Building Enterprise-Grade AI: Insights from PwC’s AI Factory Leader

Enterprise AI is entering a more demanding phase. The era of scattered pilots and isolated use cases is giving way to a harder question: how do organizations make AI work reliably, responsibly, and at scale? In this interview, Jacob Wilson, PwC’s AI Factory Leader, unpacks what that shift really means for technology and business leaders.

Wilson explains why successful organizations are narrowing their focus to high-impact workflows, why observability and orchestration have become critical to trusting autonomous agents, and how AI is reshaping leadership, talent, and sustainability strategy.

Rather than discussing AI as a future promise, this interview explores how executives can turn agentic systems into measurable business value – safely, responsibly, and at scale.

Aligned with PwC’s 2026 AI Business Predictions

PwC predicts a shift from bottom-up experimentation to top-down AI strategy. Would you like to elaborate more on this?

Wilson: What we’re seeing is a move away from one-off pilots happening in different corners of the business and toward a more intentional, leadership-driven approach. Instead of every team running its own experiments, executives are setting clearer priorities around where AI should create value and building centralized teams to help deliver it.

So it becomes less about testing everything everywhere, and more about focusing resources on the top business priorities where AI can truly move the needle.

Likewise, PwC predicts AI will both accelerate sustainability and create sustainability risks. Where do you see GenAI having the biggest impact on energy efficiency and supply chain optimization?

Wilson: While AI’s energy footprint is growing, it can also be one of the most powerful tools companies have to use energy more intelligently, especially when sustainability data is built into AI-driven decisions. That means feeding sustainability data into the same AI systems that already drive forecasting, operations, and supply chain decisions. On the energy side, AI is already helping companies run more efficiently by predicting grid conditions, deciding when energy-heavy work should run, and spotting maintenance issues early – which cuts costs, emissions, and downtime at the same time.

In supply chains, AI helps turn scattered data into something useful. It improves demand forecasts, reduces inventory waste, optimizes routes, and makes operations more resilient as climate and market conditions change.

With the right governance in place, companies can manage AI’s energy needs while unlocking much bigger efficiency and resilience gains. When done right, AI and sustainability reinforce each other instead of competing for attention.

Agentic AI: From pilot to proof

Today, agents are moving into core finance, HR, and IT operations. What differentiates an agent that is “pilot-ready” from one that is truly “production-ready”?

Wilson: A pilot-ready agent can handle a task when the conditions are controlled (i.e. happy path). A production-ready agent can do it dynamically and reliably in the real world – with guardrails in place, the right system integrations, robust security, clear human oversight, agent observability, and performance that stays steady even as things change. It’s really the jump from “this works in theory” to “this works every day, at scale.”

Enterprises are struggling with trust and predictability in autonomous systems. How can observability close the trust gap and accelerate agent deployment into mission-critical workflows?

Wilson: Observability gives teams a single pane of glass into what an agent is doing, why it’s making certain decisions, the quality of outputs based upon human feedback, and how its behavior changes over time. When you can see that in real time and understand exactly what’s working well vs not, it takes the guesswork out of adoption and ongoing improvements required to reach the full business value.

It also creates a feedback loop: teams can spot issues early, adjust guardrails, improve context, and optimize instructions to continuously improve performance. That ability to meaningfully improve performance based upon data builds confidence, reduces risk, and makes it much easier for organizations to move agents from small tests into mission-critical workflows where reliability really matters.

Is ROI from agents finally measurable? What early indicators tell you that an agent is creating real operational lift vs. just shifting work around?

Wilson: Yes, with high quality agent observability, organizations are finally able to measure business value in meaningful ways. The strongest signals show up at the workflow level, not in isolated task metrics. If you’re seeing faster cycle times, fewer exceptions, smoother handoffs, and stable or even reduced operating costs, that’s a good sign the agent is creating real lift.

