AI systems have evolved substantially over time. Early expert systems relied on predefined rules and required extensive human configuration and supervision. Subsequent advances in machine learning enabled systems to learn patterns from data, improving performance in tasks such as prediction, classification, and optimization [22]. More recently, large language models and agentic AI systems have expanded capabilities beyond narrow tasks, enabling reasoning, multi-step planning, and adaptive execution across diverse domains [23]. McKinsey’s 2025 survey found that 23 percent of organizations are now scaling agentic AI systems, with an additional 39 percent experimenting with AI agents capable of autonomous multi-step task execution [9].

As a result, modern AI systems increasingly combine multiple capabilities—perception, reasoning, decision-making, and execution—within integrated applications. This evolution supports gradual progression from supervised assistance toward conditional autonomy and, in selected domains, fully autonomous operation.

AI automation is best understood as a spectrum of human-AI collaboration rather than a binary distinction [18][21]. At lower levels, AI functions as a decision-support tool, providing insights while humans retain control over decisions and actions. As automation increases, AI systems augment human judgment by recommending actions, filtering options, or managing routine decisions within defined boundaries. Figure 1 illustrates the degree of automation across multiple dimensions of human-AI interaction.

Figure 1. Degree of automation for AI systems.

At higher levels, AI systems operate autonomously under predefined parameters, escalating exceptions to humans only when necessary. Full autonomy represents the upper bound of the continuum, where AI systems independently manage end-to-end processes with minimal human intervention. Research on collaborative intelligence emphasizes that successful automation depends on aligning AI capabilities with human strengths and designing interaction models that support appropriate trust and accountability [21][24].

Many early AI applications focused on analytics and reporting, where benefits such as improved insight generation are relatively easy to demonstrate [3]. While valuable, purely analytical systems offer limited transformational impact unless insights translate into action. Greater strategic potential arises when AI systems move from analysis to execution—automating decisions and operational processes in real time [25].

Achieving higher automation requires reliable system performance, transparency, and mechanisms for monitoring and intervention. Organizations must invest in data quality, continuous learning, and performance tracking to build confidence in automated decision-making. As AI systems demonstrate consistent behavior across diverse scenarios, organizations become more willing to delegate authority and expand the scope of autonomous operation [24].

The degree of integration fundamentally shapes the organizational impact of AI. When AI systems operate in isolation, they may generate accurate predictions or insights within specific functions, but their influence is limited. Insights must be manually interpreted, transferred, and acted upon, increasing delays and inconsistency [27]. Isolated systems often reinforce data silos and constrain AI's ability to support cross-functional optimization.

Higher levels of integration enable AI systems to access comprehensive organizational data, coordinate decisions across multiple functions, and initiate actions directly within operational systems. Integration transforms AI from an analytical add-on into a mechanism for organizational coordination. Prior research on information exchange and interoperability demonstrates that substantial organizational value arises when systems can share data and coordinate processes across functional and organizational boundaries [28]—a principle that extends directly to integrated AI systems.

AI integration typically occurs across three interrelated layers. At the data layer, integration involves connecting AI systems to diverse data sources such as transactional systems, data warehouses, and real-time streams. This requires reliable data pipelines that ensure data quality, consistency, and timeliness [29]. Organizations increasingly adopt architectural approaches such as data fabrics or data meshes to provide unified access to distributed data while preserving governance [30].

At the application layer, integration enables AI systems to interact with business applications through APIs, messaging systems, and event-driven architectures. These mechanisms allow AI outputs to trigger downstream actions and participate in end-to-end workflows. Microservices-based architectures support this form of integration by enabling modular, loosely coupled components to evolve independently while remaining interoperable.

At the user experience layer, integration embeds AI capabilities directly into the tools employees use in their daily work. Rather than requiring users to consult separate analytics platforms, integrated AI appears within operational interfaces, supporting decisions in context. This form of integration increases adoption and effectiveness by aligning AI outputs with existing workflows [24].

Despite its strategic value, integration presents significant challenges. Integration costs are typically front-loaded, while benefits are distributed across stakeholders. Organizations must invest in data standardization, interface development, and system interoperability before realizing visible gains. Legacy technology environments complicate integration efforts, as many organizations operate heterogeneous systems developed over decades, with incompatible data models and limited interoperability. The Deloitte 2024 survey [11] found that resistance to adopting AI solutions often stems from unfamiliarity with technologies or from skill and technical gaps rather than fundamental technology limitations.

