Various companies are adopting AI in different ways, but the level of use varies widely. Many still use AI only for simple purposes, while some use it to change processes.

Against this backdrop, Floodgate co-founding partner Ann Miura-Ko (앤 미우라-코) has drawn attention by classifying corporate AI adoption into six stages, borrowing from the way self-driving car capability is broken down to Level 5.

After visiting AI startups in recent weeks, as well as a company with about 1,500 employees such as Ramp, she said she came to rethink what it means to be AI native. That led to the result of breaking AI levels down into six stages.

"Cruise control is not self-driving, and lane keeping is not self-driving," she said. "At a small startup, if founders are deeply immersed in AI, the whole company can look AI native. But as you scale, the bar changes. AI has to be embedded in how the entire organisation works, not just a few founders' personal capabilities," she said.

A company that uses ChatGPT for meeting summaries and a company where agents directly query systems and run workflows can both say they use AI, but their operating levels are completely different. In this context, the author stressed that the right question is not whether AI has been adopted but what level of autonomy an organisation has achieved. As criteria for judging that, the author presented what AI can see, what AI can do, who can scale the system, and whether the organisation has actually changed. The six stages of corporate AI adoption using these criteria are summarised as follows.

Level 0 is show AI, or AI as theater. At this stage AI cannot see structured data and cannot do work that has a meaningful impact on outcomes.

If a CEO is calling for AI transformation but still runs the company the same way as in 2023, that is Level 0. Ann Miura-Ko said, "A declaration is not adoption."

Level 1 is personal productivity. Individuals use AI to write drafts or summaries, but it does not actually affect systems. If the employee who uses it best leaves, that workflow disappears as well. Miura-Ko said a statistic such as "80 percent of employees use AI weekly" falls into this stage and argued it means nothing by itself.

Level 2 is team workflows. AI workflows emerge at the team level, and people can handle work faster and more efficiently within their roles. But Miura-Ko's view is that if each department is building its own AI stack, it is only AI-enhanced silos and cannot be seen as an AI-native organisation.

Level 3 is organisational infrastructure. AI can query the entire organisation and takes real actions such as CRM updates, PR generation and invoice reconciliation. Non-engineers also build workflows and share them with other teams. If the organisational chart has visibly changed from 2023, that is a Level 3 signal.

Level 4 is a compounding operating system, where the system improves the more it is used. It is a structure in which the system learns from prior runs and improves on its own.

This includes non-engineers directly building production internal tools, and agents moving autonomously with policy-based decision authority. Having a lot of automation does not make it Level 4. Even if there are hundreds of separately built automations, if they are not connected and are not managed, it cannot be seen as Level 4.

Level 5 is effectively a self-driving organisation. Miura-Ko, while noting it is a stage that does not yet exist, defined Level 5 as one where the core operating loop senses reality, diagnoses problems, initiates tasks, executes within delegated authority, updates shared memory and improves future actions on its own. Humans take roles related to strategy, taste, risk, value judgments and exception handling. "Just as a startup is not a scaled-down version of a large company, an AI company is not a company that simply layers AI tools onto an existing business," she said. "It is an organisation rebuilt around a new operating model," she said.