Kakao outlined the development and plans for “AI for Everyone” on the 14th at its If Kakao 2026 event in Suji-gu, Yongin, Gyeonggi Province. Yoo Yong-ha, head of product and performance, said KakaoTalk would serve as the user touchpoint for an open AI ecosystem where outside companies and individual developers can register specialized agents and tools. Kakao plans to build a platform that connects participants’ specialized technologies instead of providing all chatbot technology itself.

Kakao will connect external agents and tools according to the context of users’ conversations. Tools will handle specific functions such as information searches and restaurant reservations, while companies with specialist knowledge will participate through agents. The initial group comprises 34 companies and institutions: 15 members of a Kakao-led consortium and 19 alliance members. Kakao will oversee user experience design, system integration and operations, request analysis, and the delivery of tasks and policies.

LG Uplus will manage GPU operations, including response latency and operating costs, and lead model placement and serving based on each task. LG AI Research will provide its EXAONE large-scale AI model, while Lunit will supply health-focused data. Kakao Enterprise and others will provide task tools, and FuriosaAI will conduct trials of domestically developed AI neural processing unit (NPU) chips. CrowdWorks will handle the data training and quality assessment needed to run models.

Kakao’s PlayMCP already lists more than 500 tools. Kakao plans to select tools that pass internal quality reviews and safety checks, then release them to the public in stages. It has set up a registration process for small startup teams and individual developers. Developers define a specialized task, then submit the required user inputs, permission requests, and the format of results to return when the task succeeds or fails.

To prevent agents and tools from different companies from conflicting, Kakao is preparing a shared communication contract that covers whether a function can be performed, permission checks, and task status. For example, the system must distinguish between a benefits search that finishes normally but finds no results and one that fails because of a server outage. If it cannot tell the cases apart, it may mistake an outage for “no benefits” and give users a wrong answer. For long-running search tasks, the runtime environment records progress so the system can check the previous result first when the same request comes in again.

When a step requires separate authorization, such as identity verification or payment, users will leave the Kakao chat and move to the participating company’s official service screen. The system will also explain why the user is being redirected and show the status of work done in Kakao. Kakao said shared standards for task delegation and result delivery would let participants improve their own expertise and service processes without being tied to Kakao’s platform structure.

Kakao plans to create a reward system so participants’ activity can lead to transactions and revenue, including reservations, product purchases, and subscriptions. It said even companies with low name recognition could reach customers when users discover needed information in a conversation and move to the company’s service to act on it, without knowing the company’s name in advance. External developers will be able to review total usage, task success rates, delays, and errors. Kakao plans to measure performance using detailed data and establish a model that distributes revenue in proportion to results.

Kakao will run a small beta test ahead of the planned official launch at the end of this year. Its target immediately after launch is 5 million monthly active users (MAU). It will track return visits, whether conversation context such as dates and group size carries over when an external agent changes, and the rate at which users complete real tasks such as reservations or payments.

Starting in 2027, Kakao plans a phased introduction of “one agent per person,” a personalized assistant. Users would set preferred roles and ways of working through conversation, and the agent would remember them to handle related tasks. Yoo gave the example of creating an assistant for health plans and checkup reservations, then asking, “Find a time again this time.” Based on saved preferences and past conversations, the agent would compare available reservation times and costs; Kakao said its goal is to reduce the inconvenience of having to explain the conditions from scratch each time.

Kakao outlined the development and plans for “AI for Everyone” on the 14th at its If Kakao 2026 event in Suji-gu, Yongin, Gyeonggi Province. Yoo Yong-ha, head of product and performance, said KakaoTalk would serve as the user touchpoint for an open AI ecosystem where outside companies and individual developers can register specialized agents and tools. Kakao plans to build a platform that connects participants’ specialized technologies instead of providing all chatbot technology itself.

Kakao will connect external agents and tools according to the context of users’ conversations. Tools will handle specific functions such as information searches and restaurant reservations, while companies with specialist knowledge will participate through agents. The initial group comprises 34 companies and institutions: 15 members of a Kakao-led consortium and 19 alliance members. Kakao will oversee user experience design, system integration and operations, request analysis, and the delivery of tasks and policies.

LG Uplus will manage GPU operations, including response latency and operating costs, and lead model placement and serving based on each task. LG AI Research will provide its EXAONE large-scale AI model, while Lunit will supply health-focused data. Kakao Enterprise and others will provide task tools, and FuriosaAI will conduct trials of domestically developed AI neural processing unit (NPU) chips. CrowdWorks will handle the data training and quality assessment needed to run models.

Kakao’s PlayMCP already lists more than 500 tools. Kakao plans to select tools that pass internal quality reviews and safety checks, then release them to the public in stages. It has set up a registration process for small startup teams and individual developers. Developers define a specialized task, then submit the required user inputs, permission requests, and the format of results to return when the task succeeds or fails.

To prevent agents and tools from different companies from conflicting, Kakao is preparing a shared communication contract that covers whether a function can be performed, permission checks, and task status. For example, the system must distinguish between a benefits search that finishes normally but finds no results and one that fails because of a server outage. If it cannot tell the cases apart, it may mistake an outage for “no benefits” and give users a wrong answer. For long-running search tasks, the runtime environment records progress so the system can check the previous result first when the same request comes in again.

When a step requires separate authorization, such as identity verification or payment, users will leave the Kakao chat and move to the participating company’s official service screen. The system will also explain why the user is being redirected and show the status of work done in Kakao. Kakao said shared standards for task delegation and result delivery would let participants improve their own expertise and service processes without being tied to Kakao’s platform structure.

Kakao plans to create a reward system so participants’ activity can lead to transactions and revenue, including reservations, product purchases, and subscriptions. It said even companies with low name recognition could reach customers when users discover needed information in a conversation and move to the company’s service to act on it, without knowing the company’s name in advance. External developers will be able to review total usage, task success rates, delays, and errors. Kakao plans to measure performance using detailed data and establish a model that distributes revenue in proportion to results.

Kakao will run a small beta test ahead of the planned official launch at the end of this year. Its target immediately after launch is 5 million monthly active users (MAU). It will track return visits, whether conversation context such as dates and group size carries over when an external agent changes, and the rate at which users complete real tasks such as reservations or payments.

Starting in 2027, Kakao plans a phased introduction of “one agent per person,” a personalized assistant. Users would set preferred roles and ways of working through conversation, and the agent would remember them to handle related tasks. Yoo gave the example of creating an assistant for health plans and checkup reservations, then asking, “Find a time again this time.” Based on saved preferences and past conversations, the agent would compare available reservation times and costs; Kakao said its goal is to reduce the inconvenience of having to explain the conditions from scratch each time.