The West Coast Builders — coding-agent collection
How to Ship 100x Faster as a Developer?
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 2 of 8
Merge co-founder and CTO Gil Feig explains how his team varies agent autonomy with product risk: tighter controls for a mature enterprise API, more experimentation on newer products. In conversation with Mayank Gupta, he connects this workflow to shared company context and direct customer research. The title's “100x” is an aspiration, not a measured speedup.
Key Points Covered
- Autonomy should follow product risk: Feig says Merge gives agents tighter bounds on its mature enterprise product and more room on newer products with fewer customers and edge cases. His “100x engineer” and 100-worker framing is aspirational rather than measured; he reports personally managing roughly two or three concurrent tasks 06:58–09:27(opens in a new tab).
- Enterprise delivery retains layered checks: He describes approved company context, local skills, tests and linting, AI review, and final human review, arguing that customer obligations make unattended production changes unrealistic 10:21–12:56(opens in a new tab).
- Better models may absorb orchestration work: Feig assesses OpenClaw as a useful demonstration of local agents but says it was expensive and inaccurate; in his view, model improvements, with some orchestration, will resolve more of those limitations 14:40–16:21(opens in a new tab).
- Context combines synchronized and live data: Retrieval-augmented generation (RAG) supplies a model with information retrieved from stored data. Feig argues that syncing and indexing company data remains useful alongside live API calls. He proposes a shared context service so each agent platform does not have to build its own copy of company knowledge; Merge's governance and routing benefits are first-party product claims 17:20–20:41(opens in a new tab).
- Merge claims multi-model savings: Feig says Merge Fusion sends prompts to multiple models and uses a judge to select or combine responses, claiming 20–60% savings and equal-or-better results for selected combinations. The interview does not independently verify the savings, evaluations, or scale 22:39–24:21(opens in a new tab).
- Coding has moved up an abstraction layer: Responding to whether coding is solved, Feig compares natural-language programming with earlier transitions from assembly to higher-level languages and says architecture and product judgment still require human input 41:59–43:37(opens in a new tab).
- Customer conversations beat feature research alone: Feig recommends speaking directly with prospective users and reports that Merge conducted roughly 100 interviews before launch to uncover objections that product comparisons and AI research would miss 46:12–49:46(opens in a new tab).
Full video: https://www.youtube.com/watch?v=C6u5CBIXRf0(opens in a new tab)