“Cursor is a much better product manager than I ever was”: How this PM uses AI for PRDs, Jira tickets, and replying to coworkers | Dennis Yang (Chime)

0:00We've seen so many people use tools like Cursor to write code. We actually haven't seen anybody yet just using Cursor to write. That's what you're doing. The reason why Cursor is my favorite UI for the AI is it has all the interfaces and interactions and connections into the tools that are critical for my daily product management. I think we're at this really interesting place where because these primitives are being built in the context of software engineering, you're getting these concepts of markdown, git, commits, change tracking in these tools that used to be very software engineering centric.

0:35The most useful solutions will have interoperability as one of the key things. So any system that if it feels like it's locking that contents or data away, I'm not going to prefer to use that system. I don't care why you have these modes. I'm sure there's a good reason. But that's my content. and I want all my systems to be able to access it when it needs to. It's really improving communication. I'm reducing the time I'm spending writing status and at the same time improving the status content that is being circulated both up to leadership and across the organization.

1:06Welcome back to How I AI. I'm Clara Voke, product leader and AI obsessive here on a mission to help you build better with these new tools. Today, we have a fun conversation with Dennis Yang, Principal Product Manager for Generative AI at Chime. Now, this one makes me sweat a little bit because I thought I was the alpha AI powered PM. And Dennis shows me his workflows, which are way beyond anything I've seen before. He's going to show you how to use cursor to not only write your PRDs, but push them into Confluence or Notion, read comments, reply to comments, prototype AI tools, and more. This is an awesome one for anyone out there who's curious if you can make use of cursor without writing code, and I think you're going to learn a lot.

1:53Let's get to it. AI is supposed to make work easier, but I've been there. Weeks of setup, endless back and forth with engineering, and yet another tool the team never really adopts. That's why I use Zapier's AI orchestration platform. It connects with nearly 8 ,000 apps, so I can finally put AI to work without the drama, without the delays, and without pulling engineering in every time I want to automate something. With Zapier, you can roll out AI-powered workflows that do real work across your whole company in days, not weeks. I use Zapier every single day. It automatically responds to leads with enriched personalized data.

2:28It checks my calendar weekly and offers smarter ways to manage my time. And it even drafts emails for every new request that lands in my inbox. All of that running quietly in the background so I can focus on the work that matters. And Zapier's built for scale. With enterprise-grade security, compliance, and governance, it's trusted by teams at Dropbox, Airbnb, Opendoor, and thousands more. Go to try.zapier.com slash howiai to learn more about how Zapier can bring the power of AI orchestration to your entire org. Dennis, thanks for joining How I AI. I am really excited about this episode because we've seen so many people use tools like cursor to write code.

3:11We actually haven't seen anybody yet just using cursor to write. That's what you're doing as a product manager and somebody who's thinking about strategy all the time. So before we dive in, why cursor? Why writing? You're not writing a Licka code in here for at least this use case that we're going to see. So how did you kind of get into this flow with this AI-powered IDE? To me, the reason why Cursor is my favorite UI for the AI is a few things. It has access to all of the models that I want to try. So in Cursor, you can talk to Cloud, you can talk to GPT-5, you can do DeepSeek or whatever. That's the first thing.

3:52And then the second thing is it has access to a file system so it can write things down. And then it can have access to Cursor rules, which as you use it more and more, you can start to tell it how you want to do things. And then the final big thing is Interop. It actually works with all of the different tools that I need to work with. So that's through MCP. It can talk to Jira and Notion and Confluence, look at Figma, all these things. So basically, not only does it have the UI for the AI, it has all the interfaces and interactions and connections into the tools that are critical for my daily product management.

4:29You know, and the other thing that I would say is running on the desktop is just fast. You know, it's like pretty fast at doing those things as opposed to, you know, I think we're in this real moment where there's going to be a question of, are we going to start seeing more and more desktop apps for these AI powered tools just because of the performance side of things compared to web? Web is very flexible, but I like cursor because it is, it's zippy. So this is, this is your new hub for getting work done. This is what my screen looks like. And the other thing I'm noticing is that I want bigger and bigger screens.

5:04I'm running out of space because I want my chat, right? I want the thing, the artifact that I'm working on, and then I want the file system. And then over here, I actually have the – this is my settings pane docked to the bottom because you want to make sure that it's still green, right? So I need more screen real estate. and then yeah we had lee from cursor on recently and he walked us through the three pane kind of model of cursor for those who are listening and not watching you have your file system on the left where you're looking at what files you can work with or in context of cursor in the middle you have your artifacts whether those are code um code artifacts or content artifacts on the right you tend to have your chat and then something i want to call out for for people is you also have this bottom pane you can pull up and you not i was gonna say me as a developer that pain for me is always the terminal but this is a really nice little quality of life hack that i'm gonna start to use your bottom pane in cursor is actually the tools cursor settings so you can turn on and off mcps which i actually do all the time and confirm that they're working yeah exactly so sometimes cursor would be like, I can't do that thing.

