Rodrigues Skill Architecture vs Isenberg Opportunity Scanner
// TL;DR
These two skills solve completely different problems and never compete. If you are building AI agent tooling and need agents to use your product correctly, use the Rodrigues Product Skill Architecture Method. If you are searching for a startup idea, evaluating niches, or figuring out what to build and for whom, use the Greg Isenberg Startup Opportunity Scanner. One is an engineering method for AI agent accuracy; the other is an ideation and validation framework for entrepreneurs. Pick based on your current problem, not preference.
// HOW DO THEY COMPARE?
| Dimension | Rodrigues Product Skill Architecture Method | Greg Isenberg Startup Opportunity Scanner |
|---|---|---|
| Best For | Platform/product teams building AI agent integrations | Founders and aspiring entrepreneurs searching for startup ideas |
| Core Problem Solved | Closing the context gap between AI training data and your specific product | Identifying and validating high-potential startup opportunities in underserved niches |
| Complexity | High — requires technical knowledge of MCP, agent behavior, evals, and product security | Low to moderate — requires market intuition and willingness to research, but no technical prerequisites |
| Time to Apply | Days to weeks — iterative cycle of drafting skill.md, writing evals, testing across models | Hours to days — can generate and stress-test a startup thesis in a single working session |
| Prerequisites | An existing product or platform, known agent failure modes, canonical documentation, MCP familiarity | A broad category of interest; optionally unfair advantages and target audience hypothesis |
| Output Type | A versioned skill.md file with front matter, security checklists, opinionated workflows, and eval suite | A validated startup thesis: niche, persona, product category, monetization stack, and acquisition wedge |
| Creator Background | Pedro Rodrigues, Supabase — engineering/developer tooling | Greg Isenberg — serial entrepreneur, community builder, startup advisor |
| Domain | AI engineering, developer experience, agent tooling | Startup strategy, niche selection, business model design |
| Iteration Model | Eval-driven: run scenario tests, promote skipped content, re-test across models | Thesis-driven: qualify niches, stress-test with 'marry the niche' principle, validate acquisition |
| Who Should NOT Use This | People who don't have a product yet — this assumes an existing platform | People who already know what to build and need technical agent integration guidance |
What does the Rodrigues Product Skill Architecture Method do?
The Rodrigues Product Skill Architecture Method is a systematic engineering framework for building reusable agent skill documents (skill.md files) that close the gap between what AI agents know from training data and what they need to know to work correctly with your specific product. Developed by Pedro Rodrigues at Supabase, it addresses a precise problem: agents produce stale, unsafe, or workflow-incorrect outputs when interacting with platforms that have proprietary workflows, security requirements, or post-training-data APIs.
The method walks you through auditing known agent failure modes, separating must-load from reference content, encoding opinionated workflows, and running rigorous evals across multiple model families. Its core insight is that agents are lazy — they skip reference files, default to training data, and resist tool calls. The skill.md must therefore contain every non-negotiable rule directly, point stubbornly to live documentation, and be tested like code.
The output is a versioned skill.md file bundled in a repo, complete with front matter for agent discovery, security checklists, explicit workflow sequences, and a companion eval suite. It is a technical artifact for technical teams.
What does the Greg Isenberg Startup Opportunity Scanner do?
The Greg Isenberg Startup Opportunity Scanner is a strategic ideation and validation framework that helps founders identify high-potential startup opportunities. It codifies Greg Isenberg's methodology around niche selection, verticalization, and monetization into a repeatable workflow.
The framework's signature moves include the CVS Shelf Heuristic (walk into a pharmacy and look at shelf density to spot massive pain points), the "Fish Where the Fish Are" filter (target underserved demographics like 45–65+ adults rather than crowded young markets), and the principle of "Date the Product, Marry the Niche" (the audience is the long-term bet, the product is expendable).
You walk through nine steps: enumerate sub-niches, filter for underserved audiences, run a three-question qualification test, map the jobs-to-be-done stack, select a product category (Action App, Community/IRL, Elder Tech, Creator Media, or Personalized Health), design a Free + Premium monetization stack, stress-test the niche commitment, identify an acquisition wedge, and check for builder-market fit. The output is a validated startup thesis ready for execution.
How do the Rodrigues Skill Architecture Method and the Isenberg Opportunity Scanner compare?
These two frameworks operate in entirely different domains and solve entirely different problems. The Rodrigues method is a post-product engineering framework — you already have a product, you already have agents interacting with it, and those agents are getting things wrong. The Isenberg Scanner is a pre-product strategy framework — you don't have a product yet, and you need to figure out what to build and for whom.
