Executive Summary: AI Is Crossing the Boundary From Intelligence to Agency
For the past several years, artificial intelligence has largely entered the boardroom as a question of intelligence: How effectively can machines analyze information, generate content, predict outcomes, and improve human decision-making? Agentic AI changes the nature of that discussion because AI is beginning to cross an important threshold—from systems that primarily advise people to systems that can increasingly act on their behalf.
Generative AI can prepare a customer letter. An AI agent can potentially identify which customers require attention, retrieve their histories, determine an appropriate action within established parameters, prepare and send the correspondence, update the customer record, and initiate follow-up—bringing a human into the process only when a decision exceeds its authority. What was previously a tool assisting an employee begins to become a digital participant in the operating model.
This is why Agentic AI should not be regarded simply as the next stage of Generative AI. It represents a potential restructuring of how work, decision rights, and accountability are distributed across the enterprise.
The market is moving quickly, although organizational readiness remains uneven. Gartner reported in August 2026 that only 10 percent of organizations have Agentic AI in production. McKinsey’s 2026 global AI research found that 40 percent of respondents from organizations with more than US$1 billion in annual revenue report scaling AI agents, compared with 27 percent a year earlier.
Expectations for what comes next are considerably greater. Deloitte found that 74 percent of leaders expect nearly half of their business processes to be redesigned or rebuilt around AI agents within four years. Sixty-one percent expect most agents to become largely autonomous with humans providing oversight, while 58 percent anticipate agents coordinating with one another across functions. Yet only 5 percent of organizations describe their processes as highly prepared for Agentic AI, and only 15 percent have scaled orchestrated, cross-functional multi-agent systems.
That gap between ambition and readiness may become one of the defining management challenges of the next several years. For boards and CEOs, the strategic issue is therefore not simply how capable AI becomes. It is whether organizations can redesign processes, workforce roles, technology architecture, and governance quickly enough to capture the value without creating levels of autonomy they cannot effectively control.
The leadership question is becoming increasingly clear: What decisions and actions are we prepared to delegate to machines, under what conditions, and who remains accountable when those machines act?
1. Agentic AI Is Easier to Understand Than the Terminology Suggests
The technology industry has developed an expanding vocabulary around Agentic AI: agents, agentic workflows, autonomous agents, digital workers, multi-agent systems, agent orchestration, and agentic enterprises. For board directors and CEOs, much of this language risks obscuring the essential idea. The simplest distinction is between AI that assists work and AI that can increasingly execute work.
OpenAI defines agents as systems capable of independently accomplishing tasks on behalf of users. Unlike a conventional chatbot, an agent can interpret an objective, determine how a workflow should proceed, select tools, gather information, interact with external systems, and evaluate whether the task has been completed.
Consider a CEO preparing for an important customer meeting. A conventional information system might retrieve account records. Generative AI could summarize those records. An agentic system could potentially review the CRM history, examine recent correspondence, identify unresolved service issues, analyze revenue trends, compare commitments with delivery, and prepare the executive briefing. The distinction is not simply more powerful AI. The system has been given a goal rather than an isolated prompt, together with sufficient access and authority to determine how portions of that goal should be accomplished.
As these systems mature, human involvement can move along a continuum. At one end, AI recommends and a human decides. Further along, AI acts but requires approval at defined thresholds. At the more autonomous end, agents perform routine activities independently while escalating exceptions. Every movement along this continuum transfers some degree of operational authority from a human to a machine.
Cindy’s Perspective
The board-level issue is not terminology. It is decision rights. Executives have spent decades defining who may authorize expenditures, access sensitive information, approve transactions, modify records, and commit an organization externally. Agentic AI introduces a new class of actor into that control environment.
The governance discipline should therefore begin with a simple classification: What may the agent observe? What may it recommend? What may it execute? And what must remain exclusively human? That framework is more useful than debating whether a particular technology technically qualifies as an ‘agent.’ The strategic issue begins whenever an AI system receives authority to affect a business outcome.
