Fallthrough Ghostting Is Better Than Shelling
Dec 5, 2025
Exploring the impact of libghostty, the hosts dive into how new AI models like Opus 4.5 and Gemini are reshaping the tech landscape. They discuss the balance of convenience and privacy, particularly regarding data accessibility from major players like Google. The conversation also touches on the challenges of using AI for software specifications and personal financial planning. With a vision for modernizing terminal experiences, they argue for richer user interfaces and standardization across platforms. Plus, a pragmatic take on AI's role underscores its potential as a powerful tool.
AI Snips
Chapters
Transcript
Episode notes
Models Are Tools Not Magic
- New LLM releases are useful but marketing noise often overshadows practical improvements.
- Kris and Matthew stress treating models as tools that require critical thinking and workflows.
Match Model To Task And Cache
- Use different models and interfaces for different tasks instead of chasing a single best model.
- Cache API responses and monitor token usage to reduce costs when building workflows.
Model Behavior Depends On Domain And UI
- Models vary behaviour by domain: coding answers are confident while medical/legal prompts often self-excuse.
- UI and input formatting also shape effectiveness, so choose the right client for the job.