You can also look for improvements in throughput or capacity, which suggests agentic transformation is truly amplifying the work rather than just moving tasks from one team to another. If those broader metrics aren’t improving, the agent(s) may be active, but it’s not yet delivering operational value and that’s where you can dig into the agent observability data to determine the right course of action to get back on track to business value realization.

Rise of the AI generalist workforce

What skills define this new AI-driven workforce, and how should leaders/enterprises rethink talent models to support it?

Wilson: The AI-driven workforce will look a lot less like people completing tasks and more like people directing, shaping, and improving AI-powered workflows. Instead of focusing solely on execution, employees will also need stronger skills in problem framing, data fluency (including agent observability data), judgment, and managing hybrid human-AI teams. As AI agents take on more of the day-to-day operations, human roles shift toward oversight, continuous improvement, and ongoing translation of business goals into clear agent instructions, context, and guardrails.

For leaders, this means rethinking talent models around capabilities rather than rigid job titles. The organizations that move fastest will prioritize cross-functional fluency, continuous learning, and roles that can evolve as their AI systems mature.

What does leadership look like in a world where agents, not humans, handle the bulk of execution?

Wilson: In a world where agents handle a significant amount of the execution, leadership becomes far more about direction-setting, governance, and orchestration than task management. Leaders will shift from supervising people’s work to designing the conditions within which agents operate, including objectives, guardrails, data access, and accountability structures.

Leaders will also need to cultivate trust, transparency, and responsible AI practices to make sure agentic decisions align with organizational values, while minimizing risk.

Should leaders actively reskill toward agent orchestration roles, or will the market naturally adjust?

Wilson: Leaders should actively reskill toward agent orchestration roles, because waiting for the market to adjust will leave organizations with capability gaps that slow adoption and reduce ROI. Agentic systems change not just how work is done, but who designs and governs that work, making orchestration skills central to future leadership.

Market forces alone won’t develop these abilities fast enough or in alignment with a company’s strategic priorities. By investing in targeted reskilling now, leaders can help their organizations fully leverage AI while maintaining control, accountability, and competitive pace.

Strategic Outlook for 2026–2028

If you could offer one prediction that CTO or business leaders are overlooking about AI’s business impact, what would it be?

Wilson: A key prediction many CTOs and business leaders are overlooking is that observability will become one of the most critical – and most underestimated – pillars of AI-driven operations. As autonomous agents take on complex, multi-step workflows, organizations will struggle not with what agents can do, but with understanding what they are doing, why they made certain decisions, how to intervene when outcomes drift, and how to rapidly evolve the agents as the business continues to evolve.

Traditional monitoring tools won’t be sufficient; enterprises will need new frameworks for real-time visibility and traceability. Those that fail to build robust observability will face operational blind spots, compliance risks, and an inability to scale AI with safety and security.

What’s one belief about AI-driven transformation that enterprises will regret holding onto in 2026?

Wilson: In 2026, enterprises will regret believing that AI transformation was primarily about adopting new technology rather than reinventing how their organizations operate. Treating AI as a plug-in to existing processes will leave companies with incremental gains instead of the broader business transformation improvements agentic systems can unlock.

The real value comes from redesigning workflows, roles, and decision rights so humans and AI agents can collaborate seamlessly. Organizations that fail to rethink their operating models will struggle to translate AI investments into meaningful business outcomes.

Closing thoughts:

The message from this conversation is unambiguous: enterprise AI is no longer a technology experiment – it is an operating mandate. Leaders who continue to treat AI as a collection of pilots will fall behind those building observability, orchestration, and governance into the core of their organizations.

Likewise, agentic systems are already reshaping how work is executed, how value is measured, and how leaders lead. The real risk now isn’t moving too fast, but moving forward without the discipline to scale AI responsibly.

Remember, there is still time for leaders to guide how AI transforms their organizations. But once AI systems are widely embedded in operations, it will be far more difficult to course-correct.