Generative AI Conference 2026 Participant Report — Generative AI Speeds Up Work. But Is That Work Correct?

I am Nakamura from Agile Group.
I attended the Generative AI Conference 2026. The event, themed around the social implementation of generative AI, covered a wide range of topics—such as AI agents and physical AI—from the three perspectives of industry, development, and intelligence.

Since the start of 2026, the acceleration of coding by AI agents has rapidly become more than just a vision. I personally experimented with it at a proof-of-concept level while maintaining my own personal development app, and even that limited experience made phrases like “10× productivity” sound far less exaggerated than before. As an engineer, I’m genuinely excited.

On the other hand, in my role as an agile coach, I experienced two concerns.
One is that if we can generate deliverables in large volumes, might the downstream processes—from testing to release—become a bottleneck?
The other is that if manufacturing cost constraints become less strict, requirement definitions may become sloppy, thinking “let’s just build more and figure things out later.”

At this conference, I aimed to find out how these two concerns were appearing in actual domestic settings. After all, seeing is believing.

In conclusion, the structures I was worried about have already surfaced as challenges on the ground. However, initiatives to change how work is done based on these premises have also already started.

All sessions delivered a stark, robust substance, clearly distinct from the typical “shiny examples.” From these, I’ll highlight the sessions that resonated with each of the two concerns.

Mr. Sota Omura, Sakana AI Inc.

Centered on AI and science, the session introduced the acceleration of research activities by AI and the challenges that have emerged as a result. Among several challenges presented, the following keyword left the strongest impression:

"The start and the finish are human."

The “start” refers to “defining the problem space.”
AI’s performance as a scientist has continued to improve, and research loops have accelerated. On the other hand, there is a concern that AI intervention could narrow the very problem space that science addresses.

The “finish” refers to “understanding and interpretation.”
Even if part of the thinking process can be entrusted to AI, the human task of understanding the results and interpreting their meaning does not disappear. Human processing capacity may not keep up with the speed at which AI generates results. Grant proposal reviews and peer reviews of papers were cited as concrete examples.

I also strongly resonated with the statement, “Without understanding, knowledge and creativity are hindered. There is no feedback loop.” Applied to software development reviews and acceptance testing, it’s easy to imagine that the burden would concentrate on development leaders and customers.

Mr. Hiroki Kai, Andest HD Co., Ltd. / Mr. Keito Kamoi, Algomatic Co., Ltd.

This is a case study of a “Store Manager AI Agent” that supports the store manager operations of an apparel retail store. The system enables tasks such as sales analysis and inventory search through the AI agent, and it was reported that inventory search operations were reduced by 75%.

"First, something that the field can actually use."

This statement seems like one answer to the second concern. It’s not about building something just because manufacturing costs have decreased. Instead, you first define whose what you want to change, and then verify it while having people use it. The easier it becomes to build, the more critical the quality of problem setting and feedback becomes.

What surprised me was the execution capability to permeate AI into large enterprises. They started from a clear need—to transform the store manager’s operations—and took a small-scale start with a few stores. They improved based on feedback from the field, increasing the number of target stores and expanding the scope. A cross-functional team, including non-IT departments, sets the direction and leads with ownership.

Putting it in writing, there’s nothing novel about the use cases or the approach. It’s about doing the obvious things thoroughly. I took that consistent effort to be what led to the 75% figure.

To summarize in my own way, there are two questions.

Is the work right?
Have we correctly defined customer value and selected the problems we should truly be addressing?

Is the work being done correctly?
Rather than accelerating only parts of it, is the entire flow to deliver value running smoothly?

Generative AI will speed up work. That’s why I believe facing these age-old questions becomes even more important. As a company that has positioned requirements engineering as a pillar of value, I think now is our time to shine—while making a small victory pose, I’d like to take these two questions home as my own homework.