AI agents are reaching production faster than organizations are defining accountability

Pascal Bornet highlights a widening operating-model gap in enterprise AI adoption. According to BCG’s AI at Work 2026 research, 30% of organizations have moved AI agents from pilots into live production workflows, while 50% of respondents say their companies still lack clear guidance on how humans and AI should work together.

Bornet describes this as a human-agent orchestration gap: deployment is accelerating faster than organizations are defining who owns the work, who reviews agent output and who remains accountable when an automated action fails.

The report also shows that productivity gains do not automatically become business value. Forty-two percent of regular frontline AI users report saving a full workday per week, yet 66% receive limited or no guidance on how to use the recovered time. Strategic clarity appears more important than tool access alone: employees with clear AI direction but limited tool access report measurable business impact more often than employees with strong tools but weak strategic guidance.

Supply-chain perspective: this gap is especially risky when AI agents influence forecasts, replenishment, supplier actions, inventory allocation, transport decisions or customer commitments. Every production agent should have an accountable process owner, an approved decision scope, review thresholds, escalation paths and a rollback procedure.

Organizations should stop treating login counts, prompt volumes and agent activity as sufficient adoption metrics. A governed deployment should measure the operational outcome improved, the baseline performance, the cost of intervention, the frequency of manual overrides, failure recovery and the financial or service impact of incorrect decisions.

The principal failure mode is automating an unclear operating model. When decision rights, priorities and accountability are unresolved, agents can encode and scale organizational ambiguity at machine speed.