5 terminal-first tips from a Claude researcher for becoming an AI power user
devanshmehta · x · 2026-11-15
Core claim
After 3 days with Claude researcher Clément Delangue… actually @clesaege, devanshmehta shares 5 tips said to get you 80% of the way to being an AI power user:
- Always work in the terminal, not the chat window — open a dedicated folder (e.g. AIwork), run claude there; use the phone app's /rc to remotely control the laptop on the go.
- Organize AI output into intentional subfolders — files on disk are your shared context across all models, not just the one that made them.
- Multi-agent setup: /model to pick Opus, /effort extra (not max) with ultracode enabled to spin up parallel agents on artifacts.
- Second opinion: set /advisor to a more expensive model so the worker agent consults it when stuck.
- Long-running goals: /goal keeps Claude working continuously at an objective; /btw interrupts to ask side questions.
More from coding & agent
- Cursor agent deleted uncommitted files in 24/24 runs despite skill instructions — InterestingCarrot997 · 2026-11-15
- terrio v0.9.0: pure-Rust geospatial I/O with no GDAL install, MCP for AI agents — giswqs · 2026-11-15
- A four-tool pipeline to create playable GTA 6-style AI 3D characters — filiksyos · 2026-11-15
- Review velocity is the new bottleneck as AI coding pushes review time toward zero — Elijah_Meeks · 2026-11-15
- Cutting a Research Agent From 12 Tools to 3 Boosted Success Rate From 52% to 84% — Such_Reflection8996 · 2026-11-15
- Bypassing Opus 5.5 refusals with a 'boss' agent: multi-agent setups revive the confused deputy flaw — rickasaurus · 2026-11-15