The agent that turns knowledge into action

The market is flooded with what everyone claims to be AI agents, but not all agents are created equal. Simply providing information through AI is not an agent unless that agent is able to take complex actions on the user's behalf, but of course with guardrails in place. Today, I'm going to show you how an AI knowledge agent built on Merlin Agent Builder can not only provide intelligent and thoughtful responses, but can take complex actions across multiple systems.

Let's go ahead and take a look at this agent in action then. So I'm going to start with my initial prompt. I'm going to ask it: I wanted to go ahead and take a look at the last six months of tickets related to feature x because we want to go ahead and update our FAQ page.

Now what we've got here is a site that's got a relatively straightforward set of FAQs. So you can see that it's just common technical questions and technical responses. But what I want the agent to do is go ahead and look at the last six months of tickets that exist within our Jira instance. And it should take a look and compare that against the current FAQ page to see if there's any improvements or extra pieces of information we can use to make sure that it's kept up to date in line with some of the challenges that we're seeing being asked the most frequently.

In this case, we can see the agent's come back. It's completed its feature analysis and has taken a look back at around a hundred tickets and has summarized it really nicely for us around the different areas. Some of the recommended FAQ updates because it's taking into account the current FAQ page and provided us some recommendations for things to improve as well. But it's also asking me for the ability to take some action here as well, which is great. So being able to take this knowledge and actually take that a step further and take some action, whether that's updating the FAQ page with a new content once approved or creating a Google Doc with the full analysis and proposed changes. But I'm going to ask it to go ahead and update the FAQ page, please.

And we'll see how the agent is now able to actually take action and actually go ahead and update this FAQ page with new information for us too. So you can see the agent's come back and completed all of its actions. So it's taken those two actions of one updating our FAQ page. So if we take a quick look at the current page, and we can give that a quick refresh, we should see that it's now got a much more detailed set of responses to some of these based on some of those common criteria found across our ticket. And our agent has also created for us an update summary document as well for internal analysis. So we can see kind of what it took into account. Obviously, this is different from the public facing FAQ page, but it gives us an idea of the recommendations and where they're stemming from, things like the actual load and the tickets that are analyzed there as well.

Let's take a look at a slightly more intricate example. So I'm going to ask my agent to go ahead and take a look at some analysts briefings that we've currently got sitting across a wide variety of document formats in our Google Drive folder. We also want it to go ahead and take a look at some areas for improvements against our competitors based on that. Then I also want it to have a take a look into our Productboard feature request to see if we've got common requests from customers that align with some of these improvement areas from those analysts briefings. So let's go ahead and take a look to see how the agent is able to crunch this data from a wide variety of data sources. And we'll see in just a moment how it's also able to take action on that data as well.

And so we can see the agent has come back after conducting its analysis and has broken it down into the key product areas from the scoring and also identified a few areas for improvement where the score might be slightly lower. It's also gone ahead and correlated that against some of the Productboard feature requests, providing a link through to some of the notes that they're related to. So we can see that cross correlation that is done there as well. Now we can also see at the bottom here, it's asking us if it wants to go ahead and take some action so we can go ahead and create a detailed product plan for us and also creating a new Jira project.

So I'm going to go ahead and ask it to please create this as a Jira project. And so what we'll start to see is the agent is, again, able to take action with this knowledge that it's done. So it's done that analysis, and now it's going to take that data and create that as a new Jira project for us with all the necessary epics, assigning timelines to each individual area so that we can start assigning issues to each section as well.

And so in this case, we can see it's gone ahead and created us a Jira ticket for all of these initiatives. So we've got our PEI. What we'll also see is that it's gone ahead and posted this as an update in our product updates channel as well with a link straight through. What we can do is if we click onto this, we should see this created as a new project Jira instance. And if we take a look at the timeline, it's broken down those key areas and associated sometimes with that, things like the security and compliance sprints, the platform reliability, and the integration capabilities. Again, all of these are items that are analyzed as things that needed improvement based on that analysis that has been done. So as you can start to see, the agent's really powerful in not just having access to a wide variety of knowledge sources, but also being able to take action with that data as necessary as well.

Now if you were to using a product born as an enterprise search tool, everything you saw wouldn't be done by the agent. It would be done by you. Without the ability to create powerful tools that can take action in a low code manner like you get with Tray, search tools will be able to surface loads of data, but you'll be left to figure out how to take the appropriate action on the other side. The Merlin Agent Builder allows you to take your agents to the next level. And the agent shown today is just one example of what's possible when you build on one platform for every agent.