[Op-Ed] Beyond AI “Experiments” to “Real Results”: Three Questions Companies Should Ask
IT DAILY ·
✦ AI Summary
Cho Seong-hyeon, head of technology at Databricks Korea, said the focus of AI use should move from the experimental stage to tangible results.
The 2026 "Making AI Deliver" report by The Economist Enterprise and Databricks found that about 60% of companies were using autonomous systems in actual work, and about 90% of CTOs said adoption timelines and return on investment were ahead of plan.
He said safe governance, AI Agents naturally integrated into daily work, and employee training and capability building must go hand in hand.
Cho Seong-hyeon, head of technology at Databricks Korea, wrote that companies need to shift the focus of AI use from the experimental stage to tangible results. The photo is from Databricks. He said companies are making the potential business contributions of AI more concrete.
He said this trend also appeared in the 2026 "Making AI Deliver" report published by The Economist Enterprise and sponsored by Databricks. A survey of global large companies found that about 60% were using autonomous systems in actual work. In addition, about 90% of CTOs said the AI adoption timeline and return on investment were ahead of initial plans. The share of companies that redesigned job descriptions to reflect the expansion of AI use also came in at about 75%.
Cho said companies' challenge is to turn AI expectations into real results. To do that, he explained, they must go beyond simple adoption and build an environment in which employees can use AI naturally in their daily work. He added that the goal is to improve productivity and efficiency and to embed AI in actual operations. He also suggested that companies should continuously review three questions.
As AI Agents spread, the need for SQL expertise in deriving business insights is weakening. Accordingly, the extent to which technical expertise has served as a barrier to automation is also diminishing. Within companies, conditions are expanding so that a broad range of employees can use AI without advanced technical knowledge.
This shift is leading to AI use tailored to each employee's work and to the possibility of applying AI across business-critical areas. In particular, the foundation is widening for non-specialists to use AI in line with the context of their own work.
Musinsa built automated metric monitoring using Databricks Genie. Through this, frontline users can query data in natural language and check the metrics they need. This is presented as an example of linking data analysis more closely to everyday work flows.
However, expanding the scope of AI use requires a safe testing environment. To discover new use cases, understanding how to generate results using AI tools is also necessary. A gap still remains between expectations for AI and the environment that supports actual use.
Platforms with security and governance in place help employees safely experiment with a range of scenario-based AI Agents, playing a role in narrowing that gap. On the other hand, if safeguards are lacking, companies may restrict employees' AI use for security reasons. In that case, limits on employees' AI use could slow the pace of AI adoption and expansion and also constrain AI's business impact.
As AI use expands, the need to address gaps in management and oversight has also emerged. According to the Making AI Deliver survey, 44% of companies made official governance frameworks for autonomous systems mandatory across all relevant teams. Databricks' "2026 State of AI Agents" report also confirmed the importance of governance and evaluation.
According to the report, companies using evaluation tools had about 6 times as many AI projects moved into production as companies that did not use them. In companies that applied AI governance, the performance gap in the number of AI projects moved into production widened by more than 12 times. This shows that introducing evaluation tools and governance is linked to production conversion results.
To expand AI, the scope of use needs to grow, but simply increasing adoption is not enough. Common governance principles must be applied to all AI workloads, and the intensity of control and oversight must be adjusted according to work risk levels. When operations are run in this way, trust in AI use can be improved. A safe governance foundation must also be established as a condition for AI expansion.
This type of system is meaningful because it allows employees to access the AI functions they need while also expanding access to AI and preventing harm to business security. Employees can also develop new capabilities. However, if they have to use a separate application or move to another tab, unnecessary inconvenience arises, and this way of working can affect actual AI use.
AI Agents therefore need to be naturally integrated into the actual work environment. This should extend from office workers to frontline store employees. It should also make AI available whenever needed, regardless of the work environment.
If employees can access CRM, Google Docs, and key corporate data through a single chat interface, they can minimize movement between systems, secure real-time insights, and maintain work continuity. This approach connects multiple company data sets and systems into one, allowing users to get the information they need immediately without intermediate switching and continue follow-up work.
One use case presented is directly integrating AI Agents into an intelligence dashboard for the marketing team. The marketing team can use it in sequence: first understanding operational status, then conducting a deep dive into specific areas, and then carrying out necessary actions or exploring new opportunities. For example, after identifying the cause of a traffic surge, the questions can expand in the direction of exploring how to reproduce that phenomenon.
The range of applications targeted for this expansion of AI functions includes the many apps widely used by companies. However, to enable smooth access, secure access without repeated authentication is needed. For this, automated ID management, governance consistently applied to all interactions, and business logic consistently applied to all interactions are all required.
For employees to use AI properly, the limitations of company-provided AI tools must not be excessive. Excessive restrictions on internal AI tools can lead employees to bypass internal safeguards. In addition, bypassing internal safeguards and the spread of "Shadow IT" can increase the likelihood of new governance problems.
Employees' expectations for AI go beyond simply obtaining insights. They want the ability to take direct action based on insights. They also expect AI Agents to provide new perspectives on existing thinking, guide the next step of work, and act on the user's behalf when needed.
Ultimately, the goal is for interaction with AI systems to feel as natural as collaborating with a skilled colleague. Accordingly, the role of AI Agents is presented as going beyond answering questions. AI Agents are tasked with using a company's data more broadly to deliver context-appropriate, accurate insights, provide actionable information, and help prevent interruptions or delays in the flow from analysis to execution.
The direction of this development for AI Agents is to carry out actual work on behalf of the user. The name for this advanced form of AI is "AI worker."
As AI use expands, employees need the ability to understand how to use AI tools, the ability to critically verify results, and the ability to connect verification results to judgment and action. Accordingly, companies need to invest actively in employee training and also create an environment for AI autonomous execution. The condition for AI autonomous execution is carrying out the next step of work within permitted bounds. In addition, a balance between technical infrastructure and human capability is needed, and when that balance is achieved, AI value can spread beyond a few technical teams to the entire company.
However, in the Making AI Deliver survey, only 4% of respondents cited strengthening employees' AI capabilities as a major ongoing cost. This shows that even as the need to build employee capabilities grows, it is being treated as a low priority in actual spending decisions.
The factor that determines whether AI succeeds or fails is not the number of experiments, but the foundation for use. Readiness, accessibility, and capability are also not one-time check items, and as AI adoption expands across the company, they require ongoing review and refinement by management. Readiness, accessibility, and capability are essential conditions that management must continuously oversee.
The standard for companies leading the AI race is not the number of pilots. The condition for leadership lies in having governance, environment, and capability foundations that enable all employees to access AI when needed and support safe and effective use of AI. AI should not remain a tool for a few technical teams; it must take root in daily work and decision-making processes. When AI takes root in work and decision-making in this way, the many experiments conducted up to now are connected to sustainable business results.
Source: IT DAILY · Cho Seong-hyeon
Original: https://www.itdaily.kr/news/articleView.html?idxno=241532
References
This article was produced with the help of an automated content generation algorithm.
Source: IT DAILY
View originalThis article was summarized and organized by BizCrush based on the original article from IT DAILY. For exact quotations and full details, please refer to the original article.