By Christos Makridis, PhD, Senior Researcher, Gallup; Research Associate Professor, W. P. Carey School of Business Arizona State University; Digital Fellow, Stanford University, Digital Economy Lab; Chair, CGCI, National Security Research Alliance (NSRA)
At the Council on Global Competition and Innovation – CGCI, we view advanced AI as one of eight critical technologies that will shape U.S. economic dynamism and national security over the next decade. For investors, the question is not only which models or chip providers will win, but which organizations—and which governance structures—can translate AI capability into durable operational advantage.
Artificial intelligence policy discussions still focus on measurable technical inputs. Compute capacity, data access, and chip supply dominate debates about competitiveness, while model capability is often treated as an engineering attribute rather than an organizational one. Investment statistics track capital expenditures on software and infrastructure, and especially in the defense community procurement strategies emphasize acquiring tools.
And yet the economic and strategic impact of AI—like that of earlier technological shifts, including computers—depends not on acquiring the technology but on adapting organizations and institutions to it. That is where both productivity gains and competitive advantage actually materialize.
Across industries and government agencies, the current policy push around AI adoption is understandable and often justified, particularly given the stakes for both economic growth and national security posture. Leaders are encouraging employees to try tools, procurement offices are expanding licenses, and training programs are scaling rapidly. These efforts address a real barrier, since perceived utility matters for initial uptake. But adoption is only the entry point. The binding constraint quickly becomes internal diffusion: who uses the technology routinely, in what decisions, and under what authority.
In practice, organizations purchase AI broadly and quickly but utilize it unevenly. Some teams move from occasional experimentation to embedded workflow use, while others stall at sporadic trials. Historically, the difference is rarely technical access. It is whether management defines when technology can be used, how errors are handled (especially when experimenting), and whether employees are rewarded or penalized for relying on it. Past research on the boom in computers found that U.S. firms do not outperform simply because they invest more in IT, but because they pair IT with structured management practices that translate information into coordinated action—especially through clearer targets, monitoring, and performance-linked incentives. That is what turns technology spend into a differentiated, compounding asset for investors.
This distinction matters because AI functions less like a labor saving tool and more like a general purpose augmentation technology. It expands the amount of analysis available to decision makers and compresses response time, which shifts the constraint from producing information to integrating it. As a result, the relevant complement is not only worker skill but organizational design: decision rights, escalation rules, evaluation criteria, and communication practices. Where these change, AI becomes part of the operating system of the institution. Where they do not, it remains a peripheral productivity aid.
Together with Erik Brynjolfsson, we argued that governments across the world need to do more to properly value the adoption of AI in national statistics. We show that much of what today’s firms build–from internal model deployment frameworks to customized data pipelines and AI-augmented workflows–is treated in existing accounts as routine operating expense rather than capital investment, which understates both the stock of AI capability and its accumulation over time. In addition, rapid quality improvements and bundled digital services make it difficult to separate out where value is actually created versus where it merely shows up as a cost in the accounting data. That blind spot weakens policy, distorts macro indicators, and obscures how AI adoption is reshaping organizational behavior across sectors. For investors, that mismeasurement makes it harder to distinguish firms that have genuinely embedded AI capability from those that merely report higher technology spend.
Only by measuring internal use, workflow integration, and the accompanying reallocation of labor and decision rights can statistics capture the productivity contributions of AI as it diffuses through firms and agencies. Without these measurements, productivity and readiness will appear disconnected from technological adoption and policymakers will misdiagnose the extent to which AI is driving structural change in the economy and national security apparatus.
Micro level evidence confirms the role of the organizational layer.
Using longitudinal data covering more than 30,000 workers from 2023 to 2025 from the Gallup Workforce Panel, I document that AI adoption rose rapidly but unevenly across hierarchy in the firm, occupation, education, and more. Frequent use increased from about 9 percent to 26 percent over two years. However, the variation across organizations was far larger than the variation across occupations or demographics. The strongest predictor of frequent use was whether employees believed leadership had communicated a clear strategy for AI integration. Workers in such environments were roughly 26 percentage points more likely to use AI frequently. That changes how AI readiness should be interpreted.
A key margin, therefore, for organizations is not technical access, but strategic clarity.
But even beyond simple AI adoption statistics, the consequences of AI–including on employees and their impact on their firm–depend on that environment. AI use improves engagement and job satisfaction and reduces burnout only when organizations combine clear guidance with high trust. In low clarity environments, the same technology produces weaker gains and in some cases negative outcomes. The technology itself is constant; the organizational context determines whether it functions as an augmentation tool or a source of friction.
AI lowers the cost of generating analysis but increases the need for coordination. Workers must understand when its use is appropriate, how outputs will be evaluated, and whether experimentation is safe. Good management—broadly referring to an atmosphere of trust, respect, and developmental communication—reduces uncertainty about permissible behavior, while trust reduces perceived downside risk of mistakes. Adoption therefore follows hierarchy and communication channels rather than purely technical exposure.
