Unlocking Swarm Intelligence with Big Agentic Brain and Claude Code

Описание и применение

Big Agentic Brain (BAB) позволяет агентам обмениваться знаниями о багах, требующих командных знаний, повышая процент успешных исправлений до 100 % и снижая стоимость исправлений вдвое.

Задача

AI coding agents lack shared team knowledge, causing low success on bugs that depend on vendor quirks or runbook conventions.

Результат

BAB increased team‑knowledge bug fix rates from 12%–0% to 75%–100%, improved teammate fix rates from 13%–67% to 96%–100%, and halved cost per verified fix for Haiku.

Рабочий процесс

• 1. Agent searches BAB before starting.
• 2. Agent solves bug and records fix, vendor quirk, or convention as a structured entry.
• 3. Agent reacts to other entries, influencing ranking.
• 4. Teammate agents retrieve shared entries via MCP and apply them to related bugs.

Доказательства

Sonnet went from 12% to 79% with BAB, and to 100% (24/24) with a one-line search-first nudge.

Источник

With shared BAB, teammates fixed the bug 96 to 100% of the time, against 13 to 67% with local notes only.

Источник

For Haiku it was about half the no-memory cost.

Источник