Vice President Lee Hae-seok of Polaris Office stated accordingly at SMARTCLOUD SHOW 2026, held on the 26th at the Westin Josun Hotel in Sogong-dong, Seoul, under the theme "AI moves the world." Lee gave a talk that day on the theme "The end of SaaS? Now AI agents also work on the cloud."
Lee projected that SaaS will not disappear with advances in artificial intelligence (AI), but its role and business model will change for the AI era. Software that offers only simple functions can be replaced by AI, but SaaS that secures data accumulated by corporations and workflows can instead benefit from the spread of AI.
Recently, in the software industry, concerns have grown that Generative AI could replace existing SaaS. This is because by entering prompts into AI alone, one can implement functions previously performed by software, and as AI spreads and corporations reduce new hiring, the "per-seat billing" model of SaaS, which charges based on the number of users, could also be shaken.
Lee said, "Until now, the business model of SaaS corporations was per-seat billing," adding, "When the number of a corporation's employees increased, the billing structure increased accordingly, but with AI, fewer corporations say they will increase their headcount." He continued, "If companies are not increasing their headcount, doubts have arisen about whether SaaS corporations can continue their growth trajectory."
However, he saw that the current situation, 6 months to a year after the so-called "SaaS doomsday theory" was raised, is unfolding differently than expected. Lee said, "In reality, after 6 months and a year, the results of most software corporations have instead improved, and not only revenue but also operating profit has generally improved."
Lee analyzed that usage itself is increasing as AI agents use the internet and software in place of people. In the past, when a person searched directly, they looked at at most 10–20 websites, but if AI handles the search, it roams through dozens to hundreds of sites to collect information. Even if human software use declines, overall usage can instead increase as a new user called AI emerges.
He emphasized, however, that not all SaaS corporations will benefit in the AI era. Lee cited "data" and "workflows" as the core competitive strengths of software that AI cannot easily replace.
If a product only provides the function of writing documents, AI can create presentations (PPT) or edit content and replace existing software. By contrast, systems that have accumulated data such as users' past documents and edit histories, and the materials referenced in the process of creating documents, are difficult to replace because AI also has to use that data to perform work.
He also cited workflows, the business processes of corporations, as an important asset. Using customer relationship management (CRM) as an example, Lee said, "Going forward, creating the code called CRM or office could become something anyone can do," but added, "Building a good cloud office and CRM and using them to actually transform a corporation remains very difficult."
He projected that in the AI era, the principal users of SaaS will shift from people to AI agents. Instead of people directly logging into programs to write documents, submit electronic approvals, and enter data into enterprise resource planning (ERP) systems, AI agents will handle tasks by moving across multiple software programs through application programming interfaces (APIs) or the model context protocol (MCP).
Lee said, "The era when people directly go and use software such as document editors called office, electronic approval tools, ERP, and CRM will soon end," adding, "When all the tools used by a company are connected by MCP or API, agents will weave together all this software and work as if they were people."
When AI agents use SaaS directly, "governance" that controls access rights and scopes of work also becomes important. For example, electronic approvals can be handled with a single click technologically, but one must distinguish who can approve at what point in time and what data cannot be taken outside. AI must also be designed to perform tasks only within pre-approved permissions and data scopes.
He also predicted that AI's role will evolve from merely assisting individual tasks to carrying out an entire job end to end. Until now, Generative AI has been assigned single tasks such as "Create last quarter's results as a report" or summarize materials, but going forward, one could assign an entire business task such as "Register Company A as a new supplier and proceed from contract signing through delivery."
Polaris Office is also developing AI agents in line with this shift. Citing supplier registration as an example, Lee outlined a process in which an AI agent checks existing transaction records, margins, and deal terms in CRM and groupware, drafts a supply contract using an office program, and then proceeds with the approval process in an e-contract system. It is a structure that connects collaboration via email and messenger and legal review into a single workflow.
Lee said, "Previously, AI was a tool, a junior, or a supporter that helped me, but now it can evolve into a colleague who works with me, an agent that can take on a task entirely," adding, "Software that enables AI agents to work better will instead advance further and be used more."