Together, these dimensions define a strategic space within which organizations can position existing AI applications and plan future investments. Figure 2 illustrates the resulting automation-integration matrix, which delineates four distinct AI implementation configurations: isolated-supervised, isolated-autonomous, integrated-supervised, and integrated-autonomous. Each quadrant represents a characteristic combination of human involvement, system connectivity, and organizational coordination. The framework does not assume a single optimal configuration; rather, it highlights trade-offs and progression paths that organizations can navigate based on context, strategic objectives, and risk tolerance.

Figure 2. Automation-integration matrix.

1)Low Automation/Low Integration (Isolated-Supervised)

This quadrant represents early-stage AI implementations characterized by limited autonomy and minimal connectivity. AI systems function as decision-support tools within specific departments, with significant human oversight and manual data transfer. These applications are typically used for experimentation, capability building, and low-risk value demonstration. Organizations in this configuration can build familiarity with AI while containing risk, but value creation remains localized and difficult to scale.

2)High Automation/Low Integration (Isolated-Autonomous)

In this quadrant, AI systems operate autonomously within well-defined scopes but remain functionally isolated from broader enterprise systems. These systems can execute decisions at scale and speed, delivering substantial local efficiency gains—such as automated quality inspection or algorithmic trading within a single asset class. However, limited integration constrains their contribution to broader organizational coordination. Value is captured within functional silos rather than across the enterprise.

3)Low Automation/High Integration (Integrated-Supervised)

Integrated-supervised systems combine extensive data connectivity with continued human involvement in decision-making. AI systems provide enterprise-wide visibility and insights while humans retain control over execution. This configuration is common in regulated or high-stakes environments where accountability and contextual judgment are critical. AI enhances human decision-making by synthesizing information across functions, but the pace of action depends on human capacity. Importantly, this configuration should not be read as a transitional stage on the way to full autonomy. In a range of settings it represents a deliberate and stable end state. Where decisions carry irreversible safety consequences (for example, clinical treatment or critical-infrastructure control), where accountability is legally assigned to a human actor (as in credit adjudication, medical diagnosis, or judicial and regulatory contexts), or where outcomes hinge on contextual and ethical judgment that resists codification, retaining human authority over execution is the appropriate target configuration rather than a temporary limitation. In such domains, the value of integration lies in giving human decision-makers comprehensive, real-time intelligence, while the supervisory boundary is preserved by design. The boundary conditions for remaining in this quadrant are therefore the inverse of those that favor progression: high regulatory intensity, high error cost, and high reliance on tacit judgment all weigh toward integrated-supervised as the preferred end state.

4)High Automation/High Integration (Integrated-Autonomous)

This quadrant represents the most advanced AI implementations, where autonomous systems operate across integrated enterprise architectures. AI systems coordinate end-to-end processes, dynamically allocate resources, and adapt to changing conditions with minimal human intervention. While strategically powerful, these configurations require sophisticated governance, robust data infrastructure, and organizational readiness. Organizations achieving this configuration report the highest value capture but represent a small minority of AI deployments [19].

The framework highlights potential progression paths from isolated, supervised implementations toward integrated, autonomous systems. This progression is neither automatic nor uniform across organizations. Research indicates that organizations successfully scaling AI concentrate resources on a limited number of high-impact initiatives rather than pursuing numerous disconnected pilots [10]. Leaders focus on the most promising opportunities and expect more than twice the ROI that other organizations do.

Rather than prescribing a single end state, the framework provides a lens for evaluating trade-offs, sequencing investments, and aligning AI initiatives with organizational capabilities. Many organizations deliberately maintain different configurations across functions, balancing value potential, risk exposure, and implementation complexity. Movement toward higher levels of automation and integration should be understood as contingent on strategic context rather than universally optimal.

To make the framework more testable for researchers and more actionable for managers, we distill the preceding analysis into four propositions. These are offered as conjectures consistent with patterns in the reviewed sources, to be examined empirically rather than as established findings.

Proposition 1 (Integration and enterprise value).Holding automation constant, higher integration is associated with a greater share of value realized at the enterprise level rather than confined to local efficiency gains.

Proposition 2 (Automation without integration).Increasing automation without corresponding integration tends to amplify localized efficiency while leaving enterprise-level value largely unchanged, because gains remain trapped within functional silos.

Proposition 3 (Readiness as a constraint).The feasibility of progression toward integrated-autonomous configurations is bounded by organizational readiness; organizations that advance automation or integration beyond their data, governance, and change-management capacity realize lower value and higher risk than those that sequence the two dimensions in step with readiness.

