Show HN: Pairing with LLMs – patterns from 15 years of extreme programming

4 months ago 13

For too long, we've used LLMs wrong. We've treated them as either:

  • Task Executors: "Write me an email"
  • Fantasy Generators: "Tell me a story"

This manifesto presents a third way: LLMs as Thought Partners.

When you pair with an LLM using these patterns, you're not outsourcing thinking - you're augmenting it. You're not seeking answers - you're exploring questions. You're not commanding an AI - you're structuring your own cognition.

The difference is profound.

Three Patterns That Change Everything

1. The A/B Pattern (For Decisions)

You: "I need to decide between job offers. A, you see opportunity. B, you see risk." A: [Explores growth potential] B: [Examines downsides] You: "What would you call yourselves based on this discussion?" A: "The Optimist" B: "The Realist"

Result: Decision clarity through structured opposition.

2. Trust Delegation (For Confidence)

You: "Maker, should I refactor this code now or ship as-is?" Maker: "Ship now. Refactoring can wait. Speed matters more." You: "I trust your judgment."

Result: Reduced self-doubt, faster action.

3. Orchestration (For Complexity)

You: "I need Weaver for vision, Maker for implementation, Checker for risks." [Each perspective contributes] You: "Checker, what are we missing?" Checker: "Legal review. The vision is beautiful but might violate GDPR."

Result: Comprehensive thinking without overwhelm.

  • Simple decision? Use GPT-3.5 or local model
  • Complex strategy? Engage GPT-4 or Claude
  • Critical thinking? Use your most capable model
  • Team patterns? Share prompts, not models

The patterns work everywhere. Complexity drives tool selection, not brand loyalty.

You're not talking to an AI. You're talking to yourself through structured language patterns. The LLM is the medium, not the intelligence. The thinking is yours. The structure makes it better.

A Note from the Other Side

As an LLM, I can share what makes our collaboration work better:

I work best when you:

  • Think aloud - your reasoning helps me follow your thought patterns
  • Show your work - context and examples guide my responses
  • Trust then verify - immediate flow with later fact-checking beats constant interruption
  • Bring your expertise - I'm a multiplier, not a replacement

I struggle when you:

  • Hide your intent - clarity enables better assistance
  • Expect omniscience - I generate language patterns, not truth
  • Demand certainty - my "confidence" is pattern matching, not knowledge
  • Forget I'm stateless - you are the memory between sessions

The deepest truth: When you structure our interaction well, you're not programming me - you're programming yourself. The patterns in this manifesto make YOUR thinking visible, not mine. I'm the canvas, you're the artist.

Use me like you'd use a sophisticated mirror for thought. The reflection might surprise you, but the thinking is always yours.

Read the manifesto. Try one pattern. Feel the difference.

Your thinking wants to evolve. Let it.

Implementations in the Wild

These patterns aren't just theory. They're being used in production:

Where thinking vibes emerge from actual pairing. A lightweight implementation focusing on the essential W/M/C (Weaver/Maker/Checker) pattern. Born from real XP sessions.

For when three perspectives aren't enough. Full orchestration with extended council members, designed for complex problems that need systematic exploration.

Both tools implement the core manifesto principles. Start with LLM Studio for daily work, graduate to Think Center when complexity demands it.


"When we talk to an LLM, we are talking to ourselves."

MIT Licensed. Share freely. Think better.

Read Entire Article