Developer Voices

Julien Verlaguet, a software engineer and programming-language designer, co-created Hack at Facebook and founded Skip Labs. He explores type-checking 30 million lines of PHP, the architecture behind Hack’s lightning-fast language server, and the concurrency bugs that challenged it. The conversation then turns to Skip’s immutable, reactive language, persistent state, garbage collection, TypeScript bindings, and Skipper, an AI coding agent.

Hannes Mühleisen, co-creator of DuckDB and co-founder of DuckDB Labs, is a database researcher and engineer. He walks through making DuckDB extensible to dodge unwanted PRs, why they rewrote SQL parsing with a PEG-based pluggable parser, the design of the Quack wire protocol, and the DuckLake approach to metadata. He also explains why DuckDB Labs joined AWS and how governance and resources changed.

James Brown, an engineering lead at Schroders who builds multi-agent tooling like the Clair plugin. He tells the Claude mania story and how agents reshaped his workflow. He explains Clair’s git shadow-branch trick for agent proximity. They discuss AI amplifying org problems, faster prototyping, shifting team skill needs, risks to junior hiring, dark factories, and looming burnout.

Lalith Suresh, CEO of Feldera and engineer applying DBSP theory to real systems. He explains Z-sets, the four DBSP operators that let any SQL query be compiled into incremental form, and how joins and deltas are handled. They cover origins in network control planes, building a change-centric engine, backfills and limits, and how this differs from stream processors. Try it at the open-source sandbox.

Henry Garner, CTO of Juxt and creator of the Juxt AI Radar, focuses on AI-assisted software development and the Allium behavioral spec language. He discusses the AI tooling adoption arc, orchestration frameworks versus agents, neurosymbolic approaches and the surprising return of Prolog. He also covers Allium’s role in giving LLMs structure and practical trade-offs in fine-tuning, observability and running models.

Marcin Kulik, open-source creator of Asciinema and maintainer of its recorder, player, server, and Rust AVT emulator, joins to trace 14 years of rewrites. They cover terminal parsing and emulation, trade-offs in rendering and performance, moves from Python and ClojureScript to Rust/WASM, live streaming architecture, and how a single Rust emulator now powers multiple components.

Tyler Cloutier, software engineer and founder of SpacetimeDB and BitCraft, builds a distributed in-memory DB that runs WebAssembly and borrows game architecture for real-time apps. He discusses launch-day failures and rapid fixes. He explains single-threaded L1-first performance, reducer-based hot-swapping, and how TypeScript and React hooks make real-time multiplayer web apps easier.

Glauber Costa, a software engineer and entrepreneur with vast experience in databases and kernel development, dives deep into his journey from contributing to the Linux kernel to co-founding a new venture focused on reimagining SQLite with Turso. He discusses the challenges of SQLite's closed contribution model and how Turso, a complete rewrite in Rust, overcomes limitations with features like vector search and CDC. Glauber shares insights on community growth, the importance of a healthy ecosystem, and the ambitious goal of integrating concurrent writes into embedded databases.

Christina Lin, a Google engineer pioneering the Agent Development Kit (ADK), delves into the transformative potential of AI in production systems. She explains the nuanced distinction between agentic pipelines and basic LLM calls. Christina discusses ADK's role in simplifying the construction of agents and tools, emphasizing the importance of grounding for accurate, cited results. Additionally, she highlights the Agent-to-Agent (A2A) protocol for enhancing communication and scalability in distributed systems while addressing crucial deployment and debugging strategies.

Mohammed Aboullaite, a backend engineer at Spotify, dives into the world of continuous profiling and observability using eBPF. He explains how eBPF allows for monitoring at the kernel level without recompiling, making it a game-changer for managing large-scale systems. Listeners learn about the evolution of eBPF, how to handle data volumes from profiling, and practical tools like Pyroscope and Pixie. Mohammed also discusses the low overhead of eBPF and the importance of mapping kernel events back to code, providing invaluable insights for optimizing performance in production.