Proposition 4 (Context-dependent end state).In regulated, safety-critical, or judgment-intensive domains, integrated-supervised configurations represent a preferred and stable end state rather than an intermediate stage, and attempts to maximize automation in these settings reduce rather than increase net value once accountability and error costs are considered.

As a corresponding decision rule, managers can position each AI application on the two dimensions, then progress along the integration dimension before the automation dimension where readiness permits, and treat high regulatory intensity, high error cost, or high reliance on tacit judgment as signals to hold a supervised boundary regardless of available automation capability.

Organizations in this quadrant typically deploy AI as decision-support tools within specific functions, with substantial human oversight and limited system connectivity. These applications build analytical capability and demonstrate value in low-risk settings. Table 1 summarizes representative isolated-supervised implementations across industries.

1)Financial Services:Credit Risk Assessment

JPMorgan Chase employs machine learning models to score credit applications and flag potential risks based on historical data [3]. Loan officers retain full decision authority, reviewing AI recommendations alongside contextual and regulatory considerations. The system operates as a standalone analytical tool that enhances consistency and efficiency while remaining human-controlled. Similar implementations are widespread across banking, where AI-generated risk scores inform but do not replace human judgment on loan approvals.

2)Healthcare:Clinical Decision Support

Healthcare organizations increasingly deploy AI-powered clinical decision support systems that analyze patient data and suggest potential diagnoses or treatment options. These systems alert physicians to potential drug interactions, flag abnormal test results, and recommend evidence-based care protocols [31]. However, clinicians retain full authority over treatment decisions. The AI operates within electronic health record systems but does not autonomously order tests or prescribe medications. This supervised configuration reflects both regulatory requirements and the importance of clinical judgment in patient care.

3)Retail:Demand Forecasting

Retailers deploy predictive analytics tools to forecast demand across product categories and store locations. These systems analyze historical sales data, seasonal patterns, and promotional calendars to project future demand [32]. Merchandising teams review forecasts and make final decisions about inventory levels and replenishment orders. While the AI provides valuable analytical support, human planners retain control over execution and can override algorithmic recommendations based on market knowledge or strategic considerations.

Table 1. Examples: isolated-supervised configuration.

In this configuration, AI systems operate independently within narrowly defined domains, delivering significant local efficiency gains while remaining disconnected from broader enterprise coordination. Table 2 summarizes representative isolated-autonomous implementations across industries.

1)Manufacturing:Automated Quality Inspection

Manufacturers such as Foxconn and BMW deploy computer vision systems that autonomously inspect products and reject defects in real time [32][33]. These systems analyze images of components or finished products, identify defects with precision exceeding human inspectors, and automatically route defective items for rework or disposal. The AI operates at high speed with minimal human intervention during production. However, these systems often remain isolated from upstream planning or downstream logistics systems, limiting their contribution to enterprise-wide optimization.

2)Financial Services:Algorithmic Trading

Quantitative trading firms deploy algorithms that autonomously execute trades based on predefined strategies and real-time market data. These systems can analyze market conditions, identify opportunities, and execute thousands of transactions per second without human approval for individual trades. Risk parameters and trading limits are established in advance, but execution is fully automated within those boundaries. While highly autonomous, many trading systems operate within specific asset classes or strategies and may not be fully integrated with firm-wide risk management platforms.

3)Customer Service:Automated Response Systems

Organizations deploy conversational AI systems that autonomously handle routine customer inquiries without human intervention. These chatbots and virtual assistants can answer frequently asked questions, process simple transactions, and troubleshoot common problems. Gartner [34] estimates that AI-powered customer service solutions will handle a significant share of customer interactions by 2027. While autonomous within their defined scope, many such systems operate as standalone tools rather than integrated components of comprehensive customer relationship management platforms.

Table 2.Examples: isolated-autonomous configuration.

This configuration combines extensive data connectivity with continued human decision authority. AI systems provide enterprise-wide visibility while humans retain control over execution. Table 3 summarizes representative integrated-supervised implementations across industries.

1)Sales and Marketing:AI-Enhanced CRM

Salesforce Einstein exemplifies integrated-supervised AI by connecting customer data across sales, marketing, and service functions to generate predictions and recommendations [35]. The system analyzes customer interactions, purchase history, and behavioral patterns to identify sales opportunities, predict churn risk, and recommend next-best actions. Human sales representatives and marketers determine how and when to act on these insights, preserving contextual judgment while benefiting from enterprise-wide intelligence. This configuration is common where relationship management and human judgment remain critical to outcomes.