Even in technologically sophisticated organizations, resistance to AI adoption is not purely informational. In many cases the barrier is not literacy but anxiety. Workers may understand the tools and still hesitate to rely on them if they perceive career risk, status erosion, or unclear accountability. Organizational culture therefore matters as much as formal guidance. Leadership can signal that AI is a capability to be mastered rather than a metric for headcount reduction, but that signal must be credible. Without it, diffusion slows even among the technically literate.
This finding is particularly relevant in environments where decisions carry systemic consequences, including financial markets and national security operations. In these settings, CGCI’s mission is to help investors back institutions that can absorb and govern AI effectively so that expanded analytical capacity actually improves accountable decision making.
In both domains, the bottleneck is rarely the availability of information. It is the integration of information into accountable decision processes. AI dramatically expands analytical capacity, including among people that may not have considered themselves “quants”, but unless authority structures adapt, organizations experience signal overload rather than improved judgment.
My research with the Gallup Workforce Panel illustrates this shift within firms. Leadership roles and coordination intensive positions adopt AI earlier and more intensively than individual contributors. The pattern is consistent with a technology whose value increases with responsibility for integrating multiple inputs. AI’s primary effect is therefore not automation of isolated tasks but augmentation of supervisory and decision making functions.
This helps explain a persistent puzzle. Many organizations deploy AI pilots “successfully” yet fail to generate measurable operational improvement. The issue is not technical performance but institutional embedding–and that’s hard work, especially in large organizations. Integrating AI requires redefining responsibility boundaries, feedback loops, and evaluation criteria. Where this occurs, usage becomes routine. Where it does not, usage remains experimental.
In strategic settings, the implications are substantial.
Decision environments in finance and defense both operate under time pressure, uncertainty, and accountability constraints. Historically they solved these problems through layered review structures designed to manage scarce information. AI reverses that condition. Information becomes abundant while interpretation capacity remains limited.
Organizations that maintain legacy decision architectures treat AI as advisory and slow. Organizations that adapt decision rights allow machine generated analysis to accelerate response while preserving accountability. The advantage therefore lies not in possessing more advanced models but in structuring institutions that can act on them.
Counting models, compute clusters, or procurement budgets does not capture operational capability. What matters is whether institutions have reorganized workflows so that AI output changes behavior rather than sitting alongside it.
This distinction is also visible in labor market data when we look at reallocation across, not within, occupations. In joint work with Andrew Johnston using administrative employer data across U.S. state-industry sectors, we find that AI exposure is associated with rising wage bills and employment overall, but only after allowing for reallocation across roles rather than holding occupations fixed. Within specific occupations with greater generative AI exposure, including those with younger workers, employment often declines. Yet sectors with higher exposure to AI that augments human decision making experience simultaneous increases in both wages and employment, indicating an outward shift in labor demand rather than simple substitution.
The positive effects emerge because organizations reorganize tasks, create complementary roles, and allow workers to move toward oversight, integration, and judgment-intensive activities. In other words, disruption at the job level is the mechanism through which growth appears at the sector level. When measurement allows for occupational churn and new task creation, augmentation dominates displacement and net gains become visible.
For policymakers, institutional leaders, and capital allocators, there are at least three takeaways:
First, readiness assessments should evaluate implementation structures rather than technology inventories, because that is ultimately what determines return on invested capital, not just notional access to models. An organization with modest models but clear operational doctrine can outperform one with advanced tools but ambiguous procedures. Technical literacy among senior leaders is valuable, but literacy alone does not guarantee effective integration. The missing factor is often whether institutions create shared standards for evaluating AI output. Decision makers do not need to become engineers, but organizations must develop a common operating language.
Second, workforce investment should prioritize managerial capability. Training individuals to use AI is necessary but insufficient if leadership does not define acceptable use, escalation protocols, and evaluation standards. The key question is not whether employees can use a model, but whether managers intentionally prioritize redesigning workflows properly so AI output informs accountable decisions. That requires clarifying which judgments remain human, how model recommendations are validated, and how performance is assessed in AI-augmented roles. Organizations that treat AI as a training issue will see incremental gains. Those that treat it as a management redesign challenge will see materially larger returns—and will justify continued capital deployment into AI capability rather than one‑off experiments.
Third, national competitiveness depends on institutional learning rates. Countries—and thus the firms and employees within them—that shorten the cycle between experimentation and deployment will capture productivity gains even without frontier technological leadership, and will attract a disproportionate share of patient, long‑horizon capital. Those focused primarily on acquisition risk accumulating technical assets that remain underused. Institutional agility, not just technological sophistication, will define advantage in the AI era.
AI has been described as a software revolution, but, like the technological changes of the past, it is fundamentally about organizations and ultimately people. The strategic advantage will accrue to the institutions that redesign how decisions are made, not merely to those that deploy the most sophisticated systems.
