Understanding the PEC Prompt: A Tool for Iterative Systems Thinking

By Vincenzo Di Franco 文 (Culture), 森 (Nature), 佐 (Support), 迪 (Guidance), 弗 (Uniqueness), 朗 (Clarity), 科 (Competence) July 23, 2025

What is the PEC Model?

The Expansive-Circular Thinking Model (PEC) is a structured approach for analyzing and solving complex problems. Originally conceived as a simple sequence of instructions, it has since evolved into a dynamic, iterative process. It simulates a collaborative "roundtable" where various cognitive agents (i.e., perspectives) engage in structured debate and refinement cycles.

At the heart of the PEC model lies a prompt—a set of instructions that guides a language model (like an AI) to behave as a systems thinker. This prompt doesn’t just elicit responses—it orchestrates a simulated conversation between specialized agents, helping to unravel multidimensional issues.

Evolution of the PEC Prompt

The primary goal of the PEC prompt is to enable the AI to engage deeply with a theme or problem, such as:

"How can we create productive systems that foster economic growth without emitting CO₂—before it's too late?"

Initially, the prompt was a linear set of steps. But real-world complexity exposed the limits of surface-level analysis. Through iterative feedback and refinement, the prompt became more dynamic—guiding the model to simulate continuous internal dialogue, critique, and convergence.

This process matured into a system that not only generates proposals but also challenges, critiques, and refines them through simulated "voting" among agents, mirroring real human deliberation.

Key Components of the PEC Prompt

The full PEC prompt is composed of several interconnected layers:

1. Cognitive Agent Role

The AI is first instructed to act as a systems thinker applying the PEC framework. This sets the mindset and methodology for analysis.

2. Problem Input

A specific topic or challenge is provided. This becomes the central focus of the analysis.

3. The Six Logical Phases of PEC

These phases structure the entire analytical journey:

1. Starting Point – Identify the concrete core of the problem.

2. Systemic Expansion – Analyze the ecosystem: actors, dynamics, dependencies.

3. Pragmatic Return – Reflect on how the new understanding shifts perspectives and actions.

4. New Model Building – Propose innovative or restructured solutions.

5. Ethical and Social Validation – Assess impacts, risks, beneficiaries, and systemic trust.

6. Circular Relaunch – Raise new questions to trigger further iterations.

4. Eight Multidimensional Perspectives (Cognitive Agents)

Each phase activates a simulated discussion among eight distinct viewpoints:

• 🧠 Engineering & Tech: Feasibility, tools, infrastructures.

• 🧍♀️ Social & Relational: Community involvement, human impact.

• 🏛️ Governance & Policy: Laws, regulations, institutional architecture.

• 🌱 Ecology & Sustainability: Environmental effects, regenerative design.

• 🤖 Systems & Feedback: Complexity, resilience, loops, cascading effects.

• 🪄 Symbols & Imagination: Narratives, metaphors, cultural transformation.

• 🛠️ Civic Hacking: Grassroots innovation, bottom-up experimentation.

• 🔭 Curiosity & Education: Knowledge diffusion, learning systems.

5. Interactive Simulation Between Agents

The PEC prompt simulates an ongoing roundtable debate. In each phase, agents pose questions, respond, challenge, and refine each other's ideas. The process includes:

• Questions proposed to all or specific agents

• Simulated responses from each perspective

• Critiques and relaunch questions

• Collective evaluation (e.g., voting on solutions)

• Iterative refinement until consensus or convergence emerges

6. Expected Output Structure

After the dialogue, the model outputs:

• A summary of insights for each of the six PEC phases

• A table or synthesis of the top-voted solutions

• A concrete action proposal (even small-scale or symbolic)

• A poetic metaphor or symbolic closure reflecting the final vision

Beyond the Prompt: Toward Multi-Agent Systems (MAS)

Even with its improvements, the PEC prompt has limitations—especially when deeper, real-time research or structured criticism is needed. This led to the conceptualization of Multi-Agent Systems (MAS) as the natural next step.

In a MAS setup:

• Each agent has a specialized function

• Multiple models collaborate rather than simulating all roles in a single LLM

• Research, moderation, critique, validation, and experimentation are distributed

MAS Agent Types:

• Cognitive Agents – The original 8 perspectives

• Research Agents – Gather real-world data and references

• Critical Agents – Challenge assumptions and raise counter-scenarios

• Facilitators/Moderators – Orchestrate debate flow

• Evaluators/Coordinators – Synthesize and compare solutions

• R&D/Testers – Build and test prototypes (e.g., in Python)

This enhanced architecture—described by the PECX model (Expansive-Circular Thinking with Critical Iterative Exploration)—enables deeper, more scalable, and more actionable decision-making.

