Back/Engineering/Claude Code
AdvancedEngineering

How to Build a New AI Product in 10 Weeks Using the 'No-Process' Method

Learn Gusto's radical approach to building a new product line in 10 weeks by ditching traditional tools like Jira and Figma, and relying on a perma-Zoom room and an AI coding assistant for rapid development.

How to Build a New AI Product in 10 Weeks Using the 'No-Process' Method

From 08:32 to 19:09, Eddie Kim explains the small-team build process behind Gusto Cofounder, including the trash-can method, the stack, and shipping from day one. Clip range: 08:32 to 19:09.

Before you start

What you need

  • Claude Code or another AI coding assistant
  • TypeScript and Cloudflare Workers development environment
  • Sample payroll spreadsheet data
  • Core project team with engineering and design contributors
  • Feature flag configuration for experimental UI work

What you’ll make

A working prototype of a payroll assistant with a deployable demo and feature-flagged UI components.

Tools used

Step by step

The workflow

Follow the sequence once, then adapt the prompts, checks, and handoffs to your own setup.

7 steps

Step01

Create a Rapid Prototype

Use an AI coding assistant like Claude Code to build a working prototype of your core idea. The goal is to materialize the concept quickly and create a shareable demo, not to write production-ready code.

Example prompt
I need to build a rapid prototype for a new product idea. It's a conversational AI agent that helps small business owners prepare for payroll.

Here's the core flow:
1. The user provides a Google Sheet with data like employee hours, sales, and tips.
2. The user gives instructions in plain English for calculating adjustments, like 'For every hot stone upsell, add a $15 bonus for that therapist.'
3. The agent reads the sheet, applies the rules, and presents the final payroll amounts for approval.

Please generate the code for a simple working demo using TypeScript and Cloudflare Workers. The front-end can be minimal. Focus on the logic for parsing instructions and calculating the payroll adjustments from mock data.
Step02

Generate Excitement and Recruit Your Team

Record a Loom video of your working prototype to share with a small group of senior engineers and designers. Use their feedback and buy-in to recruit the core project team.

Step03

Define Scope with a Single Whiteboard Session

Hold one initial whiteboard session to sketch out the vision and architecture. A photo of this whiteboard will serve as the sole piece of documentation for the project.

Step04

Eliminate Process Overhead

Commit to using no traditional process tools. This means no Jira tickets, no Figma designs, no tech specs, no scheduled meetings, and no standups.

Step05

Establish a 'Perma-Zoom' Room

Create a constant, 24/7 open Zoom call. This virtual room serves as the central hub for all collaboration, pair programming, and ad-hoc discussions, replacing all scheduled meetings.

Step06

Adopt the 'Trash Can Method'

Instead of writing specs, build features directly and open a pull request. The PR itself is the proposal. If the team decides against the feature, close the PR and discard the code. This method relies on AI's speed to make code cheap enough to throw away.

Step07

Empower Non-Coders to Ship

Enable non-engineers, such as designers, to contribute code directly. A good starting point is shipping 'faked' front-end experiences with canned responses behind feature flags. Engineers can then wire up the real functionality, which requires a supportive code review culture.

Example prompt
I'm a designer, and I need to create a 'faked' front-end experience for a new feature. The feature is a new dashboard widget that summarizes the upcoming payroll run.

Please generate the React component for this widget. It should include:
- A title: 'Upcoming Payroll'
- A placeholder for the total amount.
- A list of 3-4 employees with placeholder payment amounts.
- A 'View Details' button.

All data should be hardcoded. The entire component needs to be wrapped in a feature flag called [new-payroll-summary-widget]. This allows me to ship the UI for review before the back-end is ready.

What good looks like

  • Prototype parses spreadsheet data and applies plain-English adjustment rules
  • Demo can be shown in a Loom walkthrough
  • Feature-flagged UI components can be enabled without backend dependencies
  • Code changes are proposed through pull requests for review

Build your next product with ChatPRD

Turn an idea into a PRD, user stories, and a plan.

Try ChatPRD free

After the steps

Runbook notes

How to recover when the loop fails and where human judgment helps.

Recover

If it goes sideways

Payroll calculations produce incorrect adjustment totals
Validate calculations against manually checked sample payroll scenarios
Feature-flagged UI leaks unfinished functionality to users
Restrict feature flags to internal environments until reviewed
Cloudflare Worker exceeds runtime or memory limits
Reduce dataset size or move heavy calculations to background processing

Start shipping
better products.

Join 100,000+ product managers who use ChatPRD to write better docs, align teams faster, and build products users love.

Free to start
No credit card
SOC 2 certified
Enterprise ready