← BASELOAD English

Enterprise AI Agents — 62% Try, Only 23% Scale

July 1, 2026 · 20 min read 한국어 원문

Enterprise AI agent adoption is the hottest keyword of 2026, but the reality boils down to one line: “adopting is easy, internalizing is hard.” According to McKinsey, 62% of companies are experimenting with or expanding AI agents, yet only 23% have actually scaled them company-wide. If generative AI was a tool that “answers when you ask,” an agent is closer to an employee that “works when you delegate.” But just as hiring an employee doesn’t automatically transform an organization, an agent doesn’t deliver results at the push of a button. This piece examines that gap through numbers and cases.

Photo by Igor Omilaev on Unsplash

The Numbers: Market Booming, Adoption Lagging

First, the market itself is exploding. Combining industry forecasts, the enterprise AI agent software market is estimated to grow from roughly $1.5 billion in 2025 to $41.8 billion in 2030 — about 28x in five years (roughly 175% CAGR). Investment and launches are accelerating too, with a steady stream of agents specialized for specific functions like recruiting, customer support, and supply chain.

The problem is the gap between “adoption” and “results.” The McKinsey figures above — 62% trying, 23% scaling company-wide — are the crux. Most companies get as far as a pilot, but only a third of them actually embed it into real operations. The rest stall at “we tried it, but we’re not sure where to use it.” Not because the technology is lacking, but because the way work is delegated hasn’t been sorted out.

Korea’s Real Bottleneck: High Adoption, Low Internalization

Korea is aggressive about “trying” adoption. Domestic surveys put the generative-AI adoption rate at 61%. But the “internalization” rate — actually woven into how the organization works — is only 6.7%, and company-wide internalization of AI agents drops to 3.7%. That’s roughly a 10x gap between adoption and internalization.

The implication is clear. Many companies say “we’ve adopted AI,” but in reality it’s just some employees using a chatbot personally. Very few have reached the stage where an agent enters a department’s standard workflow and partly replaces or assists human judgment. Ultimately the issue isn’t tool performance — it’s the organizational design question of “what, and how far, to delegate.”

Photo by Jo Lin on Unsplash

What the Winners Did Differently: They Started From the Front Line

What the successful cases have in common is that they started not from a grand company-wide strategy but from a concrete front-line problem. According to reports, delivery platform Yogiyo (Wysiwyg Studios) built its own agents through an internal AI hackathon — a menu-improvement recommendation agent based on trade-area and review analysis, an agent for auto-classifying inquiries and optimizing delivery zones, and more. Hankook & Company likewise used an AI hackathon to surface practical tasks like supply-chain risk management and tire product strategy.

The key to both cases is that the front-line staff themselves defined “where to use the agent” — not an outside consultancy. Because the person who does the work every day pointed to “this repetitive task I want to hand off,” the resulting agent was used immediately. The difference between companies where adoption stalls and companies where it spreads lies exactly in this starting point.

So What: How Solo Businesses and SMBs Should Start

You may envy the big-company hackathons, but smaller organizations can actually internalize agents faster — decision layers are shorter and the tasks to delegate are clearer. It comes down to three things:

• Start with repetitive, rule-based work: Hand off the tasks you repeat every day or week first (compiling reports, sorting email, schedule reminders, data cleanup). The less judgment involved, the faster it sticks.
• Define the scope and accountability boundary first: Draw the line on whether the agent only “proposes” or also “executes.” Hard-to-reverse actions like payments or external sends should require human approval.
• Grow one at a time, based on results: Don’t aim for company-wide automation from the start. Confirm the time saved on one task, then expand to the next.

Closing

An AI agent isn’t “installing one more tool” — it’s redesigning “how work is divided.” The market may grow 28x in five years, but the fruit goes not to the companies that bought the technology, but to the companies that changed how they delegate. In this gap where 62% try and only 23% scale, which side you end up on is ultimately an organizational choice.

Figures in this article (market size, adoption and internalization rates, etc.) are estimates based on McKinsey analysis and domestic/international research and press reports, and may vary by source and timing. Cases are based on public reporting.