Meta's Tribal Knowledge Pre-Compute: 50+ AI Agents Generate 59 Context Files, Cut Agent Tool Calls 40%

Engineering at Meta·high signal

Meta solved the 'agents can't navigate large codebases' problem by deploying 50+ specialized AI agents that systematically read every file across 4,100+ files in three repositories, producing 59 concise context files encoding tribal knowledge. Coverage jumped from 5% to 100% of modules, and preliminary tests show 40% fewer tool calls and tokens per task. The system is model-agnostic and self-maintaining via automated validation jobs.