6 Questions Every Enterprise Has to Answer About AI

What It Covers

Presented at KPMG's 2026 tech symposium, this episode outlines six questions reshaping enterprise AI strategy as organizations shift from assisted AI to agentic AI. The conversation covers architecture design, token cost management, workforce enablement, business model disruption, and building adaptable systems for continuous change.

Key Questions Answered

• •Agentic Redesign vs. Bolt-On Strategy: Enterprises that layer AI onto existing processes underperform those that redesign workflows from scratch. The shift from "I do my work" to "I manage agents that do my work" requires structural changes, not incremental additions. Organizations treating this as a transformation problem rather than a technology problem are ahead of those still optimizing legacy processes.
• •Token Budget Architecture: AI spend no longer behaves like software licensing — it functions more like labor costs. Enterprises are burning through annual AI budgets in months, with Uber cited as a prominent example. Organizations need dedicated observability systems to track token consumption by individual, team, function, and project before they can allocate resources rationally or demonstrate ROI.
• •Systems Thinking Over Vendor Selection: Choosing the best AI vendor is insufficient. Enterprises need to design multi-model architectures that route different task types to appropriate intelligence tiers based on latency, cost, quality, and compliance. This includes harness design, data access provisioning, context management, and guardrails — not just model selection from a catalog of 11,000-plus options.
• •Workforce Enablement Gap: The upskilling requirement has escalated sharply. When AI meant better prompting, light training sufficed. Now that non-technical employees manage agents with access to critical systems, inadequate training produces real operational risk. Multiple event attendees reported agents accidentally executing on production systems due to missing guardrails, not user error, making structured enablement a risk management priority.
• •Built-In Obsolescence as Design Principle: Any agentic system built today should assume it will require significant redesign within months. Harnesses, interaction patterns, customer expectations, and policy frameworks are all in flux simultaneously. Enterprises that hardcode assumptions into their AI architectures will face costly rebuilds; those that design for ephemerality and modular swappability will adapt faster as the environment shifts.