6:19And you look and there's like a little red dot and you got to turn it back on. So. Yeah. Well, you know what, before we dive in, now that I'm curious and staring at your screen, would you mind sharing what MCPs just generally you have turned on? And then we'll go through the few that, you know, you really use on a daily basis, or at least in this workflow. Sure, sure. So I have MCP and I have two that talk to Atlassian suite. So one is our Jira Direct MCP. This is before Atlassian launched their MCP official server, first-party server. This is another one that I can just run locally. Notion, of course, talks to my Notion.

6:55Figma, talks to Figma if you have Figma running. And then GitHub talks directly. And then these two MCPs are, I'm actually going to talk about them. I'll talk about News API, which is an MCP I wrote to talk to News API. It's just an API that searches news articles. and then SEMrush is like an SEM, a search engine marketing MCP that I put together as well. Using cursor to write the MCP and then you actually can use cursor to install the MCP onto itself, which I love too. AI all the way down, as we say. Amazing. Okay. So, so you've loaded up cursor, you have it set up to write documents. Are there any other kind of cursor setup steps that you've done that you think are worth calling out for folks that are using it not?

7:40to code? Yeah, I think one big thing is as you're sort of working more and more, not just cursor, but just LLMs in general loves Markdown, right? So I'm starting to think more and more in Markdown. So I'm actually looking at something like this all day and writing and looking at Markdown. And then previewing Markdown is really important to see what it looks like, because this is actually hard to grok, right? So one of the good cursor settings here is, I think it's Markdown Preview. It's a setting that you can actually find inside. So you do Command, Markdown Preview. And in here, you basically want to automatically preview Markdown.

8:34and the extension that I'm using here, if you go to settings extensions, is called Markdown Preview Enhanced. So that's one I really like. And this one will actually automatically show in the preview pane when you click on it. Amazing. Which I absolutely love. Because it used to be that you had to know this secret code, which is like Command-Shift-V to open up the preview. But now you can just click on a Markdown thing and it pops in automatically. Yeah, I just want to call this out for folks who I know everybody loves Markdown now. It is the year of Markdown for sure. And Markdown can be rendered in a nice way.

9:17You don't have to look at a bunch of hash headers and weirdly formatted table content. So either by installing extension here or using that magic key combo there, you can get nicely formatted markdown right here in your cursor view. So, okay, we've done the setup. We've talked about extensions, cursors, your panes, your MCPs. Let's actually get into how you use cursor as a hub for workflow, how you would achieve something as a product manager, sort of end-to-end using cursor and these tools you've attached. I think it's been well-established that LLM's AI is really good for writing PRD. So you should be doing your PRD creation in whatever flow that you want, be it chat GPT or cloud or chat PRD.

10:06And for me, I actually do a lot of my writing. Now that I have this cursor set up, I do it directly in cursor. And it's working in Markdown technically. And I can preview it here. But here at Chime, it's not a single-player company. And I think that's one thing that I really want to be pushing on here is that the reason why you want to be interacting with all these other systems across not just your AI is that there are other people usually in your company that you want to be working with. So we have a ritual here at GIME, which is when PRDs are in early draft form, we'll share them out into our PRD drafts channel and then gather tons of comments.

10:44So I work inside Cursor. I'm making my PRD here. If it's in an early stage that I want some comments, I'll throw it into the PRD drafts channel. I do use Git to source control everything, but Git is not great for the whole company to go into and make comments. So I actually use the Confluence MCP to publish both into Notion and into Confluence because some people love using Notion, some people love using Confluence. um so then we have let's see here so we have here's this exact purity pushed into uh confluence um i pushed it into this uh last night and and threw it into our comments channel and already starting to get comments well what i want to what i want to call out really quickly before we go into maybe how this how this mcp actually works because i think people just like to see it is one, you know, we talked about using an IDE as basically your text editor, which is really interesting.

11:53You're not writing directly in Confluence or Google Docs or anything. You're writing in an IDE in Markdown with a nice preview. The second thing you call out that I'm curious your thoughts on is Git and source control for non-code assets or non-explicitly code assets. I think we're at this really interesting place where because these primitives are being built in the context of software engineering, you're getting these concepts of markdown, git, commits, like change tracking in these tools that used to be very software engineering centric. Yes. And I'm curious, just like, let's take a minute as product people.

12:37Do you think we're going to bridge that gap? Do you think there's going to be like Git for, well, maybe I should build it, but like Git for PMs? Or do you think we're still a little, like the abstraction is too high for it to be useful? And you need some of that UX polish that you're seeing in some of these more classic productivity tools. I have a lot of thoughts about this. So as we're sort of creating these artifacts that inform the product that we're building, traditionally today, these artifacts sit outside of the code base. So this PRD is sitting inside Confluence right here. But now what I'm realizing is since I'm using cursor and using Git to source control this artifact, it's actually now sitting inside this artifacts directory.