2. The Real Opportunity Is Process Reinvention, Not Task Automation
The first generation of enterprise AI focused heavily on productivity at the individual level. Employees used AI to prepare documents, summarize meetings, write software, and accelerate administrative work. Agentic AI expands the unit of transformation from the individual task toward the end-to-end process. That is where the economics become more interesting.
Consider customer onboarding. A typical process may involve document collection, identity verification, risk assessment, account creation, customer communications, and record updates across multiple people and systems. An agentic workflow could potentially coordinate much of that sequence continuously, escalating only those decisions requiring judgment or exceeding established risk parameters. The economic benefit does not come simply from making every activity 10 or 20 percent faster. It comes from removing handoffs, reducing waiting time, compressing cycle time, and enabling a smaller number of employees to supervise a substantially larger volume of work.
Deloitte’s research suggests executives increasingly recognize this distinction. Nearly two-thirds of surveyed leaders are reevaluating elements of their business models in anticipation of Agentic AI, while 75 percent believe collaboration between humans and agents will create greater value than automation alone.
Microsoft’s 2026 Work Trend Index provides another signal. Among surveyed AI users, 66 percent say AI enables them to spend more time on higher-value work, while 58 percent report producing work they could not have produced a year earlier. Among Microsoft’s more advanced ‘Frontier Professionals,’ that figure rises to 80 percent. Microsoft also reports that active agents in its 365 ecosystem increased 15-fold year over year and 18-fold within large enterprises.
The organizational advantage may therefore come less from substituting machines for people and more from increasing the productive capacity of skilled employees. A relationship manager might supervise agents monitoring hundreds of customer accounts; a cybersecurity analyst might direct agents screening thousands of events; a finance professional could manage automated reconciliations; and an engineer could orchestrate multiple specialized agents simultaneously. Human work moves upward—from routine execution toward judgment, prioritization, exception management, and accountability.
Cindy’s Perspective
The strategic question is not, ‘How many jobs can we automate?’ It is, ‘How much more value can our best people create when execution is increasingly augmented by intelligent systems?’
Organizations that treat Agentic AI primarily as a cost-reduction programme may capture near-term efficiencies. Organizations that redesign work around the complementary strengths of human judgment and machine-scale execution have the potential to change throughput, customer responsiveness and operating leverage much more fundamentally. The objective should be to remove low-value friction from work—not simply remove people from payroll.
3. Competitive Advantage Will Come From Better Economics, Not More Agents
Technology waves often produce activity measures that masquerade as business outcomes. Cloud programmes counted migrations. Digital programmes counted applications. Early AI programmes frequently counted use cases and pilots. Agentic AI risks repeating the pattern, with organizations reporting the number of agents deployed as though digital population were evidence of transformation. It is not. The relevant measure is whether something important changes in the economics or effectiveness of the enterprise.
McKinsey’s 2026 research provides an early signal. Forty percent of respondents from large enterprises report scaling AI agents, with adoption concentrated particularly where value is easier to quantify. Approximately 31 percent of large organizations are scaling coding agents, while 32 percent of surveyed organizations report deciding not to purchase at least one software product or feature because they could develop the capability internally using agentic coding tools.
Deloitte identifies the obstacles separating experimentation from scale. Seventy-two percent of leaders cite the absence of a unified and accessible data foundation, 70 percent cite an inability to trust and govern agents, and 67 percent identify integration cost and complexity. These findings reinforce an important point: Agentic AI does not magically eliminate fragmented data, poorly designed processes, or disconnected systems. It exposes them.
Cindy’s Perspective
Boards should move the Agentic AI discussion out of the innovation portfolio and into the value-realization agenda. Every material initiative should have an explicit economic hypothesis: What cycle time changes? What cost moves? What capacity is unlocked? What revenue improves? What customer outcome changes? What risk declines?
If management cannot articulate the expected result in measurable business terms, the initiative is probably not ready to scale. The next phase of AI therefore needs to be less concerned with proving technical possibility and considerably more disciplined about demonstrating economic consequence.
