Last July (2024), Bloomberg obtained data from OpenAI's internal meeting and reported on their AGI roadmap. It was a five-stage roadmap of chatbot → reasoning → agent → innovation → organization, and since I vaguely know how top tech companies turn their roadmaps into reality, I thought I should bet on this roadmap.

So, for the past year, I have always looked at this diagram and measured my timing. ‘When will AI-native organizations emerge and there will be no going back after that?’ At the time, the technical question was how many r's there are in strawberry, but in just over a year, we are witnessing that we have gone beyond chatbots and reasoning and entered the era of agents and scientific innovation.

Meanwhile, some of my doubts were also answered. For example, questions such as ‘Why does AI have to go beyond the 4th stage of scientific invention and discovery to go to the 5th stage of overall organization operation?’ Currently, automating the way organizations work through workflows is already possible at the third level, the agent level. In order to become an AI-native organization, innovative inventions must be a priority or a part of the way we currently perform and make money. We understand that simple workflow automation alone cannot become an AI-native organization.

This time, Google announced a lineup based on the performance of the new Gemini 3 Pro model, NanoBanana Pro, Antigravity, etc. And most of the tasks that knowledge workers have been doing in practice for the past few decades are now showing that AI can do them better or faster.

To determine this, the test I personally conducted every time a new Frontier model was announced is as follows.

1. Long Black Test - There is an online outlet called Long Black and I like their writing. Whether LLM can imitate long black writing has always been a test of interest.
2. App UI/UX test - The web landing page was well created, but the user experience was not achieved with the satisfactory UI design and flow of the iOS mobile app. Now this happens.
3. Advertising image testing - Now that not only good product development but also content marketing is essential, creating advertising images without discomfort is also critical to organizational automation. NanoBanana Pro excels in this area as well.
4. IR test - I ran NotebookLM with the data presented in front of investors and it was better than me.

Previously, coding was at a level that could compete with the results of human practitioners, but now, beyond product development, many (or most) of the things required to run a small company are being singled out to a level where it is difficult for the public to distinguish between human practitioners and the results, and I had no choice but to admit that the long-awaited and feared moment has come.

The work people will do at work will be narrowed down to two things for the time being.

1. Evaluate and be evaluated - You must be able to deliver the desired results at a high level by evaluating the model and organization. Additionally, you must be able to quickly receive feedback from evaluators such as customers/users/supervisors. (Because humans are still the point of contact between the results and the world)
2. Take responsibility - models can do everything. Except taking responsibility. (In the future, this will also be solved through a system called insurance, but for now)

Other than that, I think it would be a good idea to intentionally delegate it to a model or try automating it with an agent first.

What I want to say is, if you want to become an AI-native organization and a one-person (tiny team) unicorn, wouldn't it be a good time to start a business now? There is no set answer to business, and what is more important is the problem that the entrepreneur wants to solve by investing a considerable amount of time and effort with sufficient energy and will, but in terms of timing. There is also a sense of crisis that if we do not do it now, we may face greater uncertainty.

All knowledge workers will become self-employed, and augmenting their capabilities with their own AI agents will be recognized as their capabilities. Money will be needed to hire those agents, and cash flow management is essential, so it is inevitable that highly developed knowledge workers will become indistinguishable from businessmen. Before going in this direction, I wanted to let you know that there is a lifestyle that allows me to prove myself and move forward through the pain I choose, not through the pain of starting a business.