Turning LINE Conversations into AI-Powered Workflows

A session at LINE THAILAND DEVELOPER CONFERENCE 2026

Transcript

1. Turning LINE Conversations into AI-Powered Workflows Rachata Pratheep Na Thalang & Sittipon Kumda LINE Dev Conference 2026
2. Our starting point: Findee First prize LINE Hackathon last year LINE + MINI App AI reads and classifies files Save, find and share in LINE 2
3. From Findee to a Bigger Challenge We joined the pilot project BumbleB to tackle the broader challenges of group conversations. The project is still experimental. 3
4. Information overload Which plan? Which file? Who owns the next step? The answer is scattered across messages. 4
5. Ingest the information Files Messages URL AI Which information still applies? 5
6. Catch up What you Miss Time-scoped summaries Key group updates Summary to task draft 6
7. Never run out off ideas Ask for ideas Create & assign tasks Confirm attachments 7
8. Always know what important My tasks at a glance Overdue & upcoming Suggested priorities 8
9. Organized all the mess Task & file hub Related tasks File version history 9
10. Working in the group chat • Ask in the group's context • Reply to a starting message • View tasks by time period Product guide illustration 10
11. Private chat through the OA • Specify or select a group • Choose the time period • Receive the summary privately Product guide illustration 11
12. A web UI inside LINE • Rich Menu opens the web UI • LINE Login • No separate app installation Product guide illustration 12
13. Behind the Scenes How BumbleB turns conversations into actions 13
14. How BumbleB handles a request 14
15. Sender and time in context Sender USR_01 Time 14:20 Message Sawadee Who said it, and when? 15
16. Flex and other types Sender USR_01 Time 14:20 Type flex Content json of that flex 16
17. Stable reference IDs Illustrative example: the same data keeps the same ID Sender name USR_01 URL URL_07 Masking organizes context. Other security controls are still needed. 17
18. Caching repeated context Repeated context and a changing request Stable context Reused through cache New request Context for this task Separate what repeats from what changes. 18
19. Compaction and chat history Compaction and historical retrieval Long conversation Compacted context Retrieve past details when needed Summaries can lose details. Retrieve history when the system can access it. 19
20. Plugins, Skills and Tools Skills Guidebooks Plugin Tools Actions 20
21. Skills as guidebooks Skill: task-edit Short description Which tool to use How to 21
22. Tools as actions Illustrative action Reviewed input Tool Edit a task Action result A tool acts on its input and returns the result. 22
23. A short skill catalog Chat summaries Summarize by topic and time Task management Create and manage task lists 23
24. Reading skills on demand Read skill details for the selected task Read the catalog Select the skill Read its guidebook Load the details when they are needed. 24
25. Tools for the selected skill Limit the exposed tools to the selected skill Selected skill Backend Relevant tools The team experienced fewer wrong tool choices. Tool filtering still needs separate permission checks. 25
26. Never trust, always verify 26
27. Review before creating a task • Task name and details • Assignees and due date • Confirm before creating Product guide illustration 27
28. Overwrite or a new version? Overwrite New version Change the original file Keep a separate copy Review attachments and the choice before acting. 28
29. Quick Reply and Postback When the button uses a postback action Quick Reply button Postback action Webhook postback event The button sends action data to the backend. 29
30. What we learned Context needs provenance and time. Skills define the procedure. Users review important actions. 30
31. See you next week Follow LINE Thailand OA @linethailand 31