4. The Evidence Is Beginning to Move From Pilots to Operating Results
Perhaps the most important development in Agentic AI is that the discussion is beginning to move beyond demonstrations and vendor promises. Across financial services, healthcare, manufacturing and logistics, a small but growing group of organizations is reporting measurable operating results. The evidence remains early, and many results are company- or technology-partner-reported rather than independently audited. Nevertheless, these cases provide valuable signals because they show what happens when AI agents move beyond conversational assistance and become embedded in real operating processes.
Financial Services: Nequi Is Changing the Economics of Credit Management
Colombia’s digital financial-services company Nequi, which serves more than 28 million customers, provides one of the more compelling examples. Nequi has combined machine learning, deterministic workflows, and a multi-agent architecture within its credit operations. The company reports that its AI-driven credit approach has doubled the population eligible for preapproved credit and increased loan approvals for underserved customers by 70 percent. More importantly from an Agentic AI perspective, Nequi reports that its multi-agent approach now helps manage 80 percent of its overdue portfolio internally and digitally while reducing credit-management costs by 50 percent. The significance extends beyond automation: agentic systems are beginning to influence the economics of a core financial-services process while potentially expanding access to credit.
Healthcare: Highmark Health Is Compressing Weeks Into Minutes
Healthcare provides another compelling example because administrative complexity consumes enormous amounts of clinical and operational capacity. Highmark Health has applied AI agents to prior authorization, a process requiring clinical evidence, payer policies, documentation, and multiple system interactions. Deloitte reports that Highmark’s agentic approach can assemble the necessary clinical rationale and initiate prior-authorization requests in real time, compressing processes that historically could take weeks into minutes. Sanofi provides a complementary example of enterprise scale: its AI environment reaches approximately 50,000 employees weekly and 72,000 monthly, has generated approximately 11 million conversations, and is evolving toward an agent hub in which specialized agents can retrieve information, interact with systems, and automate workflows.
Manufacturing: Krones Is Redefining Engineering Cycle Time
In manufacturing, German industrial company Krones demonstrates how agents can move beyond administrative work into engineering. Krones has developed AI agents capable of running and refining engineering simulations and supporting digital twins. Simulation activities that previously required approximately three to four hours can now reportedly be completed in five minutes or less—a reduction of as much as 98 percent. The company plans to extend the approach to as many as 500 engineers. AGCO provides another useful signal: the agricultural-equipment manufacturer has more than 900 employees volunteering to build agents, while some quality-review processes that previously required weeks can reportedly be completed in approximately one hour.
Logistics: OPLOG Demonstrates the Economics of Agentic Orchestration
One of the strongest operating cases comes from OPLOG, a technology-driven fulfillment company operating across multiple countries. OPLOG has implemented an AI-agent orchestration system that makes thousands of operational decisions each day. The reported results are substantial: decision-making time has declined by approximately 90 percent; resource utilization has increased to 85–95 percent; operating costs have fallen 30–40 percent; and the same management team can oversee approximately three times the previous operational scale. Customer satisfaction increased eight percentage points and service-level performance improved by 3.5 percentage points. This is important because the agent is not simply helping an employee perform a task; intelligence has effectively become an orchestration layer within the operating system of the business.
Transportation and Global Logistics: DHL Is Industrializing Agentic AI
DHL demonstrates what happens when a global incumbent begins moving agents into high-volume operational processes. Agentic AI has become part of DHL Group’s Strategy 2030, with agents being applied to communications, transportation coordination, shipment management, appointment scheduling, driver follow-ups and high-priority order management. DHL Global Forwarding has disclosed an example involving approximately 2.5 million annual air- and ocean-freight booking requests. Its Agentic AI solution extracts information from emails, documents, and attachments, structures more than 70 data fields, and moves that information into shipment systems while retaining human validation where necessary. DHL reports an approximately 40 percent productivity improvement in booking and enrichment activities, together with better data quality and reduced rework. The case demonstrates bounded autonomy in practice: machines perform high-volume interpretation and execution while people remain positioned at consequential control points.
