Startup Podcast episode, co-hosts Robby and Tim talk with Catherine Jue, Co-Founder and CEO of browser infrastructure company Kernel. Their open source images acts as a browsers-as-a-service for automations and web agents.
In this episode, we break down what Kernel is building today and why browser infrastructure has quietly become one of the most important layers for AI agents. We talk about Kernel’s focus on fast, low-latency cloud browsers, why performance matters more than people expect, and how developers can connect agents via APIs or MCP servers without spinning up heavy infrastructure themselves.
We also explore the real-world use cases driving adoption - from a new wave of RPA for industries without APIs, to real-time web analysis, sales intelligence, and voice agents that need to respond instantly. Finally, we dig into Kernel’s open-source, developer-first DNA, the technical bets behind its control plane and unikernel-based browsers, and why the team believes agentic workflows are still early, but inevitable.