Groundhog

Inspiration

Every growing team has experienced the same problem: a discussion begins, people passionately debate a topic, and after thirty minutes someone inevitably asks, "Didn't we already decide this?" The answer is usually yes—but the original discussion is buried somewhere in Slack, lost among thousands of messages.

As organizations grow, institutional knowledge becomes fragmented. New employees don't know the history behind important decisions, experienced employees can't remember where they happened, and teams waste valuable time re-litigating problems that have already been solved.

We wanted to build an AI agent that doesn't wait to be asked. Instead, Groundhog proactively recognizes when history is repeating itself and steps into the conversation before another meeting, thread, or debate is wasted.

What it does

Groundhog is an AI-powered institutional memory agent for Slack that detects decision déjà vu.

When someone starts discussing a topic that has already been debated or decided, Groundhog automatically:

Detects semantically similar discussions from the past.
Searches connected Slack knowledge sources for the most relevant previous decision.
Explains why it believes the conversations are related.
Surfaces the previous decision, decision owner, reasoning, and original discussion.
Estimates the organizational cost of repeating the conversation.
Recommends whether the team should continue with the existing decision or formally review it if circumstances have changed.

Unlike traditional search tools or AI assistants, Groundhog doesn't wait for someone to ask a question—it proactively joins the conversation the moment institutional knowledge becomes valuable.

How we built it

Groundhog was built entirely inside Slack using its native platform capabilities.

We combined:

Slack Workflow Builder
AI Generate workflow steps
Slack AI Knowledge Sources
Conditional workflow logic
Threaded AI responses
Slack Canvas for decision documentation
Workflow Buttons for human confirmation

The workflow monitors discussions, analyzes the triggering message, searches connected knowledge sources for semantically similar decisions, and only responds when it has sufficient confidence that the conversation is reopening a previously resolved topic.

To improve trust and transparency, every response explains why the match was made, cites the previous decision whenever possible, and clearly distinguishes between retrieved information and AI-generated estimates.

Challenges we ran into

The biggest challenge was balancing helpfulness with accuracy.

A proactive AI agent must avoid interrupting conversations unnecessarily, so we spent significant time refining prompts to reduce false positives. Groundhog is designed to return NO_MATCH instead of forcing a connection when it isn't confident.

Another challenge was working entirely within Slack's no-code environment. Without custom infrastructure or external services, we had to creatively use Workflow Builder, AI Knowledge Sources, Canvas, and prompt engineering to deliver an experience that feels intelligent while remaining completely native to Slack.

Finally, we wanted Groundhog to be transparent. Rather than acting like a black box, it explains why it matched a previous decision and provides references so users can verify the information themselves.

Accomplishments that we're proud of

Building a proactive AI agent that interrupts repetitive discussions before time is wasted.
Creating an institutional memory system using only Slack's native capabilities.
Designing Groundhog to explain its reasoning instead of simply providing answers.
Developing a semantic decision-matching workflow instead of relying on keyword searches.
Delivering a polished, enterprise-focused user experience that feels like a natural extension of Slack.
Demonstrating how AI can preserve organizational knowledge without replacing human decision-making.

What we learned

Building Groundhog reinforced an important lesson: the biggest productivity problems aren't always about creating new knowledge—they're about helping teams remember the knowledge they already have.

We also learned that proactive AI experiences feel fundamentally different from reactive assistants. Instead of waiting for users to search, Groundhog appears exactly when it's needed, helping teams stay focused and move faster.

Most importantly, we learned that explainability builds trust. Users are much more willing to rely on AI when it clearly communicates why it reached a conclusion rather than simply presenting an answer.

What's next for Groundhog Decision Agent

Groundhog is designed to evolve from an AI decision assistant into a complete organizational memory platform.

Our roadmap includes:

Workspace-wide semantic search across organizational knowledge.
Decision health and lifecycle monitoring.
Organization-wide analytics showing time and money saved.
Automatic decision categorization and tagging.
Enterprise integrations with platforms like Salesforce.
Workspace-level customization for AI behavior and cost estimation.
Rich dashboards that measure decision quality and organizational knowledge over time.

Our long-term vision is simple:

«Never debate the same decision twice.»