Agentic Societies Need a Social Harness: Why Agents Can't Yet Collaborate Across Trust Boundaries
shishirpatil_ · x · 2026-09-18
Researchers Tapan Chugh, Arvind Krishnamurthy and Ratul Mahajan published a blog post arguing that AI agents cannot yet collaborate autonomously across trust boundaries (e.g., talking to other people's agents).
- Today: agents serve a single user while humans carry the messages — e.g., an advisor's agent flags a relevant paper, but a human still emails it to the student, whose Claude project agent compares it against the literature, and a human replies.
- The vision: agents interact directly and humans are notified afterwards, but this requires solving hard efficiency and security problems.
- Core claim: these problems closely mirror those of human societies and of distributed systems and networking, so agent societies need a dedicated "social harness" — infrastructure the authors are starting to build.
- The post links to an accompanying paper and frames itself as an early look at the field.
More from coding & agent
- Kimi Agent Swarm tested in the cloud: 16-hour autonomous training run with self-recovery — solyarisoftware · 2026-09-18
- Open-source PR Review Lab spins up sandboxes to compare GPT 5.6 Luna vs Jev reviewers — aniketmaurya · 2026-09-18
- One customer drives $10k MRR as indie AI agent product doubles September revenue — tibo_maker · 2026-09-18
- The real agent security weak point is over-privileged access, not faster exploits — code_star · 2026-09-18
- Solo dev ships hosted CVE + CISA KEV + EPSS MCP server, no API key needed — sartomiki · 2026-09-18
- Agent runs autonomously for 1.5 days to train a dexterous-hand pen-spinning RL policy — SatelliteNetSec · 2026-09-18