News

Updates from the lab

Ship notes, things we broke and fixed, and the occasional memo we’d show an investor without embarrassment. Press or bigger partnerships: start on the contact page.

All Product Lab notes Research Partnerships Engineering

Showing 3 of 15 stories

Jan 8, 2026Research

Research memo: when two agents disagree

Early findings from multi-agent task graphs tied to Sim scenarios.

We’re publishing an internal memo (summary here) on conflict resolution between orchestrated agents when recommendations diverge—especially in capex rehearsal flows. The short version: without explicit ownership of “break glass” decisions, humans tune out.

We’re testing UI patterns that force a single accountable choice with logged rationale, rather than averaging model outputs. More in Q1 lab notes.

Apr 10, 2025Research

Embedding eval harness for domain PDFs

Reproducible benchmarks for retrieval on policy and manual corpora.

We open-sourced a harness comparing chunking strategies and embedding models on redacted customer-shaped data. Goal: kill hand-wavy “RAG accuracy” claims.

Used internally on fintech compliance pilots; contributed fixes welcome.

Jun 18, 2024Research

Paper: Web3 when attestation matters, APIs when it doesn’t

Decision tree we use with legal and product.

We published a short technical note (PDF on request) mapping when chains add tamper evidence vs. when they’re theater. Spoiler: most logistics wins are signed logs + integration, not new tokens.

Used in med-logistics and industrial traceability conversations with auditors in the room.