Rolling out knowledge agents: A step-by-step guide

Ready to get your team using knowledge agents? This guide walks you through everything from planning to launch—helping you avoid common pitfalls and set up agents that your team will actually use.

By the end, you'll have a clear rollout plan tailored to your team's needs.

👥

Access Required

You'll need to be a Guru workspace Admin or relevant custom role to create and configure knowledge agents.

❗

Before you start

This guide assumes you already have:

• At least 2-3 Collections with existing Cards OR
• Sources connected to tools where your knowledge lives (Slack, Google Drive, etc.)
• A basic understanding of what knowledge agents are

---

How you know your team needs knowledge agents

Knowledge agents work best when:

• You have a large team where the same questions get asked repeatedly
• Knowledge is scattered across multiple tools and Collections
• Different departments need access to different information
• You want to track and measure how your team finds answers

If you're not sure yet, start simple. Create one agent, share it with a small group, and see what happens. You can always come back and formalize your approach later.

Time commitment: Block out 2-3 hours for initial planning and setup. You'll spend about 30 minutes per week refining based on how your team uses the agent.

---

Step 1: Figure out what your team actually needs

Let's start by understanding what questions your team is asking.

Find patterns in what people search for

1. Go to Manage > AI Agent Center.
2. Look at the questions people have been asking (even if you haven't set up agents yet, you'll see search queries for the Guru Default Knowledge Agent).
3. Write down the 5-10 most common question types you see.

Example patterns you might notice:

• Sales keeps asking about pricing and competitive features
• Support keeps troubleshooting the same 3-4 issues
• New hires keep asking about internal processes

💡

Want to go deeper?

Use Research to analyze trends across all your sources and identify systematic gaps in your knowledge base.

Map those questions to where answers live

Now that you know what people are asking, figure out where those answers exist:

Write down which Collections and Sources contain the information your team needs most.

Identify who needs access

Think through:

• Which teams will benefit from this agent? (Sales? Support? Everyone?)
• What's their role? (Are they answering customer questions? Making decisions? Onboarding?)
• Where do they work? (Slack? Browser? Guru web app?)

This helps you configure access and deployment later.

---

Step 2: Get the right people involved

You can't do this alone. Identify 2-3 subject matter experts who:

• Deeply understand the topic area
• Have time to help (15-30 min/week)
• Are respected by the team who'll use this agent

These people will get Expert or Owner access to:

• Refine how the agent responds
• Review answers for accuracy
• Champion adoption within their teams
• Help you improve the agent over time

Pro tip: Pick people who are already answering these questions regularly. They'll immediately see the value in offloading repetitive answers to an agent.

---

Step 3: Create your agent

Now let's build the actual agent. Follow our guide on Creating your first knowledge agent to:

• Set up the agent with a clear, descriptive name
• Add the Collections and Sources you mapped in Step 1
• Assign access to the right Groups
• Add your subject matter experts as Experts or Owners
• Decide if you need a custom prompt (most agents work great without one)

Common mistake to avoid: Don't add every Collection and Source you have. More isn't better. Focus on sources that directly answer the questions from Step 1.

---

Step 4: Put your agent where people will actually find it

This is the most important step. An agent no one can find is useless.

Think about where your team works most:

If they spend time in the Guru web app:

• Create a Page for their team
• Set your agent as the default for that Page
• Now when they open that Page, your agent is right there

If they work in their browser:

• Pin the agent to the browser extension
• It'll appear at the top of their search results

If they live in Slack:

• Connect the agent to relevant Slack channels
• The agent will automatically respond when people ask questions in those channels

📘

Start with one placement

Don't try to deploy everywhere at once. Pick the single place where your team is most likely to encounter the agent. Once you see it's working, expand to other places.

---

Step 5: Tell your team about it

Don't assume people will just discover the agent. You need to actively promote it.

If the agent is in Slack:

Post in the connected Slack channel:

Hey team! 👋

I've connected our new Sales Product Expert agent to this channel. 

From now on, you can ask product questions right here and get instant answers from our product docs, competitive research, and help center.

Try asking: "How does our SSO implementation work?" or "What's our pricing for enterprise customers?"

The agent will respond automatically in thread. Questions? Ask me!

If the agent is on a Page or in the extension:

Send a Slack or email announcement:

Hey team! 👋

We just launched a new knowledge agent called Sales Product Expert to help you quickly find answers about product features, pricing, and competitive positioning.

You can find it on the Sales Team Page in Guru—it'll load automatically when you open that page.

Try asking it: "How does our SSO implementation work?" or "What's our pricing for enterprise customers?"

Questions? Ask me!

Share example questions people can try. This helps them understand what the agent is for.

Ask early adopters to share wins. When someone gets a great answer, encourage them to share it with the team.

---

Step 6: Watch what happens and adjust

Give it 1-2 weeks, then check how it's going.

Check your metrics

Go to Manage > AI Agent Center and look at:

• How many questions are being asked? (Is anyone using it?)
• Thumbs up/down ratio - Are answers helpful?
• Which sources are being referenced most?
• Questions without good answers - What's missing?

Make adjustments based on what you see

Get feedback from your experts

Your subject matter experts should check in weekly (just 15-30 minutes) to:

• Review agent responses
• Identify gaps in content
• Suggest prompt improvements
• Update outdated Cards

💡

Use agent verification to stay current

Enable agents to verify Cards so they can flag outdated information. Your experts can quickly review and approve updates—ensuring your agent always references accurate information.

---

Step 7: Expand strategically

Once your first agent is humming along (2-4 weeks in), consider:

Creating more specialized agents

• Instead of one "Company Knowledge" agent, create focused agents for Product, Sales, Support, Engineering
• Each agent can have different sources, prompts, and audiences

Adding more deployment locations

• If your agent started on a Page, add it to the browser extension too
• Test connecting it to Slack channels

Connecting more sources

• Add that Google Drive folder you were hesitant about
• Connect your CRM or project management tool
• Sync meeting transcripts or call recordings

---

Advanced strategies to explore

Once you have one or more successful agents running:

Connect agents to your AI tech stack Use Guru's Model Context Protocol (MCP) to integrate agents into your existing AI tools and workflows.

Analyze trends with Research Identify patterns and opportunities across all your knowledge sources to understand what information your team needs most.

Set up a two-agent system Use one agent to monitor conversations for valuable knowledge, and another to deliver verified Cards to your team—continuously capturing tribal knowledge as your team works.

---

Measuring success: What to look for

After 30-60 days, you should see:

• Increased questions to your agent (people are finding it useful)
• 70%+ thumbs up on agent responses (answers are helpful)
• Fewer repeat questions in Slack or support tickets (knowledge is getting reused)
• Faster onboarding for new team members (they can self-serve answers)

If you're not seeing these signals, revisit Steps 4-6. The issue is usually discoverability (people can't find it) or content gaps (the agent doesn't have the right sources).

---