The Good Stuff, with Pete and Andy - Episode 21: Remote Vibe Coding & AI Agent Orchestration

Hosts: Pete and Andy (first remote recording - Pete traveling, Andy at the beach)

Episode Overview: Pete and Andy explore advanced AI agent workflows, the evolution of AI coding tools, and strategies for building sustainable development practices while Pete is away doing "vibe coding and freedom tech" work.

Remote AI Development & Wingman Evolution (03:00-15:00)Pete demonstrates how geographic separation enhances AI-powered development, using his Wingman system to continue building while traveling. Discussion of voice-driven ideation versus typing, and the cognitive benefits of speaking ideas into existence without syntax distractions.

Trigger-Based Automation & Overnight Jobs (15:00-25:00)Introduction of Wingman's new trigger system allowing file-based automation and scheduled tasks. Pete's breakthrough with overnight security reviews using "Grug Brain Developer" principles to audit code changes and identify complexity concerns while he sleeps.

Claude Code vs. Goose: The Dual-Agent Approach (28:00-32:00)Analysis of using different AI models for different cognitive tasks - Claude Code as the enthusiastic "drunk writer" for rapid development, Goose as the thoughtful "sober editor" for planning and documentation. The "dog with two dicks problem" of Claude Code's over-eagerness to build.

Documentation Architecture & Anti-Slack Philosophy (32:00-40:00)Strong critique of chat-based interfaces for knowledge work. Pete's systematic approach to moving valuable insights from chat conversations into structured Obsidian documentation, and why real-time messaging tools damage business focus and inspiration.

AI Agent Loop Problems & Root Cause Analysis (41:00-50:00)Technical discussion of when Claude Code gets stuck in debugging loops due to hard-coded values and scope creep. Solutions include using different AI models for root cause analysis and implementing background auditing agents to catch coding mistakes.

Design-First Development & Planning Orchestration (50:00-59:00)The importance of spending more time in ideation and design phases before coding. Discussion of missing orchestration tools to convert comprehensive documentation into manageable work packages, and the trade-offs between YOLO development and structured planning.

• Segregated Computing: Using dedicated machines for AI development provides security and eliminates local environment dependencies
• Voice-Driven Ideation: Speaking ideas creates more fluid thought processes compared to typing with its syntax distractions
• Overnight Labor: AI agents can provide 16 hours of cleanup and maintenance work while humans sleep
• Dual-Agent Strategy: Different AI models excel at different cognitive tasks - planning vs. execution
• Documentation Philosophy: Structured knowledge systems beat chat interfaces for long-term value creation

"If I can buy a bucket of cognition for $1 instead of $100,000, why is the truck that big? What changes?"

"I feel like it's evolved from an expensive SaaS to a cheap worker. If you said you can hire somebody for a hundred dollars a month, you'd be like, yeah, that's excellent."

"I don't want the robots to give me nice documents on a plate and serve them to me as I sit in my armchair. Come on guys, if we're going to have robot labor, why are we going to these great lengths to make it easy for them?"

"Everything's just like in Slack and you have to constantly chase people to figure out versions because nothing's being managed in a sensible way anymore."

The Good Stuff, with Pete and Andy - Episode 20: Wingman - your AI Control Tower!

Hosts: Pete and Andy (recorded at City Beach, Perth)

Episode Overview: Pete unveils his breakthrough AI agent orchestration system "Wingman" - a Mac Mini-based control center for managing multiple AI agents simultaneously. The discussion explores custom-built solutions versus SaaS subscriptions, testing AI models through games, and the emerging paradigm of humans as "control tower operators" for AI systems.

Opening & Context Setting (00:00-01:48)

• Andy returns from illness for episode 20
• Final beach recording before Pete's mysterious Atlantic adventure
• Reflecting on recent guest episode with Bethanne

The Wingman Revolution (01:48-25:00)

• Pete's Mac Mini experiment: giving AI agents their own dedicated computer environment
• Evolution from basic setup to sophisticated web-based control system
• Multiple agent orchestration: Goose and Claude Code working in parallel sessions
• Remote access from anywhere - phone, laptop, or desktop
• Obsidian integration for documentation and artifact storage
• API endpoints for automated session creation and workflows

Agent-to-Agent Workflows (20:00-25:00)