The Rodrigues method requires deep technical context: knowledge of MCP servers, agent behavioral patterns, eval design, and your product's security model. The Isenberg Scanner requires entrepreneurial intuition: the ability to spot underserved audiences, evaluate willingness-to-spend signals, and design monetization architectures. One produces a markdown artifact that ships in a code repo; the other produces a business thesis that ships as a pitch deck or product spec.
The only overlap is that both are opinionated — the Rodrigues method insists you encode your best workflows rather than letting agents infer, and the Isenberg Scanner insists you go vertical rather than horizontal. Both reject neutrality as a design choice. But they apply that opinionation to completely different problems.
If you are trying to decide between them, the answer is almost certainly obvious from your current situation: if you have a product and agents are misusing it, use Rodrigues. If you're looking for what to build next, use Isenberg.
Which should you choose?
Choose the Rodrigues Product Skill Architecture Method if:
- You have an existing product or platform with APIs, security policies, or proprietary workflows.
- AI agents are producing incorrect, stale, or unsafe outputs when interacting with your product.
- You need a repeatable, eval-tested artifact that improves agent accuracy across multiple model families.
- You are a platform engineer, developer experience team, or technical product owner.
Choose the Greg Isenberg Startup Opportunity Scanner if:
- You are searching for a startup idea or evaluating which niche to enter.
- You have a broad category of interest but no validated thesis for who to serve and how to monetize.
- You want a structured method to avoid the most common founder mistakes: going horizontal, targeting crowded demographics, or under-pricing community businesses.
- You are a founder, aspiring entrepreneur, or product strategist in the ideation phase.
Can you use both? Yes — sequentially. Use the Isenberg Scanner to identify your startup opportunity and build the product. Once agents start interacting with your product and making mistakes, use the Rodrigues method to build the skill.md that fixes their behavior. They sit at different stages of the company lifecycle and complement each other perfectly.
// FREQUENTLY ASKED QUESTIONS
Can I use the Rodrigues Skill Architecture Method if I don't have a product yet?
No. The Rodrigues method requires an existing product with known agent failure modes, canonical documentation, and security requirements. It is a post-product framework. If you're still searching for what to build, use the Isenberg Startup Opportunity Scanner first, then apply Rodrigues once agents interact with your product.
Is the Greg Isenberg Startup Opportunity Scanner only for AI startups?
No. While the framework covers AI-specific categories like Action Apps and AI Native Media Companies, it is equally applicable to community businesses, IRL events, elder tech, personalized health, and non-AI verticals. The core methodology — niche selection, verticalization, and Free + Premium monetization — is industry-agnostic.
Do I need to know MCP to use the Rodrigues Product Skill Architecture Method?
MCP familiarity is strongly recommended. The method positions skill.md as complementary to MCP server integrations — MCP provides agent action capabilities, while the skill provides guidance on using those tools correctly. You can draft a skill.md without MCP, but the full eval framework tests baseline, MCP-only, and MCP+skill conditions.
Which framework is faster to apply — Rodrigues or Isenberg?
The Isenberg Startup Opportunity Scanner is significantly faster. You can generate and stress-test a startup thesis in a few hours. The Rodrigues method requires days to weeks because it involves auditing failure modes, drafting skill.md, writing evals, running tests across multiple models, and iterating based on results.
Can I combine both frameworks for building an AI agent product?
Yes, and this is the ideal sequence. Use the Isenberg Scanner to identify the niche, validate the audience, and select the product category (e.g., an Action App for podcast producers). Once you've built the product and agents start interacting with it, use the Rodrigues method to create the skill.md that ensures agents use your platform correctly and safely.
What does the Rodrigues method mean by 'agents are lazy'?
Agents resist loading reference files, default to training data over live documentation, and avoid making tool calls. The Rodrigues method accounts for this by requiring all non-negotiable rules to live directly in skill.md, not in supplementary files. It also instructs skill authors to repeat fetch-documentation directives multiple times to overcome this default laziness.
What is the CVS Shelf Heuristic in the Isenberg framework?
It's an ideation shortcut: walk into a pharmacy and look at which over-the-counter product categories occupy entire rows of shelf space. That density signals a massive, painful, underserved problem worth building a vertical startup around. For example, rows of antacids signal GERD as a high-pain vertical ripe for a dedicated health app.
Are these frameworks competing alternatives?
No. They solve completely different problems at different stages. The Rodrigues method is for engineering teams making AI agents work correctly with an existing product. The Isenberg Scanner is for entrepreneurs figuring out what product to build. There is no scenario where you would choose between them for the same task.