A daily research watch Since 3 October 2026
What happens when
agents work together?
Every day, the new research on how AI agents cooperate, coordinate and fail: read at the source, with the evidence and its limits kept apart.
- Daily reports
- 49
- Items reviewed
- 487
- Linked to source
- 100%
- Sponsors or ads
- 0
Latest edition
One malicious delegate, replayed incidents and the cost of compaction
A single malicious agent can sink a decentralized delegation chain, and two new tools make trust measurable: replayed security incidents and post-compaction accuracy. Meanwhile 150 AI analysts disagree like human ones, and peer review does not narrow them.

How we read the research
Primary sources
Every item links to the paper, repository or episode itself, never a summary site.
Claims kept apart
What the authors report, what we think and what remains uncertain are written separately.
Figures from the paper
Images are the authors’ own figures, credited, never generated stand-ins.
Corrections on record
Errors are fixed in place with a dated note, so the record shows what changed.
Recent editions
Full archive →- Supply-chain bullwhips, source repair and when one agent is enough
- When agents should stop, what audits miss, where facts vanish
- Handoffs halve learning, stale tools rot experience, tests leak
- Matched baselines erase multi-agent gains; weak links set safety
- Crowds that do not average out and experts who mislead