agent AI Enterprise adoption to date is reminiscent of the “rogue IT” era of SaaS, with employees accessing or requesting rogue apps, sanctioned tools beginning to be deployed, and business and IT leaders scrambling to articulate a cohesive strategy for their organizations.
Granted, there are plenty of licensed pilots, but many organizations report frustration with scaling up these agent AI applications. One reason for this is that apps are designed to handle narrow tasks. Security, governance, and data infrastructure are typically works in progress; multi-agent orchestration –Enabling agents to collaborate on complex tasks and goals remains an open question for in-house developers and software vendors.
Deloitte, a Big 4 accounting firm and IT consultancy, has introduced advisory services and online tools designed to help clients draft and execute agent AI roadmaps that drive tangible business outcomes.
In this Q&A, China Widener, vice chair of Deloitte and a leader in the U.S. technology, media, and communications industry, explains how to escape “pilot purgatory” and develop a strategy for scalable agent AI. She also shared advice on how to cope Concerns about job loss due to AI By redesigning work in ways that benefit both individuals and organizations.
Mr. Widener will celebrate his 20th anniversary at Deloitte in May. She previously held executive positions in Ohio state government and served as an assistant county prosecutor.
Editor’s note: This interview has been edited for clarity and brevity.
What’s the one thing that’s stopping companies from moving beyond pilots and getting enterprise-wide ROI from agent AI?
china widener
China Widener: There is more than one thing, but at the heart of it all is the idea of clarity of vision. There is now, there is next, and there is future, and each has created value. The question is, where do you start?
This is not a technology issue. The technology is working. The question is, in what circumstances and for what purposes does it relate to the vision of the organization?
You can unlock value in the form of cost containment or mitigation, and create value through growth, new products, tools, services, or experience changes. Where you start your journey has to do with understanding what your goals are. What you identify becomes a series of questions.
More than half of the companies in our study are in the early stages or don’t have a strategy. This is difficult. Because there is no roadmap for where to go, no way to judge progress, let alone steps to take.
Deloitte’s 2026 State of AI in the Enterprise Study Insights gained We gathered input from more than 3,200 business and IT leaders directly involved in AI initiatives. Only one-third said they used AI to truly transform their business. The report suggests that this is because most companies are still focused on training employees to become fluent in AI and are not redesigning jobs enough. Why is the redesign effort so important?
Widener: There is value that comes from doing tasks faster. The idea of upskilling people and making them fluent with AI is to get them hands-on, working with AI, and leveraging it as they perform their daily tasks. Nothing has changed about their work. What matters is whether you can do it faster, more efficiently, or more accurately.
There is a step in this process that asks how AI can be used to enhance the cognitive skill set of employees. Research shows that elevating employees to create a level of cognitive equivalence is the most effective. But if you just apply it to the work they’re already doing, it makes their work more efficient, regardless of whether they need to do it anymore.
Redesigning your job allows you to change how you spend your time and what you spend it on. Currently, a single function may have 17 tasks associated with it. If we redesign the work itself, there could be only 10 tasks in the future, as some tasks can be done through some kind of automation. This frees up a lot of time and allows the same person to take on other or additional tasks. They can focus on 10 tasks that cannot or should not be automated and actually require human judgment.
Many workers are worried about losing their jobs or changing jobs so often that they won’t be able to keep up. How should organizations handle training and upskilling so that employees can trust that they will be supported to transition into new roles?
Widener: That’s the case with most things. Travel has an evolutionary path. You don’t just wake up on Tuesday and suddenly accept and adapt to changes in technology. Start giving your employees access to the work they already know how to do, so they can expand on it and do it faster. For example, research work. The end result is a more comprehensive and faster investigation. The product you produce is still your product, but having these tools available will produce a higher quality and more robust product.
It then advances the individual towards cognitive enhancement and thought process support and provides access to tools to do so. This phased approach allows employees to deploy agents that perform some of the most reproducible functions. However, when judgment and quality are required, humans assume control of the agent.
This is not a technology issue. The technology is working. The question is under what circumstances and for what purpose.
china widenerDeloitte Vice Chairman
You have to take your employees through it step by step. It’s not just about training. Some organizations believe that they simply need to improve the skills of their employees. Skill development is important, but it’s not everything.
There’s also the change management aspect. Our research shows that some things are undervalued by organizational leaders, such as role redesign. AI fluency is important because it enables adoption and execution. This is difficult to measure, such as how much time is saved or how much revenue is generated. They are specific and easily calculated.
