SK's AI Agent Push
SK Group Chairman Chey Tae-won Demands Urgent Companywide AI Agent Transformation for Enhanced Performance
Anonymous
Key Points
1
One explainer stresses that—beyond the goal/think/act/observe loop—the term “agent” is often oversold, and that “frameworks are just plumbing for running that loop at scale,” not a prerequisite to understand the concept. It also reiterates the practical requirement that “agents that don't know when they're done spin in circles.”
2
A separate technical overview highlights agent memory as a core capability split into “Short-Term Memory” (preferences and conversation context) and “Long-Term Memory” (historical data, preferences, and recurring workflows), and notes that agents can become more powerful through planning and even “multiple agents can collaborate.”
3
OpenAGI’s Lux is positioned with a specific benchmark result and comparative framing: it “achiev[es] 83.6 on Online-Mind2Web benchmark,” which the article says “outperform[s] OpenAI Operator, Anthropic Claude, and Google’s models,” and it quotes observers calling it “the first computer-use model that actually feels reliable for real production workflows.”
4
SK Group’s chairman linkages agent adoption to employee transformation data and rollout strategy: he said “More than 90 percent of our employees are already using AI,” but SK needs “AI that goes beyond individual use and genuinely helps the team by turning those efforts into organizationwide performance.” The company also discussed a “once-in-a-generation” urgency and the need to “launch AI transformation initiatives quickly on a small scale before eventually expanding them,” alongside a broader “full stack of capabilities” (including “memory chips… data center infrastructure… energy and electrification”).
5
A commentary on the human side of the shift frames AI literacy as an advantage not in “people competing against AI,” but in learning “how to use AI effectively,” comparing it to how computers became an essential skill; it argues “the biggest question is no longer ‘How will people and organisations adapt around it?’” and that “The people who learn early will have the advantage.”
SK Group Chairman Chey Tae-won ordered a full-scale AI transformation on June 14, issuing a "one agent per person" mandate across the entire conglomerate. Speaking at the 2026 New Icheon Forum, Chey warned that the window to act is narrow, calling this a "once-in-a-generation" opportunity that SK cannot afford to miss. The Korea Herald reported that more than 90 percent of SK's employees already use AI individually — but Chey said that is not enough.
The chairman's push comes as agentic AI — systems that can set goals, plan steps, take actions, and learn from results — moves from research labs into real business operations. SK is not just a user of this technology. Through SK Hynix, SK Broadband, and SK Innovation, the group controls a "full stack" that spans memory chips, data centers, and energy infrastructure, according to Let's Data Science.
From "My AI" to "Our AI" Across SK Group
Chey drew a sharp line between individual AI use and organizational performance. "More than 90 percent of our employees are already using AI," he said, but SK needs "AI that goes beyond individual use and genuinely helps the team by turning those efforts into organizationwide performance." The Korea Herald noted that Chey plans to deploy dozens of his own AI avatars to communicate with employees across SK's many affiliates — a move designed to cut through layers of corporate hierarchy.
SK's rollout strategy is deliberately staged. Let's Data Science reported that Chey urged the group to "launch AI transformation initiatives quickly on a small scale before eventually expanding them." SK Broadband is already ahead of the curve — it launched an "AI Agent Lab" in March 2026 to train 400 employees, or 20 percent of its workforce, to build role-specific AI agents.
What an AI Agent Actually Does — and Why It Matters
An AI agent is not a chatbot. A chatbot waits for a question and answers it. An agent gets a goal, thinks about how to reach it, takes an action, checks the result, and repeats. Medium explained this as the core cycle: Goal → Think → Act → Observe → Repeat. Crucially, agents that lack clear stopping conditions "spin in circles" — a practical failure that makes reliable tools and memory essential, not optional.
Memory is what separates a useful agent from a frustrating one. Short-term memory holds the current conversation and immediate context. Long-term memory stores historical data, past preferences, and recurring workflows. Medium noted that agents with both types of memory grow more powerful over time — and that multiple agents can even collaborate, dividing complex tasks between them.
OpenAGI's Lux Outpaces OpenAI and Google on Reliability
One concrete benchmark shows how far computer-use agents have come. OpenAGI's Lux model scored 83.6 on the Online-Mind2Web benchmark — beating OpenAI Operator at 61.3 and Google Gemini CUA at 69.0, according to Quasa. Lux executes actions in one second, compared to three seconds for OpenAI's Operator. Observers have called it "the first computer-use model that actually feels reliable for real production workflows."
Lux works differently from standard AI models. Rather than processing text, it reads screenshots and decides which clicks, keystrokes, or commands to make — operating any desktop app the way a human would. OpenAGI founder Zengyi Qin argues that a model trained to write a poem is fundamentally different from one trained to manage a supply chain. Running Lux via API costs roughly one-tenth of standard LLM-based agent frameworks, Quasa reported.
AI Literacy Is the New Competitive Edge for Workers
The biggest shift may be human, not technical. Medium argued that the real advantage is no longer about "people competing against AI" — it is about learning "how to use AI effectively." The comparison drawn is to the personal computer: once a niche skill, now a basic job requirement. People who learn to direct agents through multi-step tasks early will have a structural head start over those who wait.
The stakes are measurable. SK Group has committed 82 trillion won — roughly $60 billion — to AI, semiconductors, and data center infrastructure through 2030. The EU's AI Act now makes AI literacy a legal obligation under Article 4, meaning that working with agents is becoming a standard of professional due diligence, not just a competitive bonus. For SK and for workers everywhere, the time to start is now, not later.
Sentiment
Mostly PositiveNEGATIVENEUTRALPOSITIVE
Publishers
10
Articles
6
Reach
16
Topics
A
AnonymousAuthor
