[Digital Today reporter Chi-gyu Hwang] “AI adoption is a technology-related task. But embedding it is a management task.”

Oh Sarang (오사랑), AI/AX team leader at Hanwha Solutions, raised the issue on Thursday in a keynote speech at the AWS AI Wave 2026 Conference hosted by Amazon Web Services (AWS) Korea, as a condition for establishing AI transformation (AX) at an organisational level beyond individual use.

According to Oh, Hanwha Solutions, which focuses on energy and materials businesses including solar power and petrochemicals, is concentrating on AI for organisation-wide AX beyond individual use. Its goal is to embed AI into company-wide work processes.

At an early stage, it focused on automating repetitive tasks and improving individual productivity. It is now working to link AI to organisational goals and solve bigger, harder problems across its core business and value chain. As part of that, it runs AI budgets by team rather than by individual.

In this regard, Oh stressed that deployment does not mean embedding.

“Giving tools to dozens of developers and doing AX at an organisational level were completely different projects,” Oh said. “Work environments differ by design team, sales team and subsidiary. And most employees are not developers,” Oh said. “Just distributing good tools was not enough,” she added. “Infrastructure, operating systems and an enablement system that raises employees’ AI capabilities need to go together.”

If employees use external AI services through personal accounts, the pace of adoption can speed up. But there is a risk that company data could leave without controls. There is also no record of who uses what and how much.

With that in mind, Hanwha Solutions set up an environment so that all AI calls go through an LLM gateway within its AWS account and then to Amazon Bedrock. Logs and usage data also remain in the account.

Amazon Bedrock is an AI service platform provided by AWS. It supports using AI models from multiple companies, including Anthropic, OpenAI, Meta and Mistral, through a single API. Companies can build AI applications and agents while protecting their data within the AWS environment.

“At the gateway, we see which organisation uses which model and how much,” Oh said. “In an integrated management console, we track usage by subsidiary. Even if an employee enters a name or phone number, Bedrock guardrails mask that information and pass it to the model,” she said.

Hanwha Solutions began reviewing Claude as soon as a research preview allowing its use on Amazon Bedrock opened in April, and built an operating environment in three weeks. By the time of the formal launch on July 9, employees who took part in the first proof of concept were already using it in real work. It became a best practice within the group, and the number of group affiliates pushing to adopt Claude rose to 7.

“Thanks to the LLM gateway, a small team was able to expand the deployment scope to global sites, including South Korea and the United States, without building separate heavy infrastructure,” Oh said.

Hanwha Solutions overhauled the LLM gateway and deployment system on the infrastructure side. On the operations side, it focused on bringing cost control, project lifecycle, performance management and AI asset sharing into one. “Infrastructure, operating systems and an enablement system that raises employees’ AI capabilities need to go together,” Oh reiterated.

Hanwha Solutions aims to intensively identify projects that will increase business value. Next year it aims to make performance more concrete by embedding AI agents linked to team key performance indicators into work.

Governance also plays an important role in accelerating AX at a company-wide level. Oh said organisational AX can gain momentum only when governance supports it. “We started earlier than others, but what matters is embedding,” Oh said. “Because we established governance first, we were able to use it in real work even before the formal launch, and for the same reason we are now bringing AI into core businesses. Governance does not slow AI down. Rather, it is the decisive force that embeds AI into core businesses,” she said.