Dev Interrupted Sloppypasta culprits, unpacking MCP’s spotlight, and Anthropic wants your agents to work the graveyard shift
46 snips
Mar 20, 2026 They debate whether the Model Context Protocol was overhyped and why CLIs are making a comeback. They explain context anchoring to prevent model compaction during long coding sessions. They warn about AI amplifying bad processes when teams optimize the wrong bottlenecks. They celebrate the nostalgic resurgence of the decentralized small web. They call out ‘sloppy pasta’ and how to avoid it.
AI Snips
Chapters
Transcript
Episode notes
Joke About Working Nights For Cheaper Tokens
- Andrew and Ben joked Claude suggests working nights because token costs drop at off-peak times.
- They imagined 'rolling token blackouts' and collecting tokens like solar energy to use when prices spike.
MCP Solved Distribution Not Everything Else
- MCP was hypey but solved a real distribution problem for tool usage across agents.
- It shined when models were weaker by ensuring consistent tool access, but today lighter CLI-based and token-optimized approaches often suffice.
Anchor Context To Prevent Model Compaction
- Use context anchoring to avoid session compaction and preserve reasoning across long AI coding sessions.
- Capture a living feature document or lightweight substrate that records decisions, why they were made, and relevant constraints for the model to reference.