But the intangible is just as important, and that’s the change management capability, and how you think about your organization’s operating model when you have this technical capability available to you.
What use cases for agent AI are you seeing in your industry?
Widener: We were just having a conversation about agent AI and its ability to be creative and create efficiency, productivity, and effectiveness. This is a valid question whether you’re a growing organization, concerned about disintermediation, or looking for your next corporate strategy to take full advantage of agentization and the opportunities it creates.
Agentization has not affected all industries in exactly the same way. Some industries appear to be becoming more internally focused, emphasizing efficiency, and changing back-office functions. For other industries, such as the entertainment industry, it impacts the creativity of storytelling that is at the core of business. This industry and hardware have different implications for supply chain management changes.
What are the most frustrating challenges for clients when taking agent AI beyond the pilot phase and across the enterprise?
Widener: that teeth Biggest pain point. Most companies are doing some form of experimentation or piloting. Some consider this a pilot purgatory, as less than 20% of pilots successfully scale up. The value proposition of agentization is unquestionable. what teeth The question is, how can we capture and benefit from that value faster and at scale?
The challenges that arise start with the pilot itself and whether it was built to solve an enterprise problem or a specific productivity problem individually for a few or one specific use case. But should that use case be scaled? And is it scalable?
Technology and data selections are made in a pilot and cannot be expanded later. You can’t fit pilots into narrow use cases. We need to build on a broader infrastructure, understand the larger data issues that can arise, and recognize that governance is important. Data quality issues in pilots may be manageable because the pilot is small, but they become unmanageable when scaled. Scale questions should be determined as part of the pilot. It’s not about building a pilot and thinking about scale questions later.
What should companies put in place to approach agent AI from a scalability perspective and perform at that level?
Widener: Pilots tend to grow in a fairly organic way. An organization buys access to a particular LLM or AI tool, unleashes it throughout the body, and forces the people doing the work to use the tool to improve something. Then harvest the best of those ideas and consider whether they need to be scaled.
Instead, some organizations take a broader top-down or enterprise approach and say, “These are the toolsets we want to leverage, and here’s where we want to focus.”
The key to all of this is to have a disciplined approach to agentizing every aspect of your business, no matter which end of the spectrum you start from. It doesn’t matter who the stakeholders are. True consistency can be achieved if there is a consistent and disciplined approach in which ideas are evaluated and proposed value is calculated, with an understanding of the technical implications.
No organization becomes an agent overnight. Every organization begins in some way. Perhaps by competency or by business unit. But without a consistent and disciplined evaluation process, everyone will be starting from a different place, calculating different benefits, and executing on different visions.
A disciplined approach requires several components. One is to be clear about how you choose your goals. Do you mean cost? growth? experience? There are things that are amenable to agentization that can yield intermediate values, so we need to be clear. Some offer more value, but the fit will be a greater burden.
Getting to the right place for your agential bets also requires standards and protocols for evaluating them, based first and foremost on what you value. Then move your organization through the same assessments periodically to arrive at your focus. It gives you consistency and clarity so you understand what you’re spending your money on, what value you can expect in return, and can calculate and monitor that value.
Agent AI is a trend, but it’s happening in a broader context: Generation AI There are also large language models, some of which are old. What are the fundamental elements of agent AI that business and IT leaders should focus on first?
Widener: Autonomous agents operate independently, so the more autonomy an agent has, the more important governance and data quality become. I’m not going to ask for permission to run it. And the more autonomy you have, the more security you need. These are important at first, but become even more important as autonomy increases.
Deloitte has introduced a tool called Enterprise AI Navigator. What is it and how do people get it?
Widener: It was created because there was no consistent tool available that would allow organizations to have a disciplined approach that they can apply now and into the future.
This is not something we sell to our clients as a product, but rather a tool that they use to help advance their agency journey and reach both their roadmap and the outcomes they have identified. This will help you feel more confident in your choices.
Future workflows (changes to the work itself as described earlier) are suggested so you can see them before your first agent is coded. You can understand what your future will look like and begin to understand how your operating model will change and the changes you will need to make to your workforce in terms of training, upskilling, or job shifts. You know all of that comes in.
Now they are things that are discovered along the way, and that is what contributes to pilot purgatory. You need to know them when you start your agent efforts, not when you have already started your agent efforts.
David Essex is an industry editor who creates in-depth content on enterprise applications, emerging technologies, and market trends for several Informa TechTarget websites.
china widener
