Search and Agent Builder together

Agent Builder and Search are designed to work together.

Agent Builder helps you turn a business problem into an AI agent that can support repeatable tasks and questions. Search helps you explore your organisation's knowledge, find information and verify answers against their original sources.

This guide shows how you can use both to solve real business problems, starting with building an agent.

Estimated time: 10 minutes

What you'll learn

By the end of this guide you'll understand:

• How Agent Builder and Search work together
• How to identify a good agent use case
• How organisational knowledge powers your agents
• How Search can help you explore and verify information
• How to use both products to solve a real business problem

Start with a business problem

Before building an agent, start with something useful you want to solve. Good opportunities might include:

• Employees repeatedly asking the same questions
• A process people regularly need help completing
• A team spending time finding the same information
• Knowledge that's difficult for people to access
• A repetitive task that could be made easier

For example: employees regularly have questions about expenses and need help understanding company policy. This gives us a clear problem we can turn into an agent.

Step 1: Build an agent

Open Agent Builder and create an agent around the problem you've identified.

For our example, we'll create an Expenses Assistant. Its purpose could be to help employees understand what expenses they can claim and guide them through the company's expense process.

Starting with a clear purpose helps you decide what knowledge the agent needs and how it should behave.

Step 2: Connect the relevant knowledge

Next, choose the organisational knowledge your agent needs to perform its role. For an Expenses Assistant, this might include:

• Expenses policy
• Travel policy
• Approval process
• Finance FAQs

Keep the knowledge focused on what your agent actually needs to do its job.

Step 3: Give your agent clear instructions

Tell your agent what you want it to do and how it should behave. For example:

Help employees understand the company's expenses policy and guide them through the expense process. Answer questions using the available company knowledge. Keep responses clear and practical. If the required information isn't available, say so rather than guessing.

Clear instructions help your agent provide more consistent and useful responses.

Step 4: Test your agent

Before sharing your agent, test it with the types of questions real users are likely to ask. For example:

• Can I expense a taxi home?
• What's the maximum amount I can spend on a hotel?
• Who needs to approve my expenses?
• How do I submit my receipts?

Review the responses and consider:

• Is the answer accurate?
• Is it based on the right information?
• Did the agent follow its instructions?
• Is the response clear and useful?
• Would you be comfortable with another employee receiving this answer?

Step 5: Use Search to explore your knowledge

Search gives you a way to interact directly with your organisation's connected knowledge. Ask questions in natural language to find information across the sources available to you.

For example: "What does our expenses policy say about taxis?" or "What is the approval process for expenses over £500?"

Search will find relevant information from your connected knowledge and provide an answer based on what it finds.

Step 6: Verify the information

Search also helps you understand where information has come from. Review the sources provided with the answer and open the original information when you need more context.

This is particularly useful when you're:

• Checking whether information exists
• Looking for additional context
• Verifying an important answer
• Exploring what your connected knowledge contains
• Investigating why an agent isn't giving you the response you expected

Step 7: Use what you learn to improve your agent

What you find through Search can help you better understand the knowledge available to your agent.

If your agent isn't performing as expected, ask:

• Does the information it needs exist?
• Does it have access to the right knowledge?
• Is the information up to date?
• Are there gaps or conflicting information?
• Are the agent's instructions clear enough?
• Is the agent trying to do too much?

Make any necessary changes and test your agent again.

Step 8: Share your agent

Once you're confident in how your agent performs, publish it and make it available to the people who need it.

Instead of repeatedly searching for the same information or asking another team for help, employees now have a dedicated agent designed around that particular need. Search remains available when they need to explore the underlying organisational knowledge more broadly.

Another example: HR

The same workflow could be applied to HR.

Start with the problem

Employees regularly need help understanding annual leave, parental leave, benefits, sick leave and remote working policies.

Build an agent

Create an HR Policy Assistant that uses your organisation's HR documentation to answer common employee questions.

Test the agent

Ask realistic questions and see how it responds.

Use Search

Explore the underlying HR knowledge, check specific information and verify answers against their original sources.

Improve and share

Refine the agent based on your testing and make it available to the appropriate employees.

The same organisational knowledge can now help people find individual answers through Search and power a dedicated agent for a repeatable business need.

When should I use Agent Builder?

Agent Builder is a good choice when:

• You have a repeatable business problem
• The same questions come up regularly
• There's a clear task or process to support
• A particular group of users has a consistent need
• You want the AI to follow specific instructions
• You want to create something other people can use

Start with the problem you're trying to solve rather than building an agent simply because you can.

When should I use Search?

Search is useful when:

• You have a one-off question
• You need to find a specific piece of information
• You're exploring a topic
• You don't know where information is stored
• You need additional context
• You want to verify information against its original sources

Not every question or task needs its own agent.

A simple workflow to remember

Identify → Build → Connect → Test → Search → Improve → Share

• Identify. Start with a real business problem.
• Build. Create a focused agent around that problem.
• Connect. Give your agent access to the relevant organisational knowledge.
• Test. See how it performs with realistic questions.
• Search. Explore and verify the underlying knowledge.
• Improve. Refine your agent based on what you learn.
• Share. Make it available to the people who need it.

Best practices

• Start with a real business problem.
• Give each agent a clear, focused purpose.
• Connect only the knowledge the agent needs.
• Test with questions real users are likely to ask.
• Use Search to explore and verify your organisational knowledge.
• Check the underlying information when an agent isn't performing as expected.
• Don't create an agent when Search already solves the problem effectively.
• Continue improving your agents based on real usage and feedback.

Frequently asked questions

Do I need to use Search before building an agent?

No. If you already have a clear use case, you can start with Agent Builder. Search can then help you explore the organisational knowledge available, find specific information and verify the sources behind answers.

Should every common question become an agent?

No. An agent is most useful when there's a clear, repeatable need that benefits from a dedicated experience or specific instructions. For one-off questions and broader knowledge discovery, Search may be all you need.

Can Search and Agent Builder use the same knowledge?

Where supported by your organisation's configuration and permissions, the same connected organisational knowledge can support both Search and agents.

What makes a good agent use case?

Look for something that happens repeatedly, has a clear purpose, uses defined organisational knowledge, solves a genuine problem and has a clear group of users.

What should I do if my agent isn't giving me the right answers?

Start by checking the agent's instructions and connected knowledge. You can also use Search to explore the underlying information and confirm whether the knowledge your agent needs is available and accurate. Then make a targeted change and test the agent again.

Next steps

You now understand how Agent Builder and Search can work together to solve real business problems. Start with a problem in your own work, build an agent around it and use your organisation's knowledge to test and improve the experience.

You may also want to explore:

• Agent Builder overview
• Build your first agent
• Test and improve your agent
• Search overview
• Ask your first question
• Common mistakes