Global survey of 1,333 business and technology leaders across 25 countries reveals a widening gap between AI-driven productivity gains and enterprise readiness for autonomous operations, with security, risk and governance emerging as the top deployment priorities.
Artificial intelligence is delivering measurable productivity gains across enterprises, but organisations remain cautious about allowing AI systems to make decisions and execute business processes independently. According to a new global study by Tata Consultancy Services (TCS), 84% of executives identify productivity improvement as the leading source of value from their AI investments. Yet, only 10% of enterprises allow AI agents to execute decisions without human approval across multiple functions or business areas.
The findings underline a critical contradiction in enterprise AI adoption: while organisations are increasingly integrating AI into everyday operations, they are not yet ready to hand over significant decision-making authority to autonomous systems. The gap is particularly evident in the limited ability of AI agents to securely coordinate activities across enterprise applications. Just 9% of executives surveyed say their AI agents are capable of securely orchestrating workflows across different enterprise applications.
The study, titled TCS Enterprise AI Study: Intelligence that is Real, surveyed 1,333 senior business and technology leaders across 25 countries and 12 industries. It examines how enterprises are moving from experimentation and AI-assisted productivity towards more integrated, autonomous and intelligence-led operating models.
One of its most significant conclusions is that security, risk and governance have become the leading considerations influencing where and how organisations develop and deploy AI. These concerns now rank ahead of cost, speed and ease of implementation, reflecting the growing complexity of giving AI systems access to sensitive enterprise data, applications and operational workflows.
As AI evolves from a tool that supports employees into a system capable of initiating actions and coordinating processes, the implications of inadequate controls become more serious. An AI assistant recommending a course of action is fundamentally different from an AI agent authorised to execute that decision across multiple business systems. The latter requires clear accountability, access restrictions, auditability and mechanisms for human intervention.
From AI assistance to autonomous enterprise operations
The emerging Human+AI operating model is central to the TCS study. Under this model, AI is expected to move beyond supporting individual employees and gradually assume responsibility for specific tasks, workflows and decisions. However, the transition will not be uniform across all business functions.
Enterprises will need to determine which activities can be safely automated, which require human approval and which decisions must remain under direct human control. The challenge is to strike a balance between operational efficiency and the risks associated with granting AI systems greater authority.
Amit Kapur, Chief AI and Services Transformation Officer at TCS, said enterprises were making significant progress in scaling AI, but the next phase would involve a deeper shift towards autonomous action.
He noted that productivity remains the most visible benefit, with 84% of executives identifying it as the leading source of value from AI investments. The study also found that 10% of respondents had seen an uplift in revenue.
Kapur said concerns around security, governance and resilience were holding organisations back from granting AI greater autonomy, and that responsible deployment supported by appropriate guardrails and enterprise controls would be central to the next stage of adoption.
The findings suggest that enterprises are beginning to distinguish between deploying AI at scale and deploying AI with the authority to act. While productivity improvements can be achieved through employee-facing assistants, autonomous AI requires a much more mature operating environment. Organisations must be able to establish the identity of every AI agent, define its permissions, monitor its actions and ensure that its decisions remain within approved business boundaries.
This is especially important as AI agents increasingly interact with customer information, financial systems, enterprise resource planning platforms and other critical applications.
“Productivity is clearly where enterprises are seeing the greatest value from AI today, but security, governance and resilience will shape the next stage of this journey.” — Amit Kapur, Chief AI and Services Transformation Officer, TCS
Security and governance move to the centre of AI strategy
The prominence of security, risk and governance in the study reflects a broader shift in enterprise priorities. During the initial phase of generative AI adoption, organisations were primarily focused on identifying use cases, improving employee productivity and accelerating implementation. As deployments become more deeply embedded in business operations, the focus is moving towards control, resilience and accountability.
AI agents introduce a different risk profile from conventional software applications. They can interpret information, make decisions and initiate actions based on instructions and contextual data. If their access is not properly restricted, they could expose sensitive information, execute unauthorised transactions or propagate errors across interconnected systems.
Consequently, enterprises need governance frameworks that extend beyond conventional application security. These must address AI identity management, data access, model behaviour, permissions, continuous monitoring and the ability to intervene when an agent operates outside its authorised scope.
The low proportion of organisations allowing AI agents to act independently may therefore reflect a deliberate approach rather than a lack of interest in automation. For many enterprises, the priority is to establish trust and operational safeguards before increasing autonomy.
Model independence and AI sovereignty gain importance
The study also identifies growing concern about dependence on individual AI models and providers. As organisations make long-term investments in AI infrastructure, model independence is becoming an important consideration in technology and procurement decisions.
Ninety percent of respondents say model independence is important when making AI investment decisions. This indicates that enterprises increasingly want the flexibility to select, change or combine AI models according to business requirements, performance, cost and regulatory considerations.
The issue is closely connected to AI sovereignty, particularly where organisations must maintain control over data location, model selection and regulatory compliance. Overall, 34% of respondents identify AI sovereignty-related requirements as a high or very high organisational risk.
The concern is more pronounced in Europe, where 43% of executives regard sovereignty-related requirements as a high or very high risk. Regulatory obligations, data protection requirements and the need to retain control over sensitive information are influencing how organisations evaluate their AI architecture.
However, the study highlights a significant gap between recognising the risks of provider dependence and preparing to address them. Only 7% of global respondents report having organisation-wide continuity plans that would enable them to switch AI providers without disrupting operations.
