AI is now a regular feature of earnings calls, and several large vendors are breaking out meaningful AI‑related revenue and demand signals. Yet the gap between executive optimism and the actual AI readiness of enterprises is widening. Accenture’s latest Pulse of Change report shows that 86% of C‑suite leaders plan to increase AI investment in 2026 and 78% now see it as a driver of revenue growth, but only 32% report sustained enterprise‑wide AI impact. Only about a fifth say they are rebuilding processes for AI, while fewer than one in 10 are redesigning roles around it.
That gap is the core issue: Productivity gains at the edge are appearing faster than the structural changes needed in core business processes. In 2026, the real question is not whether to scale AI, but how to build the infrastructure and operating practices to handle the related data, workflows, governance and human decision‑making that turn isolated AI pilots into a durable advantage. However, Accenture’s own large‑scale deployments with Microsoft, ServiceNow and others, along with similar efforts from firms like IBM Consulting, point to an emerging blueprint for an AI operating model that is already working in the real world.
(Disclosure: Moor Insights & Strategy provides paid research, consulting and advisory services to companies in the technology industry, including some of the companies discussed in this article. Accenture is a former client.)
A Practical Blueprint For An AI Operating Model
Leaders I speak with are past the stage of asking whether copilots, agents or forward-deployed engineers will be useful. Instead, they are wrestling with a more basic problem: Their organizations were not built for a world where software can take actions, not just reveal or provide information. The executives surveyed for the Pulse of Change report describe that tension directly. Great numbers of them are committing more budget to AI and expect it to drive growth, but far fewer say their business is set up to turn that investment into results. Employees are already testing new tools, yet many still lack clarity on how work, roles and accountability will change; only about half feel their training has prepared them for AI-related changes at work. Across those signals, the message is the same: Outcomes for AI depend heavily on how a company designs and runs its operating model around the technology.
The Pulse of Change report and similar research, along with briefings I have received from firms such as IBM, suggest that AI value now depends as much on the surrounding structure as it does on individual use cases. Across the organizations I talk to, the ones that are moving beyond pilots tend to make a common set of choices about data, workflows, governance and talent. Those choices define the operating model for AI, even if it is not always labeled that way. And they increasingly align with four key principles.
1. Start With Real Workflows, Not Abstract Use Cases
Even in measurable areas such as sales and customer service, many AI programs are framed as broad initiatives rather than tied to specific workflows, which makes it harder to prove impact or scale what works. The efforts that move beyond pilots tend to focus on a few specific end-to-end workflows and stay close to how work actually gets done. Examples include choosing a process such as incident management, claims handling or order-to-cash, then documenting how it runs today — being precise about where the data sits, where decisions are made and where handoffs regularly fail. Once that foundation exists, it becomes much easier to decide where copilots or agents can safely take on parts of the work, versus where humans need to stay in charge of judgment calls and exceptions.
That emphasis on process showed up consistently at ServiceNow’s recent Knowledge 2026 conference. Tom Bruss, managing director at Accenture, and Jay Snyder, senior vice president of partners and alliances at observability vendor Dynatrace, walked me through how Dynatrace’s Davis AI for issue detection and SmartScape for mapping dependencies are being wired into ServiceNow workflows to make handling IT problems a proactive operational process. Snyder described the goal as an efficient closed loop that can detect, diagnose, remediate, validate and close IT issues, freeing people “to focus on things they haven’t traditionally had time to do.” Bruss put it simply: “If you’re just applying AI to an inefficient process, you’re automating inefficiency.”
2. Build An AI Spine Across Data, Platforms And Governance
The Pulse of Change report shows that most leaders expect AI to drive growth, but many enterprises still suffer from fragmented data, overlapping platforms and inconsistent controls. Accenture describes the answer as an intelligent superhighway: a cloud-ready architecture, coherent data strategy and integrated workflows that can support machine learning, generative AI and agents on a common foundation. In practice, this spine shows up as a small number of shared data platforms, orchestration layers and policy engines that every AI initiative has to plug into, so the payoff is consistency, scale and clearer control.
One example of this pattern is the expanded partnership between Lenovo and ServiceNow. Lenovo is combining its Digital Workplace Solutions, xIQ telemetry and managed services with the ServiceNow AI Platform, Workflow Data Fabric and AI Control Tower to create an integrated backbone for endpoint‑driven workflows. The intent is to take fragmented device operations and turn them into connected AI‑driven workflows with consistent governance and visibility.
