Henry Dowling

Dowling studied applied mathematics and computer science at Harvard. He joined Bubble’s early team in 2020 and worked there until 2024, mostly on AI. He joined South Park Commons in 2024 and co-founded Stash with Samuel Liu. His account of Stash describes capturing transcripts from agent runs and distilling them into information that can improve later work.[1][2]

Make contextual autocomplete work outside one editor

Dowling and Samuel Liu built Codeswitch, a macOS autocomplete experiment. His technical account explains how accessibility APIs, screen recognition and a browser extension collect context, while a small overlay places suggested text near the cursor. Inserting an accepted suggestion uses clipboard and keyboard behavior, with a virtual clipboard to preserve previous contents. Google Docs required separate handling because its text is rendered through a canvas.[5]

Turn a spoken interaction into working documents

His Caplog project connects a real-time voice agent to file operations. Its repository provides a browser microphone path through WebRTC and an optional Twilio phone path. A backend tool creates or appends Markdown files, letting a voice session produce documents that can be used after the conversation. The project’s setup separates the browser interface, voice gateway and file-writing service.[6][7]

Dowling’s short coding-agent checklist collects practices for removing needless code and tests. He credits Dillon Mulroy’s anti-slop linting, Lopopolo’s test-deletion approach, a deletion pass suggested by Secondmindsys and complexity goals credited to Joseph Viviano. His own contribution is the synthesis and argument that an agent can preserve a bad test as if it were an authoritative specification. The post recommends examining that assumption before adding more machinery to satisfy it.[4]