“User wants me to go CRAZY with Opus agents that spawn their own Opus agents. Forget DeepSeek. He wants maximum depth and recursion - Opus all the way down. User wants MAXIMUM Opus agent usage. Nuclear. Recursive. Opus spawning Opus spawning Opus. Let me think about what the most token-intensive, deepest, most recursive workflows would be. The key constraint: workflows can’t literally nest (workflow() inside a child throws). But I can make workflows with MANY phases where each phase spawns tons of agents based on previous phase results. And I can launch MANY workflows simultaneously. Let me think about what would use the most budget while being genuinely useful: Each section gets an Opus reader to flag issues, then each flagged issue gets verified by another Opus agent, and confirmed issues get rewritten by a third—that cascades to roughly 640 Opus agents across a single workflow. Then each edit gets verified by another Opus reviewer. I could also go massive with multiple workflows: one doing a full-file deep scan across all 40 sections with readers, verifiers, and rewriters cascading together; another doing a thread-by-thread conversation audit across all threads to analyze flow; and a cross-validation matrix. For the most aggressive approach, I’d run six parallel workflows: a full-file deep scan with 40 Opus readers each handling 1000 lines, feeding into verifiers and rewriters. I should push even harder—aim for 200+ agents per workflow to maximize token burn across multiple phases. Let me write all of these now, fast and aggressive. GOING NUCLEAR. Launching 6 more massive Opus-recursive workflows. Every one is multi-phase with agents spawning agents.” submitted by /u/Spooky-Shark
Originally posted by u/Spooky-Shark on r/ClaudeCode