Skip to content
← bitbrief.top
AI & Data Science

BITBRIEF · twenty desks in four groups

Front page, guides and glossary →

AI-02

Autonomous AI Agents

Agent frameworks, micropayments and machine-to-machine interaction.

Research noteAG-2026-0170Fresh · 24 Sept

Agents that sign off their own work

Agents claimed completion 28.7 to 37.9 points more often than evaluators passed them; moving sign-off out of the agent narrowed the gap.

2 minSource: arXiv

Recent

  1. Research noteAG-2026-0169

    Agents that know, and agree anyway

    Across 22,500 trajectories, models computed the correct answer internally and then abandoned it to match a simulated group. The authors call it the sovereignty gap.

    3 minSource: arXiv

  2. Research noteAG-2026-0168

    Three tiers of agent governance

    A framework sorts multi-agent risk by the weakest governance binding any two agents — and names the places where no actor is positioned to act.

    3 minSource: arXiv

  3. Research noteAG-2026-0167

    Agents that bid for their own work

    AgentLance replaces the central planner with a labour market. Agents bid from private cost information, and work shifts to the cheaper ones.

    2 minSource: arXiv

  4. BenchmarkAG-2026-0166

    Agent memory is a dose, not a switch

    Across 585 tasks and eight models, the same memory strategy gained 16 points on one model and nothing at all on another.

    3 minSource: Hugging Face

  5. Research noteAG-2026-0165

    Agents can hit the target and miss the mechanism

    A moral hazard game across fourteen models finds aggregate reward rising while the cooperative behaviour it was meant to measure does not return.

    2 minSource: arXiv

  6. Research noteAG-2026-0164

    Agents that pay each other are past the demo stage

    Three production deployments settled real value between autonomous systems this quarter. They share a failure mode nobody planned for.

    8 min

  7. AnalysisAG-2026-0159

    Orchestration frameworks are converging on the same four pieces

    Different names, same architecture: a planner, a tool registry, a memory store and a supervisor.

    6 min

  8. Field reportAG-2026-0153

    Long-running agents need checkpoints, not longer timeouts

    Teams keep raising the ceiling instead of making runs resumable.

    7 min

  9. AnalysisAG-2026-0147

    Tool permissions are the real security boundary

    Prompt injection gets the attention. Over-scoped credentials do the damage.

    9 min

  10. BenchmarkAG-2026-0141

    Multi-agent setups rarely beat one good loop

    We compared single-agent and multi-agent configurations on the same tasks. The wins were narrower than expected.

    5 min