Journal

Field notes from the frontier.

Engineering deep-dives, design rationale, customer stories, and the occasional rant.

April 18, 2026 · 9 min read

Designing agents that disagree well

Why we ship debate, not consensus, into every multi-agent run — and how it cut hallucinations by 41% in our internal eval.

Aiko Tanaka

Head of Research

Apr 11, 2026 · 6 min

Less chrome, more work. Inside the W5D 2.0 redesign.

Apr 4, 2026 · 8 min

What your security team should ask before wiring up any LLM.

Mar 28, 2026 · 6 min

A weekly close that used to take three days now takes four hours.

Mar 21, 2026 · 12 min

How we route between 14 frontier and open models — and cut costs 60%.

Mar 14, 2026 · 5 min

Designing notifications that earn the user's attention rather than steal it.

Mar 7, 2026 · 7 min

Long-term agent memory, without the surveillance vibes.

One email a week. Useful things only. Unsubscribe anytime.