This lack of preparedness could become a significant challenge as enterprises become increasingly dependent on external models, platforms and infrastructure. Organisations may find themselves exposed to changes in pricing, service availability, model performance or regulatory requirements if they cannot move workloads between providers.
A resilient AI strategy will therefore require more than selecting the right model. It will also involve designing systems that support portability, interoperability and continuity, while ensuring that data, applications and business processes remain accessible during a provider transition.
AI economics face closer scrutiny
Despite the enthusiasm surrounding AI, enterprises are also becoming more conscious of the economics of deployment. The study finds that 39% of organisations are still assessing whether their AI initiatives deliver sufficient value relative to token consumption and computing costs.
The finding is important as enterprises move from limited pilots to broader deployments. Generative AI and agentic systems can involve significant and recurring infrastructure costs, particularly when applications require frequent model interactions, large context windows or complex multi-step workflows.
For business leaders, the question is no longer simply whether AI can perform a task, but whether it can perform that task reliably and economically at scale.
The pressure to demonstrate returns is likely to influence decisions around model selection, workload allocation and the use of open-source alternatives. Enterprises will need to evaluate the total cost of ownership, including infrastructure, integration, security, governance and ongoing monitoring, rather than focusing only on the cost of accessing an AI model.
Open-source AI is gaining attention as part of this evaluation. According to the survey, 46% of respondents report targeted deployment of open-source AI for selected use cases. This suggests that organisations are exploring a more diversified AI environment, selecting models and deployment approaches based on the specific needs of individual workloads.
Open-source models can offer greater flexibility and control, but they also require organisations to take responsibility for deployment, maintenance, security and performance management. Their adoption is therefore likely to remain closely linked to an organisation’s technical capabilities and governance maturity.
Fragmented enterprise knowledge remains a barrier
Another challenge identified by the study is the fragmented nature of organisational knowledge. Forty-four percent of respondents say important business knowledge remains embedded within employees or scattered across fragmented documentation.
This creates a significant obstacle for AI systems that depend on reliable enterprise context. Even sophisticated models cannot consistently produce relevant business outcomes if they cannot access accurate, current and appropriately governed organisational information.
Enterprise knowledge is often distributed across emails, documents, collaboration platforms, databases and departmental applications. Some critical information also exists as undocumented experience and institutional knowledge held by employees.
For AI agents to operate effectively, organisations must address these information silos and establish reliable mechanisms for accessing enterprise knowledge. This involves improving data quality, standardising documentation, managing permissions and ensuring that AI systems retrieve information from trusted sources.
Without these foundations, enterprises risk scaling AI applications that generate inconsistent responses or act on incomplete information. The challenge is not simply to deploy more powerful models, but to make organisational knowledge accessible, structured and trustworthy.
Workforce impact remains uncertain
The study also highlights uncertainty surrounding AI’s impact on employment. The proportion of executives who say it is still too early to determine AI’s effect on jobs has increased from 4% in 2024 to 19% in 2026.
This suggests that the workforce implications of AI are evolving faster than many organisations can assess. While productivity improvements are already visible, the longer-term impact on job roles, skills requirements and organisational structures remains difficult to predict.
As AI takes over repetitive tasks and supports more complex activities, employees are likely to spend more time on decision-making, problem-solving, oversight and activities requiring human judgement. However, the pace and scale of this transition will vary across industries and functions.
For enterprises, workforce planning will need to move beyond conventional automation strategies. Reskilling, AI literacy, human oversight and the redesign of business processes will become important components of successful AI transformation.
Regional priorities reveal different levels of readiness
The TCS study also identifies differences in enterprise AI priorities across regions. North America demonstrates strength in several areas of enterprise AI foundations, with 15% of respondents reporting that AI is treated as a strategic enterprise capability.
Europe places greater emphasis on AI sovereignty, with 43% of respondents identifying sovereignty-related requirements as a high or very high organisational risk. This reflects the region’s focus on regulatory compliance, data control and the need to retain greater oversight of AI infrastructure and models.
The UK and Ireland stand out for their comparatively strong focus on operational resilience, including AI continuity planning and the integration of governance into enterprise operations.
These differences indicate that AI adoption is not following a single global trajectory. Regulatory environments, digital infrastructure, organisational maturity and business priorities are influencing how enterprises approach AI deployment.
Nevertheless, the common requirement across markets is the ability to establish trust in AI systems as they become more deeply integrated into business operations.
The next phase will be measured by trust, not just adoption
The TCS findings point to a new phase in enterprise AI transformation. The initial emphasis on experimentation and productivity is giving way to questions about autonomy, security, economic viability and operational resilience.
The fact that 84% of executives identify productivity as AI’s leading benefit demonstrates that the technology is already delivering business value. However, the finding that only 10% of enterprises allow AI agents to make decisions without human approval shows that organisations remain cautious about transferring authority to autonomous systems.
The gap is understandable. Productivity gains can be realised with relatively limited changes to existing operating models, but autonomous AI requires a deeper transformation of enterprise architecture, governance and accountability.
The next competitive advantage may therefore not belong simply to organisations deploying the largest number of AI tools or models. It may belong to those capable of combining AI-driven productivity with secure access, reliable enterprise knowledge, transparent decision-making and effective human oversight.
As enterprises move towards Human+AI operations, the measure of AI maturity will increasingly be determined by how responsibly organisations allow intelligent systems to act—not merely by how widely they deploy them.