Ledge.ai held its sixth AI study event online on September 29 and 30, 2026, under the theme “Which jobs can be entrusted to AI now, and which cannot—and why?” The eight sessions featured practitioners working in animation production, demand forecasting, research, data strategy for AI agents, company-wide AI adoption, and skills transfer on factory floors. Speakers repeatedly discussed not only AI performance, but also how it is used in work.

After an opening by Ledge Inc., megli discussed using AI in the process of completing animated works on September 29, the first day. Sniffout presented data strategy, and AI CROSS explained how demand forecasting changes ordering work. On September 30, the second day, Kaguya addressed AI literacy for research by specialists; APOLLO11 discussed database strategy for the AI agent era; Sourcenext shared 180 days of expanded AI use; and Ollo showed how video can make experienced workers’ skills visible. The sessions highlighted the difference between generating footage and completing an animated work, and the challenge of incorporating forecasts into ordering. Research raised questions about information reliability, while agent data strategy raised issues of information management and safety.

Tomoaki Kojima of Sourcenext said the company gave every employee a paid AI account, made workplace AI use a performance-review goal, and held company-wide training in vibe coding, or developing apps with AI. As adoption spread, the company found that efficiency gains in one department did not always speed up the organization as a whole, because work could wait for checks or tasks in another department. Sourcenext has begun reviewing how responsibilities are divided between departments. Kojima said using AI merely to prepare documents does not by itself amount to higher productivity; companies need to ask whether it contributes to sales or business performance.

Kai Tsumoto of Sniffout said company documents are often “human-ready,” or formatted for people to read. Complex Excel tables, PowerPoint diagrams, and manuals built around screenshots can cause AI to miss or misread information. He listed three requirements for data used by AI agents: enough information for the work, a format the AI can understand without overlooking details, and the ability for the AI to find needed information on its own. He said the work process that produces an output matters more than simply having large amounts of historical data. Companies can also record experienced workers explaining tasks to newcomers or demonstrating computer operations, then use AI to organize the steps. In a question-and-answer session, he said a “knowledge gap”—when necessary information or instructions are not fully provided—can hinder AI’s work.

Kengo Yoshida of Ollo described a system that analyzes factory-work videos with AI, identifies work steps, and supports manual creation. Comparing footage of experienced workers and newcomers can reveal steps that take longer and opportunities for improvement. The effort turns work once dependent on experienced workers’ know-how and intuition into data that can be compared and analyzed. Asked whether advances in humanoid robots could make video manuals for people unnecessary, Yoshida said work data may also help transfer skills from people to robots in the future.

The event did not set a universal rule for dividing work into tasks that can or cannot be delegated to AI. Company examples instead showed that the scope depends on each organization’s preparation and workflow design. Speakers identified defining the tasks and conditions for AI, the results expected, and how those results connect to value—and reviewing how work is done—as challenges for companies.

Ledge.ai held its sixth AI study event online on September 29 and 30, 2026, under the theme “Which jobs can be entrusted to AI now, and which cannot—and why?” The eight sessions featured practitioners working in animation production, demand forecasting, research, data strategy for AI agents, company-wide AI adoption, and skills transfer on factory floors. Speakers repeatedly discussed not only AI performance, but also how it is used in work.

After an opening by Ledge Inc., megli discussed using AI in the process of completing animated works on September 29, the first day. Sniffout presented data strategy, and AI CROSS explained how demand forecasting changes ordering work. On September 30, the second day, Kaguya addressed AI literacy for research by specialists; APOLLO11 discussed database strategy for the AI agent era; Sourcenext shared 180 days of expanded AI use; and Ollo showed how video can make experienced workers’ skills visible. The sessions highlighted the difference between generating footage and completing an animated work, and the challenge of incorporating forecasts into ordering. Research raised questions about information reliability, while agent data strategy raised issues of information management and safety.

Tomoaki Kojima of Sourcenext said the company gave every employee a paid AI account, made workplace AI use a performance-review goal, and held company-wide training in vibe coding, or developing apps with AI. As adoption spread, the company found that efficiency gains in one department did not always speed up the organization as a whole, because work could wait for checks or tasks in another department. Sourcenext has begun reviewing how responsibilities are divided between departments. Kojima said using AI merely to prepare documents does not by itself amount to higher productivity; companies need to ask whether it contributes to sales or business performance.

Kai Tsumoto of Sniffout said company documents are often “human-ready,” or formatted for people to read. Complex Excel tables, PowerPoint diagrams, and manuals built around screenshots can cause AI to miss or misread information. He listed three requirements for data used by AI agents: enough information for the work, a format the AI can understand without overlooking details, and the ability for the AI to find needed information on its own. He said the work process that produces an output matters more than simply having large amounts of historical data. Companies can also record experienced workers explaining tasks to newcomers or demonstrating computer operations, then use AI to organize the steps. In a question-and-answer session, he said a “knowledge gap”—when necessary information or instructions are not fully provided—can hinder AI’s work.

Kengo Yoshida of Ollo described a system that analyzes factory-work videos with AI, identifies work steps, and supports manual creation. Comparing footage of experienced workers and newcomers can reveal steps that take longer and opportunities for improvement. The effort turns work once dependent on experienced workers’ know-how and intuition into data that can be compared and analyzed. Asked whether advances in humanoid robots could make video manuals for people unnecessary, Yoshida said work data may also help transfer skills from people to robots in the future.

The event did not set a universal rule for dividing work into tasks that can or cannot be delegated to AI. Company examples instead showed that the scope depends on each organization’s preparation and workflow design. Speakers identified defining the tasks and conditions for AI, the results expected, and how those results connect to value—and reviewing how work is done—as challenges for companies.