From JPMorgan's COiN in 2017 to the recent LLM Suite, AI automation has been built up over nearly a decade, while Siemens has deployed AI agents to factories. Based on data, we examine the reality of AI transformation (AX) in 2026 between companies that remained stuck in pilots and those that redesigned core tasks.
According to the 'State of AI in the Enterprise' survey conducted by Deloitte between August and September 2025 and released in January 2026, 34% of companies responded that they are fundamentally reimagining new product development, core processes, or business models using AI. Meanwhile, 30% responded that they are redesigning core processes around AI, and 37% remained at a surface-level adoption stage with little change to existing business processes. The survey was conducted among 3,235 executives at the director level or higher across 24 countries. This indicates that while accessibility to AI tools has expanded rapidly, only about a third of companies have connected this to the redesign of tasks and business models.
A similar trend was confirmed in McKinsey's global AI survey, conducted in the summer of 2025 and published in November of that year. The proportion of companies regularly using AI in at least one business function increased to 88% from 78% the previous year, but organizations that scaled this to enterprise-wide value creation remained a minority. Two-thirds of the responding companies stated that they still remain in the experimental or pilot stage, and only 39% answered that AI has an impact on corporate earnings (EBIT) (with most improvements falling below 5 percentage points). This gap between 'using' and 'achieving results' is pointed out as a core variable for global AX continuing into 2026.
Background: From the 'Era of Introduction' to the 'Era of Redesign'
If the AI adoption of global companies in 2023–2024 focused on distributing productivity tools centered around chatbots and copilots, AX post-2025 is moving toward a stage where business processes themselves are redesigned around AI, and 'Agentic AI'—which autonomously performs multi-step tasks instead of humans—is deployed in the field. In McKinsey's survey, 62% of responding companies stated they use AI agents at least in the experimental stage, and 23% stated they have elevated agents to the scaling stage in some areas. However, viewed at the individual business function level, organizations operating agents at actual scaled levels remained at around 10%. A separate survey conducted by Gartner among CIOs showed that 17% of companies answered they had actually deployed AI agents. However, there are limits to directly comparing figures as definitions and measurement units for 'adoption,' 'scaling,' and 'field deployment' vary by research institution. Nevertheless, the directional consensus that core task redesign and enterprise-wide scaling are sluggish compared to experimentation and expanded accessibility is commonly confirmed across multiple surveys.
Investment scale is increasing steeply. According to Deloitte's survey, employee access to AI within companies expanded by 50% throughout 2025. Currently, 25% of companies reported that they have transitioned over 40% of AI experiments into actual production environments, while 54% responded that they expect to reach this level within the next 3 to 6 months. It is worth noting that this is a corporate projection at the time of the survey rather than actual achieved figures.
On the other hand, there is also data showing the discrepancy between expectations and reality. According to the 'AI Adoption in the Enterprise' survey (conducted from December 2025 to January 2026 among a total of 2,400 respondents, including 1,200 C-suite executives and 1,200 employees) released in April 2026 by corporate AI firm WRITER in partnership with research firm Workplace Intelligence, 79% of responding executives answered that they experienced a double-digit increase in difficulties during AI adoption compared to the previous year. Seventy-five percent of executives admitted that their AI strategy is closer to 'showcase' than practical guidance, and only 29% responded that they are reaping meaningful return on investment (ROI) from generative AI. This is evidence that the gap between flashy declarations and actual performance is substantial.
The pace of adoption itself is faster than any technology in history. According to the AI Index released in April 2026 by the Stanford Institute for Human-Centered Artificial Intelligence (HAI), generative AI reached a 53% usage rate relative to the population within three years of its mass-market release, tracing a faster diffusion curve than personal computers or the internet. The problem is that this individual-level diffusion does not immediately lead to organizational-level performance. The report 'The GenAI Divide: State of AI in Business 2025,' published in July 2025 by Project NANDA at the MIT Media Lab, analyzed 52 executive interviews, 153 surveys, and about 300 publicly available AI adoption cases. It diagnosed that 95% of the analyzed generative AI pilots failed to reach measurable profit and loss (P&L) effects, naming this the 'GenAI Divide.' While it should be taken into account that this is based on case analysis rather than a random sample survey representing all enterprises, the point that operational workflow integration and organizational learning capability rather than technical performance determine results aligns with the findings of other surveys.
Case Study ①: JPMorgan Chase—A Decade of Accumulated AX, Leading the Financial Sector with '450 Operational Use Cases'
U.S.-based JPMorgan Chase is cited as the most advanced AX case in the financial sector. The bank's AX did not happen overnight; it is the result of long-term accumulation. JPMorgan introduced the legal document analysis platform 'COiN' in 2017 to automate commercial loan agreement review tasks that previously required 360,000 hours annually. Afterward, in early 2024, it launched the in-house generative AI platform 'LLM Suite,' deploying it to over 230,000 executives and employees worldwide. Recently, it has been in the stage of expanding these two currents into numerous practical AI use cases across the enterprise. According to the company, more than 450 AI use cases are currently in the operational stage, with plans to expand this to 1,000 by the end of 2026.
