Expert Insights

Q1: You’ve often spoken about helping organizations move from AI ambition to measurable business outcomes. In your experience, what’s the biggest misconception executives have about becoming an “Agent-First” enterprise?

The biggest misconception is that Agent-First is a technology decision — pick the right model, wire up some agents, and you’re done. It isn’t.

In practice, becoming Agent-First is an organizational design decision disguised as a technology one. Executives often assume agents can slot into existing processes the way a new application does. They don’t.

An agent that can actually act, not just recommend, exposes every place where your process, your data, and your decision rights weren’t clearly defined in the first place. I’ve seen this repeatedly in deployment work: the technical build is rarely the bottleneck; the redesign of who owns which decisions, and what “good” looks like when an agent is making them—is the real bottleneck. Executives who treat this as a rollout rather than a redesign end up with expensive automation that nobody trusts enough to delegate.

To be honest, this is not an Agent-First Enterprise problem; this has been a mindset for a very long time for any digital transformation but is more relevant in the Agentic era.

Q2: Many organizations are still approaching AI as a productivity tool rather than an operating model transformation. Based on the work you’re doing with enterprise customers, what fundamentally changes when businesses begin designing around AI agents instead of simply deploying AI features?

Three things change fundamentally.

First, the unit of work shifts from a task to an outcome — you stop asking “what feature does this AI add to my CRM” and start asking “what job can this agent own end-to-end.”

Second, your data and process architecture has to become agent-first. Agents need clean, well-governed context to act reliably, which forces a level of data hygiene most enterprises have been deferring for years.

Third, and most underrated, your operating model needs feedback loops built in from day one; an agent that can’t learn from its own outcomes is just a more expensive automation script. Bolting AI features onto existing workflows optimizes the old model.

Designing around agents means redesigning the workflow itself around what autonomous execution makes newly possible. I have shared this in greater detail during the webinar that we did with Grazitti, which may be worth referencing.

Q3: Having advised enterprise CXOs for over 17 years, you’ve seen multiple waves of digital transformation. How does the current AI revolution compare with previous shifts such as cloud, CRM, or customer experience transformation? What’s genuinely different this time?

Every prior wave whether cloud, CRM, CX, etc. was fundamentally about giving humans better tools: better access, better visibility, better records. The human stayed the unit of execution throughout, and they were the USERS.

This wave is different because, for the first time, the software itself can be the unit of execution. That’s not an incremental capability upgrade, it’s a shift in who (humans) or what (AI) is doing the work.

What’s genuinely different this time is the speed of the diffusion curve combined with the depth of organizational change required. Cloud adoption gave you a decade to sort out governance because the risk profile was contained. Agentic AI compresses that timeline dramatically, and the risk profile, i.e., agents acting autonomously in customer-facing and financial processes — is not something you can retrofit governance onto after the fact.

And this is the reason that regulated industries are seeing a slower pace of adoption of Agentic capabilities. They are conservative, taking risks at that scale.

Q4: Enterprise leaders are under immense pressure to demonstrate ROI from AI investments. In your conversations with CXOs, what are the most meaningful metrics organizations should be tracking beyond productivity gains?

I’d push CXOs toward three categories beyond productivity.

First, decision quality and consistency — is the agent making the same quality of judgment call on the 1,000th case as the 1st, and how does that compare to human variance?

Second, cycle time on outcomes that matter to the customer, not just internal throughput i.e., resolution time, time-to-value, not just tickets handled per hour. This is the reason that most AI vendors have started outcome-based pricing as well.

Third, and the one most organizations skip, is trust velocity — how quickly are you able to expand an agent’s scope of autonomy because it’s earned it, versus how much human-in-the-loop overhead persists indefinitely. This is also tied to effective change management.

If that third metric isn’t moving, you haven’t actually transformed the operating model, you’ve just added a very sophisticated assistant.

Q5: As autonomous AI becomes more embedded in customer-facing operations, governance and trust become just as important as innovation. How should organizations balance speed with responsible AI adoption, particularly in highly regulated industries?

The organizations getting this right treat governance as a design input, not a compliance checkpoint at the end. Practically, that means defining the agent’s scope of authority explicitly before deployment — what it can decide alone, what needs escalation, and what’s off-limits entirely — rather than discovering those boundaries after an incident. This is exactly what I meant when I referred to agents being an Agentic Layer on top of a well-structured business process.

In regulated industries particularly, I’d advise building graduated autonomy: start agents in advisory or human-approved mode, instrument everything, and expand their authority as they prove reliable in your specific context, not based on a vendor’s benchmark. Speed and responsibility aren’t actually in tension if you sequence it this way — the tension only appears when organizations try to skip the earning-trust phase and go straight to full autonomy. A good example is putting agents to work for Internal use cases rather than external customer-facing ones, so trust is built, and eventually the exposure could be extended.

Q6: If you were advising a CEO planning their AI strategy for the next 24 months, what three priorities would you recommend to ensure they’re building long-term competitive advantage rather than simply keeping pace with the market?

First, invest in data and process readiness before you invest in agent capability. The enterprises that will win this cycle are the ones whose data is clean and whose processes are well-defined enough for an agent to act on reliably; that’s unglamorous work, but it’s the actual moat.

Second, build organizational muscle for graduated autonomy: create the governance, monitoring, and escalation structures now so you can expand agent scope quickly and safely as trust is earned, rather than relearning this under pressure later.

Third, redesign around a small number of high-value, end-to-end outcomes rather than spreading thin across many shallow use cases — depth on a few agent-owned processes builds real competitive advantage and organizational learning; breadth without depth just produces a lot of pilots that never scale.

Fourth, I’d recommend trusting one Agentic partner, instead of partnering with dozens of them. Complicating the tech landscape has been a norm in industry; with agents, it will become even more difficult to manage. Simplify AI rather than complicating it.