Cindy’s Perspective
The emerging evidence changes the Agentic AI discussion. We are no longer asking whether an agent can answer a question or automate an isolated administrative activity. Leading organizations are beginning to demonstrate material changes in the economics of entire workflows.
What is striking is the diversity of the outcomes. Nequi is changing the economics of credit management. Highmark is compressing a healthcare process from weeks to minutes. Krones is radically accelerating engineering simulation. OPLOG is enabling the same management team to supervise three times the operational scale while reducing costs by as much as 40 percent. DHL is applying agents to millions of freight transactions while preserving human validation at critical control points. These are different businesses, yet the transformation pattern is remarkably consistent.
The common denominator is not the AI model. It is process redesign. Each organization has identified a high-friction workflow, provided AI with sufficient context and authority to execute meaningful portions of it, connected the agent to operational systems, retained human intervention where judgment matters, and measured the business outcome.
As access to capable foundation models becomes increasingly democratized, sustainable advantage is unlikely to come simply from possessing an AI model competitors cannot access. Advantage will come from how intelligently an organization redesigns proprietary workflows, connects agents to trusted enterprise data, establishes appropriate autonomy boundaries, and learns faster than competitors how to convert machine agency into economic value. Boards should therefore ask for evidence of process economics rather than demonstrations of AI sophistication.
5. The Core Governance Challenge Is Defining Where Autonomy Stops
Agentic AI becomes most powerful at exactly the point where governance becomes more difficult: when a system can act without waiting for a person. Organizations are therefore not merely deploying software. They are beginning to reallocate decision authority.
The World Economic Forum has described this transition as the movement from systems of record toward systems of trust, where autonomous technologies increasingly influence or execute decisions previously controlled by human actors. This is where the concept of an autonomy boundary becomes useful.
An autonomy boundary defines the point beyond which an agent cannot proceed independently. The boundary may be financial, operational, regulatory, ethical, or reputational. A purchasing agent might reorder routine supplies within a predetermined threshold but require approval above it. A customer-service agent might issue a modest credit but not terminate an account. A cybersecurity agent might isolate a low-risk endpoint automatically while requiring human authorization before disabling critical infrastructure.
The objective is not to eliminate autonomy. It is to make autonomy deliberate, bounded, and observable. The challenge becomes greater as agents interact with one another: an incorrect instruction passed automatically through multiple agents can propagate across functions and systems at machine speed.
Cindy’s Perspective
The critical governance decision is not whether to permit autonomy. It is where autonomy should stop. Boards should expect management to define that boundary explicitly for every material agentic process. As the potential financial, customer, regulatory, or reputational consequence increases, the autonomy boundary should tighten.
Boards should also test governance through failure rather than relying on demonstrations of success. Can management identify when an agent has exceeded its mandate? Can the workflow be interrupted? Can the organization reconstruct what happened and reverse the action? The relevant standard is not perfection. It is whether an agent can fail within controlled limits.
6. Agentic AI Creates Risk at Machine Speed
The security implications deserve particular attention because agents combine capabilities that historically remained separated. A single system may interpret information, make decisions, access enterprise tools, and execute actions within one continuous workflow. That expands the attack surface considerably.
In 2026, NIST reported broad agreement among industry respondents that AI agents introduce novel security threats. OWASP’s Agentic AI framework similarly identifies risks including agent goal hijacking, tool misuse, identity and privilege abuse, supply-chain vulnerabilities, unexpected code execution, memory poisoning and cascading failures among interconnected agents.
Organizational preparedness is not keeping pace. Deloitte estimates that roughly 80 percent of surveyed organizations do not yet possess mature Agentic AI governance capabilities, including clearly defined decision boundaries, real-time monitoring and comprehensive audit trails. The concern is therefore not simply that an agent can make a mistake; it is that an agent can execute or propagate a mistake at a speed and scale fundamentally different from normal human operations.
