Fixing Failing AI Pilots

"Agentic scaffolding is one of the tools we’ve built to really help companies move from pilots to actually scaling their AI efforts across the enterprise,” Rima Safari, a partner in PwC’s US data, analytics and AI practice, told me in a video interview. “At this point, every major AI transformation that PwC takes on with a client, they’re using scaffolding for it as a tool to drive that operating model change.”

How Agentic Scaffolding Works

With agentic scaffolding, business leaders can input strategic objectives and process requirements into an agentic scaffolding tool. The scaffold is an application layer that can visually orchestrate AI workflows while generating the underlying code needed to execute and manage those agents.

“Agentic scaffolding allows you to think about, what are those bottlenecks that don’t allow you to scale, and how do you surpass them? So one bottleneck oftentimes is the change management required…in terms of the evolution of roles, so agentic scaffolding helps with that,” Safari said.

In short, scaffolding provides an early look at what a new process would look like automated with AI. For instance, a team can create a visual simulation of how a process would work, including simulated steps, required tasks, handoffs, exceptions, validators, evidence requirements, data flows and documentation.

The scaffold can also be stress tested to identify ways to improve performance before the initiative moves to production. This approach helps to fine-tune policies and controls for each agent.

But how effective is this approach? PwC claims that one Fortune 500 insurer it works with has used agentic scaffolding to help create and validate agentic workflows for intake, quoting, underwriting, enrolment and billing, cutting the quote-to-bill time by 50% to 80% depending on case complexity.

Reimagining Change Management

As more companies look to leverage AI agents alongside human employees, enterprises require a more agile approach to change management to keep up.

“Previous change management playbooks assumed time. Long planning cycles, sequential rollouts, structured resistance management. That model is now broken. Agentic AI does not give you a stable target to manage change toward because the system keeps evolving after deployment,” Shashi Bellamkonda, principal research director at Info-Tech Research Group, told me via email.

“What I am seeing is that the enterprises making real progress have stopped treating change management as a phase that follows implementation. It has to be the foundation the implementation is built on. The compressed timelines are not a temporary condition. They are the new operating environment,” Bellamkonda said.

Bellamkonda argues that traditional change management was designed for a bounded transformation with a defined end state, whereas agentic AI has no end state. Agentic systems learn, the workflows evolve and now the workforce also has to evolve with them continuously.

Building AI Initiatives

Rebuilding workflows to accommodate AI agents is a tall task. Chai Atreya, CPO and CTO of the autonomous marketing platform ActiveCampaign and an ex member of the Amazon Alexa team, also shared some fundamentals for rolling out AI pilots with success.

Companies that want to implement pilots effectively need reusable capabilities, context layers, governance models, agent infrastructure, evaluation loops and teams that can continuously apply AI to new business problems. Beyond that, there needs to be a company-wide change to support success.

“The biggest requirement is organizational. AI pilots cannot sit off to the side as innovation theatre. Product, engineering, design, data, legal, security, and business teams need to work together around a clear outcome," Atreya said.

PwC’s deployment of scaffolding, alongside Bellamkonda and Atreya’s insights into modern workflows, points toward a shift in the tech industry in terms of how AI pilots are built and brought to production.