• Return-to-base protocol for session completion and handoffs
• Webhook systems enabling autonomous agent chains
• Vision of fully automated code review and development cycles
• Moving from human-centered to machine-centered processes

Custom Tools vs. SaaS Fatigue (25:00-40:00)

• Economics of $15/month software subscriptions adding up
• Pete's approach: "Fast fashion for software" - build exactly what you need
• YouTube reel generator case study: custom-built vs. existing solutions
• The open-source developer superpower now accessible via AI

Management-Style AI Interaction (40:00-50:00)

• Transition from "button-pushing" to strategic oversight
• Using specialized agent personalities (Grug Brain Developer for code reviews)
• Dueling agents: building features in parallel and choosing the best
• Planning and auditing as core human value-add

Gaming as AI Benchmarks (50:00-01:09:00)

• Settlers of Catan as model testing environment
• Game mechanics as proxy for enterprise negotiation and resource management
• Tabletop wargaming automation concepts
• Testing strategic thinking, negotiation, and long-term planning capabilities

The Control Tower Paradigm (01:09:00-01:13:00)

• Humans as air traffic controllers for AI agent fleets
• Multiplayer mode implications for teams
• Terminology evolution: logbooks, flight plans, control towers

Dedicated AI Environments: Running agents on isolated hardware eliminates privacy concerns while providing full system access and persistent operation.

Agent Orchestration Architecture: The future involves managing 1-100 agents simultaneously across different projects and functions, requiring sophisticated coordination systems.

Custom vs. Commercial Software: AI development capabilities are shifting the economics from subscription services to custom-built solutions tailored to specific workflows.

Human Role Evolution: The transition from direct execution to strategic oversight and agent management represents a fundamental shift in how knowledge work gets done.

"I don't have a good way of explaining how just fucking nice this is" - Pete on Wingman's seamless operation

"The data is in the computer. You were in the computer. Talk to it." - Pete on the future of AI interaction

"It's the same way that junior developers fuck stuff up... we just need to use the same processes" - Pete on managing AI agents

"Fast fashion for software" - Andy's description of rapid custom tool development

Looking Ahead: Pete's system represents a glimpse into the future of human-AI collaboration, where individuals can operate sophisticated agent networks from anywhere, fundamentally changing the nature of knowledge work and business operations.

The Good Stuff - Episode 19: The Human Edge

Hosts: Pete and Bethan Winn
Location: City Beach, Perth

Episode Overview: Pete hosts with special guest Bethan Winn, his wife and author of "The Human Edge: Critical Thinking in the Age of AI." They explore the intersection of human thinking and artificial intelligence, discussing when and how to use AI tools while maintaining human agency and creativity.

Book Link: The Human Edge: Critical Thinking in the Age of AI

Key Discussion Points:

Introduction and Book Launch (01:01-04:00)
Bethan discusses her motivation for writing "The Human Edge" making critical thinking training more accessible as AI becomes ubiquitous. Her five years of corporate training experience revealed growing need for critical thinking skills in the AI era.

The Writing Process (04:00-07:00)
First-time author experience, early morning writing sessions, and using dictation to capture ideas. The challenge of moving from workshops to written format with AI assistance.

AI as Cognitive Offloading (07:00-12:00)
Comparison to historical abstractions like GPS replacing map knowledge and software development layers. The "AI veganism" concept - consciously choosing when to use AI versus manual processes for different purposes.

Pete's AI Development Experiments (12:00-18:00)
Multi-agent conversations, facilitator roles, and panel discussions between different AI models. Discovery that AIs can generate genuinely new concepts through interaction, challenging assumptions about creativity.

Human vs AI Creativity (18:00-25:00)
Debate about manufactured music, authenticity, and the value of creative processes. Whether AI-generated content diminishes human artistic expression or simply changes the distribution landscape.

Critical Thinking in Practice (25:00-32:00)
Using AI conversationally to challenge thinking rather than replace it. The book's focus on structured conversations and frameworks for better decision-making.

The Value of Difficulty (32:00-40:00)
Importance of uncomfortable experiences for growth. Pete's development of mobile AI tools that enable work while walking, maintaining human movement and reflection.

Knowledge Democratization (40:00-48:00)
How AI breaks down gatekeeping of specialist knowledge. The shift from information scarcity to the need for better filtering and evaluation skills.

