The West Coast Builders — coding-agent collection
OpenClaw Creator built App to Control Claude Code
Eight selected interviews on harnesses, multi-agent workflows, context, and review
Eight selected conversations with Jinjing Liang, Gil Feig, Ben Vinegar, Thorsten Ball, Li Yin, Quinn Slack, Mario Zechner, and Peter Steinberger from Mayank Gupta's broader interview playlist. Topics include agent harnesses, parallel worktrees, inspectable session history, review, architecture, model economics, and durable context. Product claims and forecasts remain the guests' accounts, not independently verified results.
In this collection Browse 8 summaries 8 of 8
Developer Peter Steinberger joins host Mayank Gupta to discuss his Claude Code workflow and demonstrate VibeTunnel. The recording does not demonstrate or discuss OpenClaw despite the current video title.
Key Points Covered
- Detailed voice prompts reduce ambiguity: Steinberger uses Wispr Flow to dictate long prompts. He repeats goals and constraints from several angles so Claude Code has more context for a one-shot implementation 01:52–03:53(opens in a new tab).
- Concurrent agents still need boundaries and review: Steinberger runs two to four agents in one shared directory, deliberately choosing low-overlap areas. He then reviews diffs in Tower, tests, commits, or reverts. This personal workflow does not establish that shared-directory concurrency is generally safe 03:53–06:55(opens in a new tab).
- The throughput claim is a personal report: Steinberger says he built a live analytics site with Supabase and Postgres in about four hours while doing other work. But the recording supplies no benchmark, preserved comparison, or independent quality assessment 08:39–11:40(opens in a new tab).
- CLI constraints can sharpen focus: He values Claude Code's text-only interface while keeping an IDE nearby for reading files and diffs when needed 12:42–13:47(opens in a new tab).
- External research remains part of implementation: When an agent struggled with a virtual keyboard, Steinberger researched newer browser APIs and found detailed documentation. He passed its URL back to Claude Code rather than expecting the model to solve an unfamiliar domain unaided 13:47–14:50(opens in a new tab).
- VibeTunnel preserves a local workflow: The project began from wanting to check and steer long-running local agents from a phone without moving Mac development into a cloud pull-request workflow 23:38–25:33(opens in a new tab).
- The beta demonstration exposes unfinished behavior: VibeTunnel mirrors terminals and accepts remote input. But Steinberger explicitly shows broken scrolling and describes monitoring features that are not finished yet 25:33–28:48(opens in a new tab).
Full video: https://www.youtube.com/watch?v=fu7th5HiADo(opens in a new tab)