Notes from our work building autonomous research systems for markets — the questions we're chasing, and what we've found so far. We publish what we can.
Feasibility was the easy part — an agentic loop can already discover better strategies. The harder question is design, and the answer surprised us: across five research paradigms, less debate, less context, and lighter coordination beat maxing out every dimension.
Read →Running a fleet of agents spends two currencies at once: wall-clock time and tokens. More agents in parallel finish sooner but burn more tokens; fewer agents are cheaper but slower.
Read →In almost every market, most participants lose, a few win a great deal, and it is very hard to see who is who, because nearly all of the trading is private. Hyperliquid is the rare exception.
Read →FunSearch and AlphaEvolve showed that an LLM in an evolutionary loop can rewrite code toward better solutions on deterministic targets — matrix multiplication, bin packing, compiler heuristics.
Read →