Productivity and Attention (48:00-55:00)
OODA loops in software development, the challenge of managing multiple projects when AI accelerates execution. Social media's impact on attention and the value of deliberate focus.

Social Media Censorship Debate (55:00-01:19:00)
Discussion of Australia's social media ban for children. Arguments for education over restriction, the dangers of information control, and the importance of teaching critical evaluation skills.

Belief Change and Growth (01:19:00-01:26:00)
The "ocean analogy" for confronting uncomfortable truths - the discomfort of changing deeply held beliefs compared to entering cold water, but the growth that follows.

Human Connection in the AI Age (01:26:00-01:32:00)
The continued importance of empathy and understanding different perspectives. Bethan's approach of assuming the best in people and the value of genuine human connection.

Closing Thoughts:
"The AI doesn't care about the outcomes that you get, but you do" - emphasizing human agency in an AI-augmented world.

Book: "The Human Edge: Critical Thinking in the Age of AI" by Bethan Winn

The Good Stuff, with Pete and Andy - Episode 18: Building and Shipping

Hosts: Pete and Andy (recorded at City Beach, Perth - the van is back!)

Episode Overview: A follow-up to their accountability session from last week, Pete and Andy report back on actually shipping products to completion. This episode covers their successes with Fat Controller and Hype Man, explores new development workflows with Claude Code, and discusses the evolution of AI-assisted coding practices.

• Follow-up from last week's therapy session about finishing projects
• Pete releases Fat Controller 20 minutes before recording
• Andy gets Hype Man operational as an autonomous CMO
• The importance of shipping something people can actually use and pay for

• Built to solve Pete's timezone posting problem on Nostr
• Features: GIF support, satellite CDN integration, secure keychain storage (Mac only)
• Evolved from local-only to supporting remote signers for hosted version
• Redesigned signing architecture for better remote compatibility
• Released as open source with value-for-value model

• Andy's first official week on Nostr - hasn't felt compelled to check Twitter
• More invigorating and motivating environment
• Zap payments add meaningful interaction layer
• Less negative energy compared to Twitter/X

• Multi-agent system that extracts insights from podcast transcripts
• Creates and publishes social media content automatically
• Iterative improvement process: removed research agent, added review agent
• Cost-constrained testing with DeepSeek, upgraded to Claude for better output
• Reflection agents as key pattern for improving AI output quality

• Claude Code excelling at small, targeted changes
• Cursor better for complex, orchestrated implementations
• Claude Code's agentic loops provide better multi-step problem solving
• Both tools serve different phases of the development process

• Mac Mini as dedicated AI development server with 24/7 availability
• Terminal-based workflows accessible from phone via voice transcription
• Menu-driven system for common development tasks
• Segregated environment philosophy: give AI full access to designated machines
• Mobile coding: voice-to-terminal transcription enabling "vibe coding" from anywhere

• Three modes: drunk idea capture, sober editing, drunk coding
• Need for seamless idea-to-implementation pipeline
• Tutorial mode: AI explaining reasoning behind architectural decisions
• Balancing speed vs learning in AI-assisted development

• Danger of spending more time optimizing AI workflows than building actual products
• "Sharpening the axe forever" - when process improvement becomes procrastination
• Finding balance between efficiency and productivity

• Debate: Transform existing businesses vs. build AI-native competitors
• Unit economics advantage of AI-native companies
• Venture funding opportunities for dramatically improved cost structures
• Private equity interest in AI transformation strategies

• Weekend project to replace Screen Studio subscription
• Real-time vs post-processing architecture decisions
• Part of broader strategy to reduce SaaS subscriptions and build local tools
• Working prototype with keyboard shortcuts and lightweight toolbar interface

• Vision for idea-to-prototype pipeline
• Speed Run workshop concept: build entire company in one day
• Integration of existing tools (Pipeline, Wingman, Claude Code) into cohesive workflow
• Vibe startups: from idea to launched product with full departments

Hosts: Andy and Pete (recorded on the balcony at an undisclosed location)

Episode Overview: Pete and Andy explore the challenges of shipping projects in the age of AI, discussing their experiences with Claude Code, the paradox of infinite possibilities versus completion, and strategies for moving from ideation to deployment.

