Tutorial, Product Demonstration
How to Make Claude Code Your AI Engineering Team
- Gary Tan, President and CEO of Y Combinator, asserts the industry has entered the "agent era," where software is built using AI teams structured with human-like roles, processes, and reviews.
- Tan built GStack, an open-source framework that transforms AI coding assistants into a specialized engineering team, surpassing Ruby on Rails in GitHub stars within three weeks of release.
- Tan reports writing more code in the past two months than in all of 2013, attributing this productivity to the "thin harness, fat skills" architecture of GStack.
- Current AI coding limitations are defined not by model intelligence, but by the lack of scaffolding; without context, models "guess" and produce plausible code that "silently breaks."
- GStack includes a skill called Office Hours, a distilled version of YC partners' advisory sessions that forces founders to answer six reframing questions before coding begins.
- In a live demonstration, Tan used GStack to architect a tax app that extracts 1099 documents from Gmail, utilizing a "browser automation" strategy rather than Plaid or OAuth to avoid credential storage risks.
- The AI identified a wedge strategy for the tax app: aggregating documents ($2–$5/month) could evolve into a higher-margin lead generation funnel for tax preparers (percentage of transaction).
- GStack's Adversarial Review skill automatically subjected a design document to scrutiny, identifying and fixing 16 issues, thereby raising the design score from 6/10 to 8/10.
- Tan introduced Design Shotgun, a visual brainstorming tool that generates multiple UI variants (e.g., "Command Center" vs. "Friendly Progress") using OpenAI's image generation capabilities in approximately 60 seconds.
- Tan created a custom QA and Browse tool by wrapping Playwright and Chromium at the CLI level, enabling agents to perform complex browser interactions, screenshot analysis, and regression testing.
- GStack aims to elevate developers to "Level 7" of a software factory, enabling the management of 10 to 15 parallel Claude Code sessions running simultaneously on different projects or branches.
- Tan's current workflow involves processing 400 pending PRs for open-source projects and generating 10 to 50 pull requests daily, eliminating the need for a traditional to-do list.
- GStack mitigates supply chain attack risks by allowing the AI to review, test, and merge code in isolated, parallel environments before deployment.
- The framework now includes a Ship Tool to verify pull request readiness, completing the end-to-end engineering lifecycle from ideation to production.
- The GStack repository has accumulated over 70,000 stars, with users reporting that 80–90% of their development time is now spent in strategic planning and review phases rather than manual coding.
- Tan encourages developers to utilize the "slash office hours" command at
github.com/garytan/gstackto leverage institutional YC startup thinking.