Cindy’s Perspective
The governing principle should be unequivocal: Delegation of activity does not transfer accountability. If an agent exposes confidential information, executes an inappropriate transaction, or adversely affects a customer, accountability remains with the organization that designed, authorized, and deployed the system.
Every material agent should therefore have a named accountable executive, narrowly defined permissions, observable behaviour and a reliable mechanism for intervention. If accountability is ambiguous, autonomy has moved further than governance.
7. Workforce Transformation Will Determine Whether Agentic AI Creates Durable Value
One of the most underappreciated consequences of Agentic AI is its potential to reshape organizational design itself. Traditional enterprises are structured around people executing processes through functional departments. Work moves between marketing, sales, finance, operations, and technology through meetings, approvals, and handoffs. Agents do not necessarily respect those organizational boundaries.
An orchestrated workflow can potentially move from customer inquiry through account analysis, pricing, inventory verification, contract preparation, and finance without stopping simply because those activities sit within different departments. This creates the potential for much faster execution, but it also challenges traditional roles, management spans, and organizational structures.
Microsoft’s research indicates that readiness remains uneven. Only 26 percent of surveyed AI users believe leadership is clearly and consistently aligned around AI. Sixty-five percent fear falling behind if they fail to adapt quickly, yet 45 percent say it feels safer to remain focused on existing objectives rather than redesign work. Only 13 percent say they are rewarded for reinventing work when experimentation may not immediately produce conventional performance outcomes. These are not technology barriers. They are leadership and organizational barriers.
Employees will increasingly need to learn not simply how to use AI, but how to delegate to it, supervise it, evaluate it, and intervene when necessary. Managers may eventually supervise teams containing both people and digital agents. Job descriptions may need to specify not only human responsibilities but which agents an employee oversees and where human judgment remains mandatory.
Cindy’s Perspective
The workforce question is ultimately one of organizational architecture. If intelligent agents can execute work across functions, leadership cannot simply insert them into existing jobs and assume the operating model should remain unchanged.
Organizations should instead begin with three questions: Which work should remain human? Which work should move to machines? Where must judgment transfer between the two? This requires more than reskilling; it requires leaders to reconsider roles, decision rights, spans of control, performance measures, and career paths. Companies that automate without redesign may simply make inefficient processes move faster.
Executive Conclusion: The Agentic Enterprise Will Be Designed, Not Installed
Agentic AI remains early enough that boards should be skeptical of predictions suggesting enterprises will become fully autonomous in the immediate future. Gartner’s finding that only 10 percent of organizations currently have Agentic AI in production provides an important counterweight to the market enthusiasm. Yet the operating evidence is beginning to matter.
Nequi reports a 50 percent reduction in credit-management costs. Highmark is compressing portions of prior authorization from weeks to minutes. Krones reports engineering simulation improvements approaching 98 percent. OPLOG reports 30–40 percent lower operating costs and three times greater management scale. DHL reports approximately 40 percent productivity improvement in a workflow processing millions of annual freight requests. These results do not prove that every organization will achieve similar economics, but they demonstrate that the Agentic AI discussion is moving from technological possibility toward measurable operating outcomes.
AI is evolving from answering questions toward completing work. Employees are moving from prompting machines toward delegating objectives. Individual AI applications are connecting through orchestrated workflows. Decision-making is shifting from entirely human execution toward bounded machine autonomy.
The defining leadership challenge is not deployment. It is design. Organizations must determine where machine agency creates genuine competitive advantage, where autonomy should stop, how accountability will remain unmistakably human, which processes require fundamental redesign, and how workforce capabilities must evolve.
The emerging competitive divide is therefore unlikely to be between companies that adopt AI and companies that do not. AI adoption will eventually become pervasive. A more consequential divide may emerge between organizations that add agents to existing processes and organizations that redesign their operating models around agency. The second group has the potential to create fundamentally different economics.
Twelve Questions Every Board and CEO Should Be Asking
1. Where could Agentic AI fundamentally change the economics or customer experience of our business rather than simply automate isolated tasks?