Opening & AI Agent Development (00:00-12:00)The hosts discuss Pete's "Hypeman" project - an autonomous agent that ingests podcast transcripts, conducts web searches, and tweets throughout the day. They explore the challenges of managing Claude Code's tendency to over-engineer solutions and add unnecessary features.

Claude Code vs Planning Environments (12:00-24:00)Discussion of using different environments for planning versus execution. Claude Desktop for read-only planning and code review, Claude Code for implementation. The "dog with two dicks" analogy - Claude Code's overwhelming enthusiasm to build everything immediately rather than following structured plans.

The Old Bull and Young Bull Framework (24:00-30:00)Pete and Andy introduce their management philosophy for AI agents: needing both the enthusiastic "young bull" (execution agent) and the wise "old bull" (planning/orchestration agent) to achieve optimal results.

The Completion Problem (30:00-40:00)Deep dive into the challenge of finishing projects when AI makes starting new ones so easy. The dopamine hit of new ideas versus the grind of deployment. Recognition that they're developing sophisticated vibe coding setups but struggling to ship finished products.

System Development vs Product Shipping (40:00-50:00)Exploration of whether time spent refining AI coding systems is valuable capital development or procrastination. Discussion of calendarizing projects and creating external deadlines to force completion.

Cost Optimization in AI Agents (50:00-56:00)Pete shares how iterating on Hypeman reduced costs from $20-30 per transcript to 50 cents by strategically using different models (DeepSeek vs Sonnet 4) for different tasks.

Social Media Strategy & Nostr Focus (56:00-01:08:00)Comparison of engagement across platforms - genuine human interaction on Nostr versus bot-heavy Twitter. The value of being able to send sats for valuable content. Discussion of LinkedIn's corporate facade versus Twitter's AI innovation content.

The Australian Tech Landscape (01:08:00-01:14:00)Concerns about government surveillance, economic dependency on government spending, and the contrast with AI's promise of infinite leverage for entrepreneurs.

Action Items & Accountability (01:14:00-01:15:36)Commitment to ship projects by next episode. Plans for content recycling tools and increased Nostr presence. Recognition that shipping, not starting, is their bottleneck.

"The cost of doing something goes to zero, then the only real constraint you have is what you should do."

"I feel like the bit that makes it the wrong idea is the opportunity cost... we just narrowed that window down to a few hours."

"You've got to go and spend all your time and energy in the system that you want to win."

Next episode will feature accountability check-ins on shipped projects and progress on deployment systems. The hosts commit to focusing on completion over ideation.

Episode Theme: The challenge isn't building with AI - it's knowing what to build and actually shipping it.

# The Good Stuff Episode 16 - LLM Subscriptions, Goose AI Agent, Beacon & Bitcoin

**Summary:**

In this episode, hosts discuss subscription-based AI services and their rate limits, sharing their frustrations with Claude and Cursor's recent policy changes. They explore their personal workflows using various AI tools for coding, writing, and idea development, discussing the benefits of using multiple specialized tools versus all-in-one solutions.

The conversation concludes with an in-depth look at "Beacon," a project designed to bring AI tools and Bitcoin functionality to users in areas with limited internet access through platforms like WhatsApp.

## Key Themes:

**Subscription Models and Rate Limits in AI Services** (00:00 - 27:00)

- Discussion of Claude implementing weekly rate limits

- Comparison with Open AI's subscription model

- The economic realities of providing AI services

- Issues with communication and "bill shock" when companies change terms

**Personal AI Tool Workflows and Setups** (27:00 - 50:00)

- The hosts' approaches to using different AI tools for specific purposes

- Discussion of Claude desktop vs API performance differences

- Challenges with integration across different platforms

- Running local models on dedicated hardware

**Agent-Based Writing Tools and Philosophy** (50:00 - 73:00)

- Building dialogue-based writing companions versus editing tools

- The importance of focusing on the philosophy behind writing

- How agent-based conversations can help develop ideas more effectively

- Creating systems that don't interrupt creative flow

**Vibe Coding and Game Development** (73:00 - 84:00)

- Using voice-based coding for game development

- The potential of AI for nostalgic game recreation

- Making coding more accessible through voice interfaces

- Teaching coding through game development

**Beacon: AI and Bitcoin for the Underserved** (84:00 - 1:14:00)

- Introduction to Beacon, a system bringing AI and Bitcoin to users via WhatsApp

- How it works around censorship and limited internet access- Creating community financial tools for areas with limited banking

- The importance of bringing information freedom to underserved populations

Episode 15- Moar Vibe Coding Talk!

