Having been an engineer, a founder, a product manager, and a designer, I have a lot of experiments running at any given time but the format and goals vary widely. But it wasn’t until I joined an advance reading group of The Experimentation Machine, by Jeff Bussgang, that I imagined a more systematic way of approaching what has always come very organically to me.
I began my “experiment of experiments” in a work context, but have expanded and interated on it in my personal life. My ultimate goal is to surface my working environment here for curious onlookers.
Iteration 1
A system created within the Cisco IT security context using Visual Studio Code and our approved internal enterprise LLM. Leverages an agents.md framework to coordinate between a few key agents to refine ideas, write PRDs, and build out working prototypes.
The agents
The hub contains 4 active agents that work together in a structured workflow to create, document, and build experiments:
- 🎯 Strategist - Strategic coaching to frame experiments with proper problem validation and success criteria
- 📋 Product Manager - Creates comprehensive Product Requirements Documents balancing technical/business/experience goals
- ⚡ Technical Architect - Builds minimal, performant prototypes using reusable components and UX heuristics
- 🎨 Design Advisor - Reviews prototypes and povides research-backed UX analysis and recommendations
The workflow
- Experiment. This often starts with a question I ask myself or a need I perceive in my workday, such as “Why can’t I take a figma file and automatically check it for new pattern or component candidates?” Then the main agent helps me create the framing for the experiment.
- PRD. The PRD agent asks questions and pushes at assumptions so that a simple document can be created. I can modify the PRD or gut check if I’ve over-complicated things.
- Prototype. The Prototype builder knows some basic technical preferences of mine, as well as the basic tech stack I need to be using. Once the prototype exists I iterate on it from a functional standpoint, both via chat as well as in individual files.
Iteration 2
In my second iteration I geared it more explicitly toward my personal interest in product and startup ideas. I still leverage an agents.md structure for defining my collaboration team, but with Cursor, OpenAI, and Claude as the central orchestration stack.
The agents
- Experiment Creator — Refines raw ideas into structured experiments with clear statements, directory structure, and metadata. Creates the foundation for tracking and validation.
- Market Research — Conducts market analysis and calculates TAM/SAM/SOM estimates. Analyzes competitive landscape, market trends, and business opportunity to inform go/no-go decisions.
- PRD Writer — Creates Product Requirements Documents that balance user needs, technical feasibility, and business goals. Translates experiments into actionable product specifications.
- Prototype Builder — Builds minimal, performant prototypes from PRDs using appropriate tech stacks and best practices. Generates initial code structure and implementation guidance.
- Design Advisor — Reviews PRDs and prototypes for design compliance, providing UI/UX feedback, accessibility audits, and component suggestions. Automatically invoked by PRD Writer and Prototype Builder at key checkpoints.
The workflow
- Experiment. Starts with a question or need. The Experiment Creator helps refine the idea, frame the statement, and set up the structure. It asks clarifying questions and waits for approval before creating directories and metadata.
- Market Research (optional). When I want to validate the business opportunity, Market Research analyzes the market, calculates TAM/SAM/SOM, and assesses competition. It presents estimates for approval to inform the PRD.
- PRD. The PRD Writer asks questions and challenges assumptions to produce a clear document. Before finalizing, it automatically calls the Design Advisor to review UI/UX sections—checking for missing design specs, accessibility requirements, and component needs. I can modify or simplify the PRD, then approve it before it's saved.
- Prototype. The Prototype Builder uses my technical preferences and required stack. It analyzes the PRD, proposes structure and tech stack for approval, then generates the initial code. After generating code, it automatically calls the Design Advisor to review for design compliance. I approve the design improvements, and the prototype structure is finalized. Once it exists, I iterate functionally via chat and in individual files.
The workflow has approval checkpoints, so I control when to proceed. Design quality is reviewed at both the specification (PRD) and implementation (prototype) stages automatically.