Build to Launch follows Jenny Ouyang as she tests AI tools on real work, keeps the receipts, and turns each result into a system you can use. Episodes cover AI agents, Claude Code, vibe coding, automation, and the practical work between an idea and a shipped result.
Reading the feed…
OpenAI folded Codex into the ChatGPT desktop app, so I handed it four real jobs I hadn't been able to finish reliably in Claude: editing my first YouTube video, turning that video into an automated podcast, making my short-form Notes system visible enough to debug, and finding and restoring a task I had buried by accident. This audio edition walks through what each job produced, the interface details that made me stay, and why I kept both ChatGPT and Claude on one shared system underneath instead of switching. The full written build is in the companion Build to Launch article.
I turned my Obsidian vault into a connected memory layer Claude can read and update. This audio edition explains why Markdown, WikiLinks, properties, and Obsidian's index matter; how a 4,663-file test ran about 60 times faster through the index; and what changed when Claude could retrieve decisions, link new work, and write results back. The full implementation stays in the companion Build to Launch article.
I gave Hermes Agent three real jobs: research, a 47-task backlog, and a broken tool setup. This episode covers what kept working after I stepped away, where the run failed, and the three rules I now use before handing Hermes a bigger job.
