Your team is building a structured data extraction pipeline that processes 10,000 legal contracts per day. The current architecture uses an agent team where a classification agent, an entity extraction agent, and a clause validation agent all coordinate via a shared task list — messaging each other directly to reconcile conflicting field interpretations before writing to the output schema. Post-deployment analysis shows per-document token costs are 4× higher than projected, and P95 latency has grown to 12 minutes per document. Logs confirm that each agent's deliberation is thorough and accurate, but the inter-agent messaging is extensive. What architectural change would most directly address the cost and latency problem?