The Good Stuff, with Pete and Andy - Episode 9: Tools, Tools, Tools

Hosts: Pete and Andy (with bonus ambient drum and bass from a nearby camper van)

We dive deep into the practical tools we're using for AI development, exploring the difference between AI as tools versus human-at-the-edge workflows, and discussing the technical complexity of building local AI systems.

Key Discussion Points:

The Cold Open: Screen Time and Digital Minimalism (00:00-05:52)**

Pattern Matching vs. Reasoning in AI (07:00-17:20)

Apple's recent paper questioning whether LLMs truly "reason" or just do sophisticated pattern matching
How thinking models work
The relationship between human thought and AI pattern matching
How AI systems handle novel problems and the role of entropy

AI Tools in Practice (18:50-32:00)

Why Cursor has gained such traction compared to alternatives
The importance of context management and local file access
Pete's experience with OpenAI's Codex vs. local tools like Cline and Cursor
The dopamine feedback loops that make certain tools more engaging

Local vs. Cloud AI Systems (30:00-40:00)

The benefits of running AI systems locally rather than in web apps
Avoiding the complexity of SaaS
How local processing leverages your computer's existing power and storage
The privacy advantages of keeping personal data on your own machine

Memory and Knowledge Graphs (38:52-50:00)

The limitations of basic RAG (Retrieval Augmented Generation) systems
Introduction to graph RAG systems that provide richer context
How Stakwork uses self-improving graph databases for better AI performance
The importance of solving the "memory problem" for effective AI systems
LLMs as translation layers between human language and structured data

Personal Knowledge Graphs (50:00-58:00)

Pete building a personal knowledge graph system
Using Docker containers and API interfaces for local AI development
The challenge of managing context across multiple AI tools and workflows
Bethan's book "The Human Edge"

Building AI Systems: Technical Complexity (58:00-01:10:00)

How accessible it is for non-developers to build AI systems with current tools
The "slow code" approach: treating development as a learning experience
Apply Git liberally!
Andy's experiments with N8N for workflow automation and content creation pipelines

Workflow Automation vs. Autonomous Agents (01:16:00-01:22:00)

Comparing deterministic workflows to autonomous agents
Why most business tasks are better suited to static workflows
The role of humans in AI systems: providing intent and experience
Enumeration vs. abstraction: building specific workflows rather than trying to create universal solutions

Development Stack and Tools (01:13:00-01:16:00)

Pete's current toolkit: Cline, Visual Studio, O3, Claude, Codex, Code, Personal Graph....
Plans for a local Nostr-based ebook reader with cross-device syncing
Paying for all the tools!

Conspiracy Corner: Moon Mysteries and Dead Internet Theory (01:25:00-01:35:00)

Discussion of moon landing anomalies
Dead Internet Theory and how algorithms shape both content creation and consumption
The decline of film quality
AI-generated content and the future of creativity

Key Quotes:

"The job of the LLM isn't to be everything... what LLMs are specifically good at is translating stuff into and out of human language"

"The price of bullshit is also dropping to zero"

"We're not here to raise low agency children!"

"Everyone loves a sausage!"

Was Benjamin Franklin America’s First Scientist?

The Michael Shermer Show

Franklin versus Jefferson: commerce, cities, and paper money

The AI-powered Podcast Player