How AI agents coordinate. And when they shouldn't.
Research Paper · Apr 2026
When to start communicating.
An adaptive gate that activates stigmergic coordination only under population-level distress signals outperforms both always-on and always-off baselines by +34% on the hard task across 10 seeds — and the same principle transfers to LLM agent collectives.
Read paper→
Pilot Study · Apr 2026
A growing memory field.
On a 10-seed, 300-task longitudinal delivery simulation, a shared growing memory field lifted a fine-tuned 9B model from 85.4% to 97.3% success. All 10 seeds positive, sign-test p = 0.00195. Pilot evidence; ablations pending.
Read paper→
Stay in the loop
New papers, pilot results, and things we're learning.
We publish a short note every time we run a new experiment or ship a finding worth reading. No cadence, no newsletter padding — just the work.