Dev Interrupted Virtual pets in your terminal, ads in your pull request, & no more CSS in your browser?
32 snips
Apr 3, 2026 They debate ads appearing inside pull requests and the trust problems that creates. They unpack a massive code leak that revealed hidden model features and implementation gaps. They explain Shopify’s play to cut AI inference costs by 75x with smaller self-hosted models. They explore Pretext’s trick for measuring text before the DOM and playful terminal “virtual pets” from leaked tooling.
AI Snips
Chapters
Transcript
Episode notes
Opt Out Of Copilot Training Data Immediately
- Opt out of Copilot data-sharing if you're not on an enterprise plan to prevent private repo data being used for training.
- GitHub defaults non-enterprise users to opt-in; uncheck the box before April 24th if you want privacy.
Self Hosted Small Models Cut AI Inference Costs
- Self-hosting smaller, task-tuned models can massively reduce inference costs while improving quality for narrow tasks.
- Shopify switched from a large paid model to Qwen 3 fine-tuned and multi-agent orchestration, cutting costs ~75x and doubling output quality.
Orchestration Enables Efficient Model Choice
- Task-level model selection plus agent orchestration reduces overall AI costs by using expensive models only when necessary.
- Ben Lloyd Pearson argues orchestration lets you pick mini models for simple tasks and reserve Claude/Opus for hard problems.