Blog — Agent Design Notes & Research
Thoughts on agent design, research outcomes, and experimental records from building the Actoviq Agent SDK. Less "how to use the API," more "what we tried, what we measured, and what we concluded."
Posts
Model Team & Multi‑Model Agent Design — A Research Report
2026‑06‑17 · design + experiment
Does letting an agent convene a team of models actually produce better results? Across 48 graded runs with multi‑dimensional scoring and 3× trials, the autonomous expert panel proves quality‑neutral on average — with a consistent citation/structure edge at a ~20% cost premium — and the agent convenes it on its own ~39% of the time. Includes the Model Team design, the two‑track benchmark methodology, full results, and five design lessons (forcing collaboration backfires; single‑run benchmarks lie; composites hide trade‑offs; verify execution objectively; beware demand‑induced fabrication).
About this blog
Posts live under docs/blog/ as Markdown files with front‑matter:
---
title: Your Post Title
date: 2026-06-17
---
# Your Post Title
*Published 2026‑06‑17 · short kicker*
…To publish a new post: add the file, then link it from the Posts list above and register it in the blog sidebar (docs/.vitepress/config.mts). English only for now.