Build a new prototype in one pass, then test the working surface before making narrow follow-up changes.
Source episode · 01:35
How I AI: My First Impressions of Claude Opus 4.8 – Coding, Strategy, and Where It Shineswith Claire Vo
01:35 to 04:02: Claire tests a one-shot greenfield build, then finds that follow-up edits and existing-code edges need tighter scope and direct preview checks.
A functional prototype that demonstrates the primary path and exposes the assumptions behind the one-shot build.
Anthropic AI assistant
Anthropic AI assistant
Step by step
Follow the sequence once, then adapt the prompts, checks, and handoffs to your own setup.
5 steps
Outline the complete architecture and functionality for a new feature you want to build from scratch. Provide specific instructions, as more detail leads to a better one-shot build.
Describe the user, primary action, data, empty state, error state, and finish condition.
In the Claude Code environment, give Claude Opus 4.8 a single, detailed prompt. Specify the complete architecture and functionality for the prototype you want it to build in one shot.
Create a fresh branch, use the existing stack, and build only the described feature.
Allow the model to enter its planning mode and then autonomously code the feature. For a complex task, this process can take around 20 minutes to complete.
Run the app and complete the primary path plus the named edge states before stopping.
After the code is generated, push it to a preview branch or another testing environment. Verify that the feature is live and works as expected from the one-shot build.
Deploy a preview tied to the commit and report the exact test URL and sample data.
Use caution when asking the model to make iterative changes or fix bugs on the initial code. It may introduce new bugs or hallucinate fixes when asked to iterate on the one-shot build.
Limit this correction to the observed preview failure and rerun the same reproduction steps.
Turn an idea into a PRD, user stories, and a plan.
After the steps
How to recover when the loop fails and where human judgment helps.
Recover
Keep building
Use a hybrid AI workflow to bridge design and development. Generate high-level design concepts and tokens with Claude, then use Codex for robust implementation of the technical components in your codebase.
Implement a QA loop for AI-generated code using Devin. This playbook combines automated code reviews with an agent-recorded video walkthrough for visual verification, ensuring high-quality merges at scale.
Build a reusable 'Watchdog' playbook in Devin to automatically monitor customer accounts. The agent proactively checks for errors, synthesizes a prioritized list of critical issues, and saves you time.
Join 100,000+ product managers who use ChatPRD to write better docs, align teams faster, and build products users love.