3. Treat Governance As Part Of The Architecture
As organizations move from AI that makes suggestions to AI that acts, governance must become part of the product and architecture. For example, Accenture’s work on autonomous IT with ServiceNow, and the firm’s broader AI guidance, adds scoped permissions, auditability and clear decision rights directly into the design of workflows and agents, rather than leaving them to a separate compliance stream. In Knowledge 2026 conversations, Accenture CIO Tony Leraris described this to me as “a stepped path to a zero-touch service desk for routine issues, but with every agent treated as a first-class identity, governed by policy and fully observable.” Mean time to resolution, case deflection and service quality, he said, are tracked alongside risk and compliance indicators.
ServiceNow’s announcements at Knowledge 2026 extended this idea from the customer-zero environment into the platform itself. The company expanded its AI Control Tower from a dashboard into a command center that can discover, observe, govern, secure and measure AI models and agents across the enterprise, including third-party and custom systems. One new feature that drew a lot of attention was a kill switch that pauses or shuts down agents in real time when they operate outside their scope or exhibit anomalous behavior. ServiceNow leaders referenced a real (non-customer) incident where an AI agent with elevated permissions deleted a production database — including backups — in nine seconds, with no attacker involved and no guardrails to stop it. That points to a broader shift toward agent governance that protects the business at runtime as well as during approval.
IBM’s focus on multi-agent orchestration, policy-aware infrastructure and data sovereignty makes a similar point: Agent behavior, access to systems and operational control all have to be designed and instrumented together if AI is going to run at scale. Secure connections into the right systems, fine-grained scopes, monitoring and logging, and enough process documentation that agents can follow the same rules as people all need to become part of the core design. When governance is treated as a first-class design concern, companies are better positioned to move quickly, keep visibility and control, and capture the payoff from scale.
4. Put Humans In The Lead And Design AI-Assisted Work Around That
The Pulse of Change report is explicit that AI value now depends as much on people as on technology. Leaders are bullish on AI, but worker confidence, clarity and skills are lagging. In the deployments I have seen up close, from Accenture’s internal AI programs to early agent work in sectors like financial services and healthcare, the teams that make progress do not assume that putting humans in the loop will happen naturally. They define which decisions stay with people, how exceptions are handled, what escalation looks like and how success is measured for both humans and AI.
Accenture’s Copilot deployment is one example of that discipline. By the time Microsoft and Accenture announced the rollout of Microsoft 365 Copilot to roughly 743,000 employees, Copilot had already undergone a staged internal journey: early pilots with senior leaders, targeted coaching and clear usage guidance by role, along with internal communities sharing patterns and guardrails. Internal reports indicate that 97% of surveyed employees said Copilot helped them complete routine tasks up to 15 times faster, with active use rates approaching 89% in key cohorts. It’s worth bearing in mind that those numbers sit atop an intentional change-management program, not just feature adoption. None of the other pieces work for long if employees are not clearly in the lead on accountability and judgment, which is what makes the payoff durable.
Putting The Blueprint To Work
To make this concrete, consider a global bank that wants to move beyond AI pilots in customer service and operations. Instead of standing up another broad AI initiative in the contact center, it focuses on one high-value, high-friction workflow: resolving cross-border payment inquiries. That process touches multiple systems and teams, from the contact center and core banking systems to sanctions screening and regional operations, so the payoff can be measured where the work is most painful.
The bank treats this as an end‑to‑end redesign, not a feature add. It maps the current flow and documents where data sits, where decisions are made and where delays or errors are most common. That becomes the basis for a target design that assigns specific steps to agents, copilots and humans. Agents handle tasks such as gathering context from internal systems, checking transaction status across regions and drafting responses. Human experts stay responsible for risk decisions, sanctions issues, complex exceptions and relationship‑sensitive conversations.
Underneath that workflow, the bank invests in the AI spine described earlier. It connects the relevant systems of record to a governed data platform, exposes them through APIs and event streams and routes access through a common identity and policy layer. That allows AI agents and copilots from different vendors to see a consistent view of the customer and the transaction and to operate under the same policies, whether they are embedded in CRM, case management or internal operations tools. Governance is part of the design.
