Lenny Pruss emigrated from the Soviet Union as a refugee and grew up in Silicon Valley with engineer parents who transitioned into tech, giving him an early view of technology as an engine of upward mobility, while his pivot from computer science at UC Berkeley to venture capital left him with deep respect for engineering combined with a focus on exploring the future through investing. He deliberately focused on infrastructure over consumer apps early in his career, specializing in cloud, DevOps, data infrastructure, and developer tools, and has become known for backing what he calls the lunatic fringe—deeply technical founders who are obsessed with solving problems they've felt viscerally, emphasizing that he seeks builders willing to spend a decade on solutions rather than chasing traditional business signals.

• A strong conviction that the software stack for AI agents will fragment into specialized components rather than staying as all-in-one platforms. He compares the current state of AI agents to the early days of cloud computing, arguing that while vertical platforms like Heroku emerge first, the market eventually shifts toward distinct building blocks for storage, orchestration, and messaging.
• Interest in deep infrastructure startups that rebuild core technologies from scratch to leverage modern hardware. He highlights companies like CedarDB for reimagining database layers for modern multicore systems and TigerBeetle for using the Zig language to handle memory allocation differently than legacy SQL systems.
• Focus on reliability and rigorous testing as the critical bottleneck in software development. He frequently discusses Deterministic Simulation Testing and tools like Antithesis, arguing that as AI makes writing code cheaper and faster, the real value shifts to verifying that the code actually works and is bug-free.
• Appreciation for durable execution and workflow orchestration as essential infrastructure for AI agents. He points to Temporal as a prime example, noting that successful agents require production-grade plumbing to handle retries, state management, and long-running processes without failing, rather than just clever prompting.
• Admiration for engineering craftsmanship and first-principles thinking. He praises specific internal methodologies, such as coding standards that require multiple assertions per function and designing for performance before profiling, rather than relying on legacy assumptions.