PwC’s AI agents are now your consultants — whether you’re ready or not

PwC's move to deploy AI agents directly to clients signals a fundamental shift in how consulting outputs—and by extension, data architecture decisions—will be validated. For data teams, this means consulting recommendations will increasingly bypass human review, creating new failure modes we need to architect around. We're accustomed to challenging consultant assumptions through dialogue; an AI agent won't engage in that pushback. From an operational standpoint, this accelerates the need for robust data contracts and observability frameworks. If an AI consultant recommends a medallion architecture or specific orchestration tools, we need instrumentation to quickly validate or reject those suggestions empirically. The broader trend here intersects with platform engineering: as AI agents become decision-makers in technical architecture, data teams must invest in clear metrics, automated testing, and guardrails that prevent bad recommendations from reaching production. My recommendation is straightforward—start treating AI-generated architecture recommendations with the same skepticism you'd apply to junior consultants, then build the monitoring and validation infrastructure to prove or disprove them automatically before implementation.