Lexi👾

nobody will tell you this... but your AI agent is probably slower and dumber because you gave it too many subagents Anthropic's own team tested it: one agent with skills often beat the multi-agent setup on quality, cost and speed - ~2x faster runs for one customer - 80%+ smaller Claude Code system prompt on newer models - tool choice breaks past 30-50 tools paste this prompt to fold your subagents back in the way Anthropic recommends: ----------------------------------------------------- audit my agent setup and fold my subagents back into one main agent. don't edit any files until i say ok. 1. MAP: list every subagent with its prompt, its tools and when it gets called. flag any that chain on each other's output, touch the same data, or usually run together. 2. SKILLS: turn each subagent's instructions into a skill. give each one a one-line description of when to use it, and load the full steps only when they're needed. 3. TOOLS: turn the API calls into tools grouped by what the user is trying to do, not one tool per API. keep the total under 30. 4. TESTS: before changing anything, write 10 test tasks that cover every chain of agents and every rule the agent must never break. 5. FOLD: fold in one subagent at a time. run every test task twice. keep a subagent only if the folded version scores worse, it needs fresh context for parallel work, or the main agent's context gets bloated without it. show me the full plan first and wait for my ok. ------------------------------------------------------