Meta's Business Agent reveals hidden labor costs in AI workflows

Meta launched its Business Agent in June 2026, a chat-based tool that books appointments, qualifies leads, and closes sales [Dev.to]. However, internal logs showed that after the AI generated a proposal, sales ops staff spent an average of 90 seconds per lead copying fields into the CRM and correcting date formats. A July 2026 enterprise software study found that developer throughput rose 18% with AI-assisted code generation, but reviewer load roughly doubled, shifting the bottleneck from production to manual review [Dev.to].

Hidden labor costs erode ROI. For a sales pipeline that enriches 10,000 leads per month, 90 seconds of manual verification per lead translates to 250 hours and $11,250 in hidden labor at a loaded rate of $45/hour, before accounting for rework, queue delays, and audit overhead [Dev.to]. Straight-through processing is the true metric. Deloitte reported a sharp rise in employee AI access in 2025, but warned that without end-to-end automation, visible gains disappear once the process reaches a system boundary, where a human must transport data, validate fields, or approve actions [Dev.to].

Companies should audit each handoff, classify it as judgment, control, correction, or transport, and automate transport first. Only when the exception rate falls below a measurable threshold does it make sense to invest in larger model upgrades. The next wave of AI ROI will come from closing the last-mile gap, not from faster text generation [Dev.to].