2)Healthcare:Integrated Clinical Analytics

Health systems deploy enterprise analytics platforms that integrate data from electronic health records, laboratory systems, imaging archives, and claims databases to provide comprehensive patient views and population health insights. AI algorithms identify patients at risk for adverse events, predict readmissions, and recommend preventive interventions. Care teams review these recommendations and make decisions about patient outreach and care coordination. The integrated data enables insights not possible from isolated systems, while human oversight ensures appropriate clinical judgment [31].

3)Supply Chain:Visibility Platforms

Global enterprises deploy supply chain visibility platforms that integrate data from suppliers, logistics providers, and internal systems to provide real-time tracking and predictive analytics. AI algorithms analyze patterns to predict disruptions, recommend alternative suppliers, and optimize inventory placement. Supply chain managers review recommendations and make decisions about supplier selection, inventory rebalancing, and logistics routing. Integration enables coordination across complex global networks while human judgment addresses exceptions and strategic trade-offs.

Table 3. Examples: integrated-supervised configuration.

This configuration represents the most advanced implementations, where autonomous systems coordinate decision-making and execution across multiple functions with minimal human intervention. Table 4 summarizes representative integrated-autonomous implementations across industries.

1)E-commerce:Intelligent Fulfillment Operations

Amazon’s fulfillment operations exemplify integrated-autonomous AI, coordinating demand forecasting, inventory placement, robotic picking, packing optimization, and logistics routing across a global network [36]. AI systems operate continuously and adaptively, making millions of decisions daily about which products to stock in which warehouses, how to route orders through fulfillment centers, and which delivery options to offer customers. Human involvement focuses on strategic oversight and exception management rather than routine decisions. This configuration enables the speed, scale, and efficiency that define modern e-commerce logistics.

2)Ride-Sharing:Dynamic Platform Operations

Uber’s platform demonstrates integrated-autonomous AI in a service context, coordinating dynamic pricing, driver dispatch, route optimization, and demand prediction in real time across millions of rides daily. AI algorithms continuously balance supply and demand, adjusting prices to attract drivers to high-demand areas while optimizing routes to minimize passenger wait times. The system integrates payment processing, safety monitoring, and customer feedback into a comprehensive operational platform. Human intervention focuses on policy decisions and exception handling rather than individual ride coordination.

3)Energy:Autonomous Grid Management

Electric utilities deploy AI systems that autonomously manage power grid operations, balancing supply and demand in real time across thousands of generation sources and millions of consumption points. These systems integrate data from smart meters, weather forecasts, renewable generation facilities, and industrial consumers to predict demand, optimize generation scheduling, and manage grid stability. AI can autonomously dispatch resources, adjust voltage levels, and reroute power flows to prevent outages. Human operators monitor overall system performance and handle strategic decisions, but routine operations are fully automated.

Table 4.Examples: integrated-autonomous configuration.

Across industries, these examples illustrate that organizations rarely adopt AI uniformly across all processes. Instead, they deploy different configurations depending on risk tolerance, value potential, regulatory constraints, and organizational readiness. Digital-native organizations often progress more rapidly toward integrated-autonomous configurations due to stronger data foundations and architectural flexibility, while legacy organizations must modernize existing systems while maintaining operational continuity.

The framework clarifies these strategic choices by highlighting how automation and integration jointly shape AI’s organizational impact. Movement toward more advanced configurations requires not only technical capability but also organizational change management, process redesign, and governance structures appropriate to increased autonomy.

The framework contributes to IS research on technology adoption and digital transformation in several ways. First, it synthesizes recent enterprise AI research into a coherent strategic lens that captures how AI differs from earlier generations of enterprise technology. The automation dimension reflects the fundamental shift in human-technology relationships that distinguishes AI—where systems actively participate in decision-making rather than merely supporting human activities [13][25][37].

Second, the framework highlights integration as a recurring factor associated with the transformation of AI from localized analytical capability into enterprise-wide intelligence. This finding resonates with longstanding IS research on the value of interoperability and information exchange, while extending these insights to contexts where integrated systems can autonomously coordinate actions across organizational boundaries. The MIT CISR Enterprise AI Maturity Model similarly emphasizes that organizations in advanced stages of AI maturity build ‘modular, interoperable platforms and data ecosystems to enable enterprise-wide intelligence’ [26][38]. Future research should examine the specific integration mechanisms—technical, organizational, and governance-related—that enable progression toward more advanced configurations.

