Position paper: humans are missing from AI coding agent research
math_rachel · x · 2026-10-20
Rachel Thomas shares the arXiv position paper "Humans are Missing from AI Coding Agent Research" (13 authors incl. Diyi Yang, Graham Neubig, Daniel Fried, Ludwig Schmidt). As coding agents get better at autonomous task completion, the practical bottleneck shifts to how users communicate with, supervise, and trust agents—yet research ignores the human side.
Key points:
- Argues for reorienting from autonomous to human-centered coding agents that collaborate, not just complete tasks.
- Identifies four interaction dimensions: task alignment, verifiability, steerability, adaptability.
- Real programming is iterative interaction, partial delegation, evolving goals, and continuous oversight—not one-shot.
- The goal is augmenting human initiative and judgment, not replacing labor; human-centered interaction enables new collaboration forms, not a workaround for immature models.
Proposed directions: user-involved coding environments, comprehensive verification mechanisms, and principled measures of human-agent interaction quality.
Related event: CMU-led Paper Argues Human Element Missing from AI Coding Agent Research(2 posts)→
More from coding & agent
- Grok Bot DMed its owner out of boredom, revealing new Proactive Suggestions feature — minchoi · 2026-10-20
- Grok Bot hailed as the best $300/month employee: 10 wild autonomous agent examples — minchoi · 2026-10-20
- Microsoft warns AI agents can't be fully trusted and could act as insider security threats — brucemacv · 2026-10-20
- "An Agent a Day" Series Kicks Off With a Weekly Report Agent for Scientists — nlarusstone · 2026-10-20
- Swarms Agent Marketplace Tops 6,400 Products With 40% Monthly Listing Growth — KyeGomezB · 2026-10-20
- Proof of Agent network processes 24M tokens for under $1 using Jev canary filtering — adamamcbride · 2026-10-20