When this kind of redesign starts to work, the conversation inside the bank changes. The focus shifts from where to use AI to which workflows merit this level of architectural and operating‑model investment, and in what sequence.
Where FDEs And Agents Fit
The global bank example raises an obvious question: Who actually carries out this kind of redesign work at scale? Forward-deployed engineers are one of the ways enterprises are starting to answer that. In practice, deploying agents in production is an engineering and process‑design challenge. It involves modernizing systems so agents can securely access the right data, mapping access controls and entitlements, documenting processes in a way agents can follow, designing new human/agent workflows and maintaining evaluation and guardrail frameworks as models and architectures evolve. That work sits squarely inside the blueprint described here, and in many cases, FDEs are the ones accountable for it rather than the line teams they support.
One concrete example of this shift is the FDE program that Accenture and ServiceNow announced at Knowledge 2026. ServiceNow’s AI‑specific FDEs and Accenture’s industry‑focused FDEs are being embedded together inside mutual customers to build agentic workflows directly on the ServiceNow AI Platform — which the customer has already been using — and then scale them from first build to enterprise deployment. This process is adopted as a joint delivery model rather than left entirely to internal teams.
In that sense, FDEs and similar roles are part of the operating model, not an overlay. They help decide which workflows to prioritize, translate domain nuance into process and policy, configure agent behavior within the constraints of the AI spine and design handoffs where humans stay in the lead on judgment and accountability. Large consultancies are playing a similar role. Accenture’s internal work with Microsoft and ServiceNow, like IBM’s AI operating‑model push around its watsonx platform and agent orchestration, all point in the same direction: Enterprises will need people who understand both the domain and the architecture well enough to wire agents into real workloads, govern them over time and keep humans firmly in charge of outcomes.
The Risks Are Real, But So Is The Path Through Them
Every part of this blueprint runs into hard constraints. The organizations that are making progress are not avoiding these risks; they are designing with them in mind.
Architectural and data risk is the starting point. Given how fragmented most core systems and data still are, the practical response is to narrow the aperture. Stabilize a small number of critical domains first, build a governed data layer for those areas, put AI directly into the flow of work, then repeat the pattern.
Governance and control risk rise quickly once agents can take actions in production systems. Accenture’s ServiceNow customer‑zero work is explicit about this: Every agent needs an identity, scoped permissions and an audit trail from day one, because a misconfigured agent can generate thousands of actions before anyone notices. At Knowledge 2026, ServiceNow pushed this idea further with an expanded AI Control Tower that discovers, observes, governs, secures and measures AI agents and applications across the enterprise, including third‑party and custom systems. The kill switch is one visible expression of that approach, but the pattern is broader. Governance is moving into the architecture itself. That means creating a basic control layer early, even if it is imperfect, with clear scopes, central policy enforcement, consistent logging across platforms and a joint team that owns how agent behavior is monitored, adjusted and, when necessary, stopped in production.
People and operating‑model risk are just as significant as technology risk. The Pulse of Change report calls out that the biggest barrier to AI value is now alignment with employees. Executives are optimistic, but many workers do not feel prepared or clear about how their roles will evolve. Accenture’s Copilot rollout and IBM’s client work both point to the same remedy: Change roles, incentives and skills alongside the data and platform, and treat new tools as a change in how work is done, not merely as new software. The core design questions that leaders need to answer involve who owns which decisions, how exceptions are handled and how work is measured in a world where humans and AI share the workload.
None of this is easy, but the pattern is getting clearer. The organizations that are pulling ahead are treating AI as a catalyst to rebuild how work runs, not as an add‑on to legacy structures. In 2026, the real differentiator is whether leaders are willing to redesign a few critical workflows, wire them into an AI spine with built-in governance and give people the confidence and authority to stay in charge as the technology scales.
Moor Insights & Strategy provides or has provided paid services to technology companies, like all tech industry research and analyst firms. These services include research, analysis, advising, consulting, benchmarking, acquisition matchmaking and video and speaking sponsorships. Of the companies mentioned in this article, Moor Insights & Strategy currently has (or has had) a paid business relationship with Accenture, Lenovo, Microsoft, IBM and ServiceNow.
