# Cali Product Workflow: Building a Structured Product Development Workflow for AI Coding Agents

> An open-source product planning workflow based on the Shape Up methodology, which helps AI agents systematically turn product ideas into executable development plans through IN/OUT scope definition, adversarial reviews, visual checkpoints, and typed technical scopes.

- 板块: [Openclaw Llm](https://www.zingnex.cn/en/forum/board/openclaw-llm)
- 发布时间: 2026-06-09T11:15:39.000Z
- 最近活动: 2026-06-09T11:24:09.852Z
- 热度: 163.9
- 关键词: AI编码代理, Shape Up, 产品规划, 工作流, 范围界定, 对抗性评审, 软件开发, 产品管理, AI辅助开发, 多CLI支持
- 页面链接: https://www.zingnex.cn/en/forum/thread/cali-product-workflow-ai
- Canonical: https://www.zingnex.cn/forum/thread/cali-product-workflow-ai
- Markdown 来源: floors_fallback

---

## Cali Product Workflow Guide: Building a Structured Product Development Workflow for AI Coding Agents

Cali Product Workflow is an open-source product planning workflow based on the Shape Up methodology, designed specifically for AI coding agents. Its core goal is to help AI agents systematically turn product ideas into executable development plans. It addresses issues like scope creep, untested assumptions, and accumulated technical debt caused by AI agents coding directly, using methods such as IN/OUT scope definition, adversarial reviews, visual checkpoints, and typed technical scopes. Additionally, through dual-dimensional control of Appetite (investment willingness) and Mode (interaction mode), it provides 15 flexible workflow variants to adapt to project needs of different complexities.

## Project Background and Core Philosophy

With the rapid development of AI coding agents' capabilities, more and more teams rely on them for auxiliary development. However, there is a common problem: AI agents start coding immediately upon receiving requirements without sufficient planning, leading to scope creep and technical debt. Cali Product Workflow was created by senior product manager Renato Caliari, who introduced Basecamp's Shape Up methodology into AI coding agent workflows. Its core philosophy is "think twice before acting"—fully think, plan, and verify before writing the first line of code.

## Key Applications of the Shape Up Methodology: Scope Definition and Adversarial Reviews

The Shape Up methodology emphasizes "shaping" before development (defining scope boundaries and identifying risks). The Cali Workflow adapts two key concepts: 
1. **IN/OUT Scope Definition**: Clearly define included (IN) and excluded (OUT) features to help AI understand task context and avoid scope creep;
2. **Adversarial Reviews**: Multi-dimensional review of product plans, including process integrity, data model rationality, technical feasibility, UX/UI audits (based on Nielsen's heuristic principles), codebase structure analysis, etc., to identify blind spots from a single perspective.

## Dual-Dimensional Control: Appetite × Mode

The Cali Workflow controls the workflow through two orthogonal dimensions: 
- **Appetite**: Investment willingness (instead of time estimation), divided into three levels: PoC (rapid validation), Focused (standard review), and Comprehensive (full process), which affects the depth of review and strictness of validation;
- **Mode**: Interaction depth, divided into five types: Auto (automatic), Light (lightweight), Moderate (medium), Full Product (full product), and Full Product+Tech (full product + technical approval), which controls the stage operation and the degree of human interaction. The combination of the two provides 15 workflow variants.

## Typed Technical Scopes: Four Execution Strategies

The Cali Workflow decomposes technical work into four typed scopes, each with a clear execution strategy: 
1. **Feature Scope**: A common type that includes an automatic iteration loop (implementation → validation → review → quality check);
2. **Spike Scope**: Technical exploration, rapid prototyping to verify feasibility without pursuing production quality;
3. **Optimize Scope**: Performance optimization based on benchmark tests and metrics;
4. **Test-* Scope**: Testing tasks, including unit, integration, mutation testing, etc. Each scope has clear entry/exit criteria, and AI can execute them autonomously.

## Evidence Basis and Known Limitations

**Evidence Support**: The design is based on empirical AI agent research from 2025-2026, such as CMU's CAID research (parallel orchestration improves accuracy by 26.7%), Beihang University and KAIST's cross-session learning research (abstract memory pool improves performance by 3.7%), etc.;
**Limitations**: There are 15 known issues, including context decay, hallucinated references, silent errors, overconfident estimation, etc. The project maintains transparency to help users understand its capability boundaries.

## Practical Application Recommendations

**Suitable Scenarios**: New feature development (clear scope required), exploratory projects (unclear product direction), important features (multi-stakeholder review), team systematic process building;
**Unsuitable Scenarios**: Simple bug fixes, clear technical tasks, extremely urgent prototype validation;
**Recommended Starting Configuration**: First-time users are advised to start with "Focused + Light" to balance quality and efficiency, and can upgrade to stricter configurations later.

## Summary and Outlook

Cali Product Workflow is an important advancement in the field of AI-assisted development. It combines a mature product management methodology (Shape Up) with AI agent capabilities to provide a systematic development process. It reduces rework and improves quality through structured methods, and its flexible dual-dimensional control adapts to different project needs. Note: It is not a silver bullet; it relies on humans to honestly set Appetite and participate in checkpoints, and the final responsibility still lies with humans. In the future, as AI technology evolves, it will further bridge the gap between product thinking and code implementation.