Performance is also being disclosed in concrete figures. According to Forbes in July 2026, employees using the LLM Suite report a 30% to 40% improvement in work efficiency, and CEO Jamie Dimon stated that through these overall AI initiatives, including cost savings, up to $2 billion in annual business value is being generated. What is noteworthy is JPMorgan's approach. According to multiple media outlets such as The Digital Banker and Emerj, JPMorgan chose a structure that selects models from multiple vendors—such as OpenAI and Anthropic—by use case within a secure control environment, embedding data protection and compliance from the platform design stage to suit the characteristics of a regulated industry. It is evaluated as increasing the speed of diffusion by deeply embedding AI into actual workflows, such as customer meeting preparation for wealth management advisors and drafting performance evaluations, rather than stopping at tool deployment. The core of JPMorgan's case is that it is not a one-off automated case from 2017, but a trajectory accumulated over nearly a decade of automation experience and expanded into generative AI.
Case Study ②: Siemens—'Industrial AI Agents' Deployed to Factory Floors
In manufacturing, Germany's Siemens sets the benchmark for industrial-site AX. Siemens has expanded its generative AI-based 'Industrial Copilot' to its own factories and customers, and reports of improved code quality and development speed among engineers have emerged at Thyssenkrupp Automation Engineering. In June 2026, the company unveiled 'Intelligence Center X,' which bundles industrial data, workflows, and AI agents into a single controlled system. In early customer adoption cases disclosed by Siemens, manual burdens were reportedly reduced by up to 95% and production issue resolution speed increased by up to 85%. These figures are based on initial cases disclosed by the company and actual effects may vary depending on the industry, process, and scope of adoption.
Siemens' strategic direction is clear: the next battleground for AI is not the office, but the physical world—factories, power grids, and infrastructure. To this end, the company is partnering with Nvidia to build an 'Industrial AI Operating System,' stepping up step-by-step to build the world's first fully AI-based adaptive manufacturing site starting in 2026, using its Erlangen electronics plant in Germany as the first blueprint. In fact, in April 2026, a humanoid robot based on Nvidia's physical AI stack successfully demonstrated 8.5 hours of autonomous logistics operations at the Erlangen plant, marking its first milestone. At CES 2026, it also unveiled a cooperative case with PepsiCo to convert U.S. manufacturing and logistics facilities into high-precision 3D digital twins to simulate overall factory operations and supply chains. The calculation is to make 175 years of accumulated industrial data and domain expertise a moat that frontier AI companies cannot challenge.
Analysis: Three Common Traits of High-Performing Companies
Synthesizing advanced corporate cases and major survey data, the common traits of companies achieving substantial results in AX can be largely summarized into three points. First, they start from workflow redesign rather than tool adoption. According to McKinsey, only about 6% of total companies are 'AI high performers' that achieved over 5% in earnings improvement through AI, and these companies fundamentally redesigned work flows themselves at a rate approximately three times higher than other companies. Deloitte's diagnosis is also that redesigning a core task from start to finish (end-to-end) around AI and then horizontally expanding it increases the probability of success.
Second, they establish governance and 'human-in-the-loop' systems prior to scaling. In McKinsey's survey, 65% of high-performing companies had formalized procedures for humans to verify AI outputs, compared to only 23% of other companies. In a situation where security and risk are cited ahead of regulatory uncertainty as the biggest obstacles to the diffusion of agentic AI, control systems are becoming a prerequisite that enables scaling rather than a cost that slows down speed. Third, they set growth as a goal alongside efficiency. McKinsey analyzed that companies extracting the greatest value from AI tend to include growth and innovation in their goals alongside cost reduction. In reality, JPMorgan manages expanded customer coverage per advisor, while Siemens manages the expansion of new software businesses in parallel as performance indicators for AX.
The pace of diffusion across industries is not uniform. Technology, software, finance, and telecommunications industries belong to the axis leading field deployment. On the other hand, industries combining physical facilities and operational technology (OT), such as manufacturing and energy, face high difficulty in data connection and legacy system integration, resulting in fundamentally different diffusion methods. Healthcare is a somewhat complex case. While interest is high enough that 85% of healthcare leaders in McKinsey's survey answered they have already adopted or are exploring generative AI, the rate of actually deploying agents into production remained low compared to other industries, classifying it as an industry that is 'fast to explore but slow in field deployment.' However, given that the absolute scale of value created when manufacturing is combined with physical assets—such as digital twins, predictive maintenance, and process optimization, as seen in the Siemens case—is massive, observers note that 2026–2027, when lagging industries begin aggressive catch-up, could be an inflection point for reshaping industrial landscapes.
Outlook: Checkpoints for the Second Half of 2026
Deloitte projected that by the end of 2026, the gap between companies that have transitioned from pilots in at least one core function to redesigned operations and those that have not will become visible in earnings data and board agendas. In Deloitte's survey, only 4% of companies answered that they regularly report AI value creation to the board of directors, but Deloitte forecasted that this practice will realistically become an expected capability for large listed companies by the end of the year. For South Korean companies, three checkpoints deserve attention. First is the field deployment rate of agentic AI. While Gartner's survey shows 17% and McKinsey's business function-based scaling criteria show around 10%—differing by institution—the degree to which both metrics rise during the second half of the year can serve as a common yardstick to gauge the speed of next-generation AX competition. Second is the industrial gap. While technology and financial sectors lead, the catch-up speed of manufacturing and energy sectors can serve as a reference point for determining the response timing of domestic flagship industries. Third is the standardization of ROI measurement frameworks. Major consulting firms universally advise that only companies tracking changes in cycle times, decision-making structures, and output quality—rather than adoption rates—can prove the sustainability of AX investments. 'What was redesigned,' rather than 'how much was adopted,' is becoming the standard for evaluating global AX in the second half of 2026. Now that the race for enterprise-wide adoption rates has effectively concluded, the target for domestic companies to benchmark is not the tool list of advanced enterprises, but the execution pathway itself of redesigning a single core task from end to end.