Slackbot guided players through MrBeast’s complex puzzles.

A Trusted Puzzle Partner

To handle the volume and complexity of the puzzle challenge, Salesforce turned to Slackbot to create an interactive guide that brought the entire experience to life. Hidden clues scattered across the Super Bowl commercial, YouTube videos, and social channels set the hunt in motion — igniting a thriving community across Discord, Reddit, and social media where players debated, theorized, and collaborated. As discoveries unfolded, players used Slackbot to test ideas, refine theories, and make sense of it all — structuring their thinking, surfacing patterns, and keeping the investigation grounded in context.

“We turned ambitious game requirements into something truly agentic — an AI-powered gaming platform,” said William Dressler, Regional VP – AI and Data Architecture. “No one had ever built a gaming platform on Salesforce before.”

Trusted And Fair Gaming

At the heart of Slackbot was the Puzzle Vault — a Kanban-style board where players created and managed “guess cards” to visualize their theories. Slackbot used these tools to deliver a trusted and fair gaming experience by:

• Providing visibility for teams: Out-of-the-box dashboards in Agentforce Observability gave teams real-time insights into player interactions, helping them fine-tune the AI agent while preserving privacy and trust.
• Maintaining zero-leakage guidance: Slackbot never revealed puzzle solutions and always based its reasoning on players’ own submissions and guesses, ensuring that every discovery was earned by the participant.
• Measuring engagement through exploration: Metrics focused on guess cards created, conversation depth, and session patterns, which allowed players to experiment and iterate freely.
• Capturing every action: All pivots, hint requests, and puzzle switches were logged in the session tracing data model within Data 360, enabling comprehensive monitoring and debugging.

Conversations Powered By Context

Behind the scenes, Slackbot reasoned and took action through Salesforce’s LLM Gateway — a high-throughput layer that handled tens of thousands of LLM calls per minute while staying securely connected to Salesforce. The gateway efficiently routed requests, managed concurrency, and preserved conversational memory across sessions, so players could pick up right where they left off, even at peak traffic.

Before generating any response, Slackbot dynamically pulled in each player’s puzzle state, chat history, and guess cards, giving it the full picture of their current thinking. This allowed for highly specific, actionable guidance: if a guess card contained Morse code, Slackbot could act as a translator; if a theory was on the right track, it could nudge the player forward — without ever confirming or revealing the solution.

Semantic search against a vector database let Slackbot validate reasoning without holding the answer key, while a Redis cache ensured real-time progress at sub-second response times, even under heavy load. Multilayered protections kept the experience secure: a machine learning–based prompt injection detector blocked attempts to trick Slackbot, toxicity scanning ran on every input and output, and zero data retention by external LLM providers ensured that all player conversations remained private.