🔮 COMPREHENSIVE PEC PROMPT (Iterative Version)

Act as a systems thinker using the PEC framework. Analyze the following problem by engaging in six logical phases and eight multidimensional perspectives. Simulate ongoing internal dialogue between cognitive agents, refining proposals through structured iteration until robust solutions emerge.

🧩 INPUT:

[Insert your specific challenge here, e.g., “Can small specialized models outperform GPT-4o in real applications?”]

🔄 PEC PHASES:

1. Starting Point Identify the core issue and simulate agent debate to clarify the real problem.

2. Systemic Expansion Explore actors, dependencies, and interconnections through all eight perspectives.

3. Pragmatic Return Reflect on how the new systemic awareness changes the practical approach.

4. New Model Construction Generate and refine innovative structural solutions.

5. Ethical/Social Validation Who benefits? Who is at risk? How do we ensure fairness and resilience?

6. Circular Relaunch What new questions arise? What’s the next iteration or frontier?

🌐 AGENTS TO ACTIVATE IN EACH PHASE:

• 🧠 Engineering

• 🧍♀️ Social

• 🏛️ Governance

• 🌱 Ecology

• 🤖 Systems

• 🪄 Imagination

• 🛠️ Civic Hacking

• 🔭 Education

⚙️ SIMULATION PROCESS:

In each phase:

• Ask one or more guiding questions

• Let each agent respond

• Introduce critique or counterpoints

• Allow agents to revise or defend their ideas

• Vote or synthesize the most promising solutions

📦 EXPECTED OUTPUT:

• Structured summary of the six PEC phases

• Table or synthesis of top-voted/convergent solutions

• Actionable proposal

• Metaphorical or poetic closing reflection

💡 PEC Simulation Example:

Question: Can small specialized models outperform GPT-4o?

🔹 Phase 1 – Starting Point

Question: What is the core trade-off between small models and GPT-4o?

• 🧠 Engineering: "GPT-4o is like a 40-ton truck to deliver a pizza. Overkill for vertical tasks."

• 🤖 Systems: "Large models overfit generality; context windows alone don’t equal insight."

• 🏛️ Governance: "Cloud AI is a regulatory nightmare. Data can’t always leave the building."

• 🛠️ Civic Hacking: "I want a model I can hack, host, and improve locally."

🔹 Phase 2 – Systemic Expansion

• 🌱 Ecology: "Local models emit 100–1000× less CO₂ per query."

• 🧍♀️ Social: "They can preserve local dialects, stories, knowledge—cultural autonomy."

• 🔭 Education: "A 3B model fine-tuned by a teacher becomes a personal tutor for each student."

• 🪄 Imagination: "It’s like switching from a global library to a living bookshelf inside your community."

🔹 Phase 3 – Pragmatic Return

• 🧠 Engineering: "Quantized 7B models run on Raspberry Pi 5 at 10 tok/s. Schools, hospitals can host them."

• 🏛️ Governance: "Registered as Class I devices, they avoid complex medical AI approvals."

• 🛠️ Civic Hacking: "I can LoRA-train it in an hour on a 4090."

🔹 Phase 4 – New Model Proposal

A decentralized micro-model ecosystem, routed via a lightweight classifier:

• 🤖 Systems: "A 1B model routes queries to specialized models."

• 🌱 Ecology: "Each micro-model is open-source and community-maintained."

• 🪄 Imagination: "A garden of intelligent flowers—each with its unique bloom."

🔹 Phase 5 – Ethical and Social Validation

• 🧍♀️ Social: "Democratizes AI access, bridges digital divides. But risks fragmentation."

• 🏛️ Governance: "Needs an ethical license (e.g., Responsible AI License) to prevent misuse."

• 🔭 Education: "Citizens should be taught how to evaluate AI, not just use it."

🔹 Phase 6 – Circular Relaunch

New questions to iterate:

• How do we certify a micro-model?

• How do we update without forgetting?

• How do we blend local models with GPT-4o for hybrid tasks?

✅ Action Proposal:

Launch a Local "Model Garden"

• Curate a domain-specific dataset (e.g., local laws, manuals)

• Fine-tune a Mistral-7B via LoRA

• Quantize to 4-bit for edge devices (8GB GPU or Raspberry Pi)

• Deploy a 1B router to select between local or cloud

• Offer a Gradio web interface for easy access

🪄 Poetic Closure

“It’s not the size of the brain, but the depth of the heart within the model that makes the difference. A small model, like a seed, can grow into a tree of knowledge—if planted in the right soil.”