2. Which AI systems currently have authority to act, and does leadership understand precisely what each is permitted to do?
3. What is the autonomy boundary for every material agent, and which decisions must always return to a human?
4. Who is the named executive accountable for each material Agentic AI deployment?
5. Can we reconstruct every significant action an agent takes, including the data, systems, and tools it accessed?
6. Can we stop an agent quickly, contain an error, and reverse an inappropriate action?
7. Are agents operating according to least-privilege principles, or are we unintentionally creating digital identities with excessive enterprise access?
8. How are we testing for goal hijacking, tool misuse, privilege abuse, memory poisoning, and cascading failures?
9. What measurable business value is each major Agentic AI initiative expected to produce, and are those outcomes reflected in enterprise KPIs?
10. How will agents change workforce requirements, skills, management spans, and organizational design?
11. Which processes should be fundamentally redesigned rather than simply adding AI to existing workflows?
12. Are we moving quickly enough to remain competitive while maintaining sufficient control to protect customers, employees, shareholders and enterprise reputation?
Cindy’s Closing Perspective
Agentic AI is significant because it shifts the enterprise AI conversation from capability to authority and from experimentation to operating-model design. That is the inflection point boards should focus on.
When AI primarily generated insight, the dominant leadership questions concerned accuracy, adoption, and productivity. As AI increasingly executes work, the questions become more structural:
-
Where should decision rights reside?
-
What authority should machines possess?
-
Where should autonomy stop?
-
How should people and agents work together?
-
How will economic value be measured?
-
And how will accountability remain unequivocally human?
These are CEO and board questions, not technology-department questions.
The evidence from Nequi, Highmark, Krones, OPLOG and DHL is instructive because these organizations operate in very different industries, yet the underlying transformation pattern is similar. Value is emerging where organizations move beyond adding AI to individual jobs and instead redesign high-friction processes around data, intelligent execution, human judgment, and measurable outcomes.
This is also why the strategic objective should not be maximum autonomy. Maximum autonomy is a technology aspiration. Optimal autonomy is a business design decision. The appropriate level of agency will differ by process, risk, customer impact and regulatory environment. A logistics routing decision and a medical decision should not share the same autonomy boundary simply because the underlying AI is capable of executing both.
Boards must therefore resist two extremes. The first is excessive caution that protects the existing operating model while competitors redesign theirs. The second is premature autonomy that transfers consequential decisions to machines before data, controls, and accountability are ready. The strategic opportunity lies between them.
The organizations that lead the Agentic era will not necessarily possess superior access to AI models. Increasingly, those capabilities will be widely available. Advantage will come from proprietary process knowledge, trusted data, thoughtful human-machine design, disciplined governance, and the organizational ability to learn faster than competitors.
The ultimate leadership question is no longer simply, ‘What can Agentic AI do?’ It is: ‘What should our organization become now that intelligent systems can increasingly act?’ The answer will shape far more than an AI strategy. It will shape the future operating model of the enterprise.
Lady Whistledown’s Agentic Society Papers
Dearest Gentle Readers,
It appears the corporate world has acquired a fascinating new employee. They require neither an office nor an expense account. They do not complain about meetings, rarely request a vacation, and appear entirely willing to work through the night. Most intriguing of all, management would now like to give them authority.
Only recently, artificial intelligence was considered rather accomplished if it could prepare a respectable memorandum. How innocent those days suddenly seem. The newest machines would prefer to read the correspondence, examine the customer record, determine what ought to happen next, update the system, send the response, and ask another machine to complete whatever remains.
One cannot help admiring the initiative. One might also ask who exactly is supervising them.
Corporations have spent centuries developing approval limits, audit committees, segregation of duties and compliance departments designed to ensure that human beings remain within their assigned authority. And yet some enterprises now appear remarkably eager to grant their newest digital employees access to multiple systems and simply hope everyone behaves. One suspects the auditors may eventually wish to join the conversation.
