Enterprise AI has to work with the systems, data, permissions, and processes a company already has. Shannon Bell, OpenText’s executive vice president, chief digital officer, and chief information officer, argues that accumulated IT complexity—not AI capability alone—can keep organizations from turning experiments into useful, scalable deployments. Her argument is an executive viewpoint, not proof that complexity explains every company’s results. The practical response is to simplify where complexity no longer serves a purpose, strengthen data governance, and give agents only as much authority as their demonstrated performance and risk justify.
Why complexity can hold enterprise AI back
An AI agent does not operate in isolation. It may need to retrieve information from several systems, interpret inconsistent records, and follow a workflow shaped by years of integrations and exceptions. Each connection adds dependencies to operate, secure, and maintain. Customized or fragmented systems can also leave an agent facing edge cases that make its overhead greater than the value of the task.
As an Amazon Associate I earn from qualifying purchases.
Bell says technical debt from past integrations is a barrier to scaling AI. Her CIO opinion article reports that 68% of CIOs identified that problem, but it does not provide the underlying survey methodology; treat the figure as a statistic reported by Bell, not as a fully documented standalone benchmark. She also cites McKinsey figures that nearly two-thirds of enterprises worldwide have experimented with agents while fewer than 10% have scaled them to tangible value. The article does not identify the exact McKinsey report or its fieldwork, so the figures should likewise be understood as Bell’s secondary citation.
Other surveys point to related obstacles, though their findings are not directly interchangeable. Tray.ai, an integration-platform vendor, says a survey of more than 1,000 enterprise leaders found 38% citing integration complexity as the biggest barrier to scaling AI agents, 57% naming security concerns as the top barrier to agent success, and 79% expecting data challenges to affect rollouts. Its infographic does not state survey dates or fuller sampling details, and it promotes the company’s product, so these results are useful context rather than neutral industry benchmarks. Read Tray.ai’s infographic.
#1 Best Overall
Complexity is not exclusively a technology problem. In a Harris Poll survey for CapTech, conducted May 8–22, 2026, among 302 director-level-and-above IT decision-makers at US organizations with at least 250 employees already using AI beyond the pilot phase, 56% said stakeholder misalignment and decision-making hindered modernization more than technical barriers; 44% pointed to technical barriers. In that same survey, 86% agreed their organizations could implement AI rapidly but internal decision-making slowed progress, and 90% believed modernization decisions would improve with a shared understanding of tradeoffs. CapTech is a consultancy that sells related services, so its findings describe that defined US respondent group, not all enterprises. See CapTech’s survey announcement.
Simplify the stack without removing necessary complexity
The goal is not a perfectly uniform architecture. Bell describes OpenText as working across data centers, multiple public and private clouds, and sovereign environments because different workloads have different requirements. Security, regulation, data sovereignty, and business fit can make architectural variation necessary.
The more useful distinction is whether a dependency serves a present need or remains because of historical growth and past integration choices. Bell puts it this way: “The important distinction is between complexity that serves a purpose and complexity that has simply accumulated.” The CIO opinion article reports that OpenText began with more than 1,500 applications and later reduced its landscape by more than 300 applications while consolidating over 40 data centers. Those are Bell’s descriptions of her company’s transformation—not industry benchmarks or targets every organization should adopt.
For each proposed AI workflow, review its dependencies before adding another layer of integration:
- List the applications, data stores, services, and handoffs the workflow needs.
- Identify which dependencies are required for security, regulation, sovereignty, or business operations, and which persist mainly because of past choices.
- Consider whether a redundant connection, duplicated data path, or unnecessary process step can be removed or consolidated.
- Check what new operational and security work each connection would create, including how it will be maintained and monitored.
Make data and permissions ready for the workflow
An agent can only use the information it can access and interpret. Data quality affects whether its output has enough context to be useful; data location, governance, and permissions affect whether retrieval or processing is appropriate. These are workflow design questions as much as platform questions.
Before deployment, document the information the agent needs, where it resides, who governs it, what may be exposed to the AI system, and where sensitive information will be processed. Define role-appropriate access rather than granting broad permissions for convenience. Make the agent’s activity and the data it uses visible to the people responsible for oversight, then monitor outcomes after launch.
Independent reader supportYour contribution helps us test, update, and keep practical guides available for everyone.Start with a bounded task, then earn more autonomy
Choose an initial use case with understood inputs and outputs. Map the systems it touches, the information it requires, the permissions it needs, and the result people expect. Where errors could have material consequences, begin with analysis and recommendations rather than allowing the agent to take consequential action.
- Define the task. Specify the expected input, output, and limits of the agent’s responsibility.
- Establish a review process. Have a person check recommendations or actions while the workflow is being evaluated.
- Record performance. Track whether outputs were correct and what differed when they were not; use those findings to identify data, integration, or process failures.
- Expand authority only when justified. Increase what the agent can do as evaluation shows reliable performance, with oversight calibrated to the consequences of failure.
- Keep controls active after launch. Preserve visibility into behavior and data, monitor outcomes, and maintain a way to disable the agent.
Bell describes OpenText’s network and security operations team using a resolution agent to analyze incidents and recommend a resolution, with a human making the final decision at the time described. This is an example from her account, not an independently tested case study. Her general principle is that “autonomy should be earned through evidence.”
Best Value
Make ownership and decision-making part of the design
Technical readiness does not settle who owns a workflow, which tradeoffs are acceptable, or what counts as success. Those decisions shape the agent’s scope, permissions, evaluation criteria, and escalation path. CapTech CTO Brian Bischoff argues that organizations should establish “clear ownership, governance, and measures of success early” to move initiatives toward business outcomes. His comments accompany CapTech’s survey and should be read in that consultancy context.
Before expanding a deployment, make sure the relevant technical, security, data, legal, and business stakeholders agree on the workflow’s purpose, risk, access boundaries, review responsibilities, and success measures. If decision-makers cannot agree on those points, adding integrations or autonomy will not resolve the underlying organizational bottleneck.
What to compare when choosing an enterprise approach
The sources cited here do not provide a neutral, product-by-product performance comparison. Evaluate approaches against the needs of the specific workflow rather than assuming a single platform or architecture will fit every environment.
The Tool Desk
Outbyte Driver Updater FREEScan for outdated or missing drivers - takes under a minuteDriver Scan →Outbyte PC Repair FREEClear out junk files and repair common Windows errorsFree Scan →Quick Recap
- Integration burden across existing systems and the effort needed to maintain connections.
- Data quality, governance, location, and control over sensitive information.
- Security and access controls that match the agent’s defined responsibilities.
- Support for hybrid, private, or sovereign deployment requirements where applicable.
- Clarity of ownership, human review, escalation, and rollback or disablement options.
- Visibility into agent activity and data, plus a way to evaluate outcomes over time.
- The organization’s ability to agree on tradeoffs, decision rights, and measures of success.
Product prices and availability are accurate as of the date/time indicated and are subject to change. Any price and availability information displayed on Amazon at the time of purchase will apply.