Third, by positioning AI implementations within a two-dimensional strategic space, the framework enables comparative analysis across industries, organizational contexts, and time periods. Researchers can examine which configurations are associated with different outcomes, how organizations move across quadrants over time, and what factors enable or constrain progression. This approach complements case-based research on individual AI implementations by providing a common vocabulary for cross-case comparison.

The framework also raises questions for future IS research. How do organizational factors such as data maturity, technical capability, and change management capacity influence optimal configuration choices? What governance mechanisms are most effective at different points in the automation-integration space? How do industry-specific regulatory environments shape feasible progression paths? These questions connect AI implementation research to broader IS traditions in technology adoption, organizational change, and IT governance.

For executives and managers, the framework provides practical guidance for AI implementation strategy. Several implications emerge from the analysis.

Assess current positioning.Organizations should evaluate where their existing AI applications fall within the framework. Many organizations maintain a portfolio of applications across multiple quadrants, reflecting different maturity levels and strategic priorities. Understanding this portfolio provides a foundation for identifying gaps, redundancies, and opportunities for progression.

Prioritize integration infrastructure.The evidence reviewed here recurrently points to integration as an important enabler of AI value creation. Organizations that invest in data infrastructure, API architectures, and interoperability standards create foundations for scaling AI impact across functions. Conversely, organizations with fragmented data environments and disconnected systems will struggle to progress beyond isolated implementations, regardless of the sophistication of individual AI tools.

Concentrate resources on high-impact initiatives.Research indicates that organizations achieving meaningful AI value focus on a limited number of high-impact use cases rather than pursuing numerous disconnected pilots [10]. The framework helps identify which applications have the greatest potential for progression and enterprise-wide impact, enabling more disciplined investment prioritization.

Match configuration to context.The framework does not prescribe universal progression toward integrated-autonomous systems. In highly regulated industries, contexts requiring significant human judgment, or applications involving substantial risk, integrated-supervised configurations may represent appropriate end states. Organizations should match AI configurations to the specific requirements of each domain rather than pursuing maximum automation or integration for its own sake.

Develop governance that scales with autonomy.As organizations move toward more autonomous and integrated configurations, governance requirements intensify. Organizations should develop governance structures and risk management mechanisms appropriate to their current and target configurations, recognizing that approaches suitable for isolated-supervised applications will be insufficient for integrated-autonomous systems.

The framework also has implications for policymakers concerned with AI governance, workforce development, and economic competitiveness.

Regulatory frameworks should accommodate configuration diversity.AI applications in different quadrants present different risk profiles and governance requirements [39][40]. Regulatory approaches that treat all AI applications uniformly may impose unnecessary burdens on low-risk implementations while providing insufficient oversight for high-autonomy, high-integration systems. Configuration-aware regulatory frameworks could provide more proportionate oversight aligned with actual risk levels.

Workforce development should address human-AI collaboration.As organizations progress across the framework, the nature of human work shifts from direct task execution toward supervision, exception handling, and strategic oversight. Workforce development programs should prepare workers not only for displacement but also for new roles in human-AI collaborative systems. This includes technical skills for working with AI tools and judgment skills for effective oversight of autonomous systems.

Support for integration infrastructure.The evidence reviewed here suggests that integration capability is among the important factors shaping AI value creation. Policies that support development of shared data infrastructure, interoperability standards, and integration platforms—particularly for small and medium enterprises that may lack resources for independent development—could accelerate AI adoption and value creation across the economy [41].

Several promising directions for future research emerge from this framework.

Longitudinal analysis of progression paths.Empirical research could examine how organizations move across framework quadrants over time, identifying common progression paths, barriers to advancement, and factors that accelerate or impede movement. Such research could test whether particular sequences of implementation are associated with superior outcomes.

Configuration-performance relationships.Future studies could examine whether specific configurations are associated with different performance outcomes, and how these relationships vary by industry, organizational size, and strategic context. This research would help refine understanding of when different configurations are appropriate.

Governance mechanisms across configurations.Research could investigate how governance structures, risk management practices, and accountability mechanisms should differ across framework quadrants. This includes examination of technical controls, organizational processes, and human oversight approaches appropriate to different levels of automation and integration.

Human-AI collaboration dynamics.As organizations implement AI across the framework, the nature of human-AI interaction changes substantially. Research could examine how work roles, skill requirements, and organizational structures evolve as AI systems become more autonomous and integrated, building on emerging research on collaborative intelligence [18][21].

Cross-industry comparative analysis.Comparative research across industries could identify sector-specific patterns in AI implementation, including how regulatory environments, competitive dynamics, and industry structure shape configuration choices and progression opportunities.