Reconcilable Differences 280: How Dangerous I Can Be
Feb 12, 2026
They geek out over RFCs, machine-naming rituals, and the early web tools that shaped their curiosity. They argue about agents that write code, prompt engineering, and how half-baked skills can be dangerous. Practical tinkering appears in a long dive into sink holes, soap dispensers, and clever repurposing ideas. Side threads touch on passkeys, tiny sites, and the Australian Open 1 Point Slam.
AI Snips
Chapters
Transcript
Episode notes
HyperCard And BBEdit Sparked A DIY Web Habit
- Merlin recounts early HyperCard and FileMaker tinkering that taught him web publishing and semi-relational databases.
- He used BBEdit includes to generate pages from a CD-based exhibitor database, sparking his lifelong joy in DIY tooling.
Match Tasks To Tool Strengths
- When you use a tool, learn what it's good at and design work to fit those strengths rather than forcing it to be everything.
- Merlin compares this to vacation planning: match expectations to the resort's strengths to avoid frustration.
Advanced Beginners Gain Outsized AI Leverage
- Advanced beginners extract disproportionate value from AI coding agents because they know enough to ask for useful outputs but not enough to overfit complex agent workflows.
- Merlin warns that 'a little learning is a dangerous thing' with these systems.