Summary: In this episode, Andy and Pete discuss their evolving coding workflows with AI tools, focusing on how Claude Desktop and Cursor have created more structured and efficient development environments.

They share insights about implementing sequential prompting methods, organizing work in Obsidian, and using AI-driven panel discussions to explore ideas.

The conversation highlights how AI tools are changing their approach to productivity and shifting focus from execution to more fundamental questions about project purpose.
00:11 - 05:47: Improved AI Coding Workflows

• Discussion of Claude Desktop for ideation and technical planning
• Implementation of sequential prompting with validation checks
• Creating structured folder systems in Obsidian for better organization
• Example of 3GS scene development using this workflow
  

05:48 - 11:00: "Vibe Coding" vs Structured Approaches

• Comparing different coding approaches with AI assistance
• Benefits of implementing checks and balances in code generation
• Creating change logs and indexes to track progress across projects
• How this system helps manage multiple concurrent projects
  

11:01 - 15:36: AI Environment Improvements

• Discussion of GitHub MCP setup for code checking
• Benefits of super whisper for voice control
• Plans to run local models on a new Mac Mini
• Using Goose as an agent on dedicated hardware
  

15:37 - 23:23: AI Panel Discussions for Knowledge Generation

• Using AI agents to have conversational exchanges about topics
• How dialogue between AI models can generate new insights
• Testing different models (Claude, Grock) to identify bias patterns
• Value of these synthetic discussions for idea exploration
  

23:24 - 31:20: Shifting Focus from Tasks to Purpose

• Discussion about capturing organizational knowledge from meetings
• The limitations of task-focused productivity tools
• How AI automation shifts focus from execution to purpose
• The importance of asking "why" rather than just "what" and "how"

# Episode 14: Vibe Coding that works!

In this episode, the hosts discuss their development of "slow code," a more intentional approach to coding with AI that contrasts with "vibe coding." They explore how this methodology creates a structured workflow combining human planning with AI execution, resulting in higher quality code while maintaining the speed benefits of AI assistance.
## Key Themes:
**Understanding Vibe Coding** (00:02:20 - 00:08:00)* Vibe coding involves speaking instructions to AI and ignoring the underlying code* Democratizes coding by allowing anyone to build in natural language* Major downside: potential security issues and poor code quality
**Evolution to Slow Code** (00:08:00 - 00:19:40)* Hosts developed methodologies through experimentation with AI tools* Initial frustrations with existing AI coding tools led to refinement* Slow code combines intentional planning with AI execution
**Three-Phase Methodology** (00:19:40 - 00:25:30)* Ideation: Using Claude Desktop for exploration and brainstorming* Planning: Converting ideas to structured documents and roadmaps in Obsidian* Execution: Having AI implement based on detailed specifications
**Tools and Implementation** (00:25:30 - 00:40:00)* Using Obsidian for planning and documentation* Connecting to AI tools via MCP (Model Context Protocol)* Root Code for implementation with orchestration capabilities* Benefits of separating environments for different phases
**Multi-Agent Conversations** (00:40:00 - 00:51:00)* Creating pipelines where multiple AI agents discuss ideas* Using different models (Claude, Grok) to avoid single-model biases* Value in seeing how ideas develop through agent conversations
**Benefits of Slow Code** (00:51:00 - 01:01:30)* Creates reusable, higher quality code versus one-shot vibe coding* Maintains human intentionality and design control* Still much faster than traditional development* Parallels the "write drunk, edit sober" approach to creative work
**Future Applications** (01:01:30 - 01:15:00)* Integration with Goose for continuous development and scheduling* Creating dedicated AI development environments* Applications beyond coding (writing, research, analysis)* Plans to share methodology through tutorials and community
**Contrasts with No-Code Tools** (01:15:00 - 01:26:00)* Visual no-code tools add unnecessary complexity* Slow code leverages AI's strengths while maintaining human oversight* More flexible than constrained visual interfaces