Agent Frameworks Considered Harmful — Rémi Louf, .txt
Agent frameworks force humans to stay on-task, undermining the promise of autonomous agents
“SSH with vibes”
Rémi Louf, CEO of .txt, argues that current agent frameworks fail at true autonomy — despite powerful underlying models (citing a step-function improvement around Claude Opus), tooling still requires constant human presence, like riding a self-driving tractor rather than leaving it unattended. He coins the critique 'SSH with vibes' to describe agent app interfaces that dress up manual supervision as automation. The talk is a practitioner-level signal that the gap between model capability and usable agent infrastructure remains the key unsolved problem.