Still, this author would not advise resisting these ambitious newcomers. A well-governed agent may become one of the most productive colleagues an executive has ever encountered. An ungoverned one, however, possesses the remarkable ability to make thousands of mistakes before the first employee has finished breakfast.
And so, gentle readers, while every fashionable board prepares to announce its Agentic AI ambitions, perhaps it should begin with a decidedly unfashionable question: What, precisely, have we given the machine permission to do? The answer may tell us considerably more about the quality of corporate leadership than it does about the sophistication of the AI.
Until our next gathering, dear reader,
Lady Whistledown
Research Bibliography
Amazon Web Services. Nequi Success Story. Nequi scales financial inclusion and credit management using multiagent AI on AWS. Source
Amazon Web Services. OPLOG Case Study. OPLOG accelerates fulfillment decision-making with Amazon Bedrock AgentCore. Source
Amazon Web Services. Ryanair: Agentic AI Innovator Case Study. 2026. Ryanair applies agentic AI to high-volume customer interactions. Source
Amazon Web Services. Sanofi Success Story. 2026. Sanofi builds an agentic AI hub for a global workforce using Amazon Bedrock. Source
Deloitte. 2026 US Health Care Outlook: Three Critical Strategies Shaping a More Resilient, Technology-Enabled Future. December 11, 2025. Deloitte Insights. Includes the Highmark Health prior-authorization example. Source
Deloitte. Agentic AI Is Scaling Faster Than Guardrails. April 24, 2026. Deloitte Insights. Source
Deloitte. AI Agents Are Only the Beginning: Deloitte Survey Examines the AI Readiness Gap and Reveals How Enterprises Can Prepare for Agentic Success. August 12, 2026. Deloitte US press release. Source
Deloitte. AI Agents Are Only the Beginning: The Path to Agentic Transformation. August 12, 2026. Deloitte Insights. Source
DHL Group. 2025 Annual Report / Reporting Hub. 2026. Strategy 2030 and enterprise AI transformation reporting. Source
DHL Group. DHL Global Forwarding Capital Market Briefing. March 2026. London. Includes Agentic AI booking and enrichment use-case metrics. Source
Gartner. Agentic AI Value, Cost and Skills: A C-Suite Guide. August 7, 2026. Analysts: Frances Karamouzis and Kristen Richardson. Source
McKinsey & Company. The State of AI in 2026: On the Road to ROI. August 25, 2026. Global survey. Source
Microsoft. 2026 Work Trend Index: Agents, Human Agency, and the Opportunity for Every Organization. May 5, 2026. Microsoft WorkLab. Source
Microsoft. AGCO Scales Employee-Built AI Agents with Microsoft Copilot Studio. July 6, 2026. Microsoft Customer Stories. Source
Microsoft. Krones Cuts Simulation Run Times by Up to 98% with AI Agents and Azure High-Performance Computing. July 28, 2026. Microsoft Customer Stories. Source
National Institute of Standards and Technology (NIST). Summary Analysis of Responses to the Request for Information Regarding Security Considerations for AI Agents. May 18, 2026. Riggs, J.; Hamin, M.; Perry, N.; Edelman, B.; and Cihon, P. NIST Trustworthy and Responsible AI Report 800-5. Source
OpenAI. A Practical Guide to Building AI Agents. Business guide on agent design, orchestration, tools, guardrails, and human intervention. Source
OWASP GenAI Security Project. OWASP Top 10 for Agentic Applications for 2026. December 9, 2025. Globally peer-reviewed framework for security risks in autonomous and agentic AI systems. Source
World Economic Forum. AI Agents in Action: A Playbook for Trusted Adoption, Authorization and Scaling. May 26, 2026. Developed in collaboration with Capgemini. Source
World Economic Forum. From Systems of Record to Systems of Trust: A Board-Level Playbook for Governing Agentic AI. April 28, 2026. Board-level perspective on delegation boundaries and governance. Source
Evidence note: Company and technology-partner case studies are useful operating evidence but should not be interpreted as independently audited performance studies unless explicitly stated by the source.