13:22And what I'm wondering is that if, I think I want to start to see if the artifacts are actually sitting inside the repo into which the code is being developed, that adjacency actually encourages the engineers and the AI coding assistant to continually have access to this. You don't have to give it an MCP to Confluence. This is the source of truth. So you can read it. And when things happen, so PRDs are typically a snapshot in time of this is what we thought the thing should be at this moment. In development, we constantly learn and iterate and learn and iterate, but rarely do we go back to the PRD to edit it.

14:05So what if I put a cursor rule in that says, hey, if I'm working on this feature and I'm learning things about the feature as I'm working on it, remember to update the PRD with the latest of how you're thinking about this. That's how the ChatPRD repo works. So we have a docs directory inside our sort of like repo for ChatPRD. It has a combination of product documentation and engineering documentation listed there. Even to-dos, like we've even pulled some of the like just task tracking into the repo because then, you know, everybody can work on it, AI or our human engineers. And so we've actually moved all of our documentation, PRDs, technical documentation, and user-facing support docs into the repo.

14:58Because, you know, when I'm working on the product, I'm 90 % of the time just working directly in the code. And the code needs the context of the documentation. The documentation needs the context of the code. And so I think it'll just be really interesting to see where the source of truth content pieces lie. I mean, the other thing that's kind of interesting is because you're in like a cursor or something that is a MCP client, you could just put the link to Confluence in there. And then it could call the MCP and look it up on this other source of truth. So, you know, who knows where all this data will accrue?

15:35There will be battle of the enterprise software source of truth. platforms. But I do think there's a lot of flexibility and optionality. And I do really believe, just like you said, code and documentation are going to start living in the same place more and more. Exactly. And that brings me to another thing that I'm really realizing is in this future, the most useful solutions will have interoperability as one of the key things. So any system that, if it feels like it's locking that contents or data away, I'm not going to prefer to use that system. Right? Yeah. Like I don't care why you have these modes.

16:13I'm sure there's a good reason. But that's my content. And I want all my systems to be able to access it when it needs to. Yeah. So I think that's going to be increasingly more important. Well, and what's really interesting is, you know, you've been in product management for a long time in B2B. And we used to always say in B2B, like your number one competition is like an Excel spreadsheet or a Google sheet. Completely. And now like your number one competition is like a source controlled markdown file. Like if you cannot outperform the source controlled markdown file as a SaaS offering, then you're not adding enough value.

16:50So I think it's gonna be really interesting. Okay. Apologies to our audience. You got two product people in here. I have a lot of thoughts about where SaaS is going. Let's get back to the workflow. So you're making your PRDs in cursor. You use the MCP, and it would be just really helpful for people who haven't used an MCP before, just to show how simple it is to do something like this. So let's say you wanted to push this PRD into Notion or into Confluence. Can you show us how that works for people who've never seen it before? Right. So this PRD looks great. actually this I need to tell look at that exclamation mark you're so polite so let's publish it into Confluence and I already pushed it so I'm going to tell it into don't overwrite the other one we already pushed make a copy please and I'm demoing this capability.

17:54All right, so we're demonstrating this capability. So this is basically all that's easy. So you always want to look to see that your MCP server is green. It's planning its moves out. And then basically, this is it. It's just running. It knows all of its tools. One thing, if you haven't seen how MCPs are set up, if you click on Tools Enabled, you can mouse over each one of these and you can get an understanding for how each of these tools works or what they do. And for me, just when I show people the MCP descriptions here, you start to form a mental model on what tools you've given your AI and what it could do.

18:36And how to ask for it. And so the other thing is not only how to ask for things from your MCP, so really understanding the naming of the tools, but the other thing is to look across your tools, And I'll tell you all a little painful issue that I ran into is a lot of SaaS products have things called projects or files or docs. And when I say update the project, it's like, do you want me to update the Confluence space? Do you want me to update the ChatPRD project? And so I think this ability to toggle on and off tools in the context of what you're working helps it just narrow in on what task you want to do.

19:12So a little lesson learned from Claire and named poorly named, maybe not in specifically named MCPs. Exactly. The current MCP server seems to be having issues. Is this what we get for doing it live? Yeah. Well, and here's the thing that you all, you know, here's the learning, how I AI. A couple of things. I think MCPs are really opaque to people who have not set up their first one. to set one up you're usually given at least on a hosted mcp a url you paste a couple lines of config into cursor or whatever your client is it'll like boop you to log into the app just like you would sort of go off another app or put in an api token and then that should give you access to it and then you do exactly what we're watching dennis do right now um if you all want to know the state of this highly stable technology and platform which is you toggle this button on and off over and over again until it turns green.

20:06Until it turns green. So we're going to give you the benefit of the doubt and say what really is going to happen is it's going to call these tools. One of these tools is going to say, push a Confluence document to a space, and then that's going to show up in your Confluence space as you prove in-house. This is the thread of conversation. This is what it should look like. You say publish this PRD into Confluence, and it effectively writes it. So when the MCP works properly. and then in my PRD itself I actually have a place where it can put its own published URL right so this is what it should look like once you've published it it really is this easy once the underlying plumbing is working properly yeah I do think that's worth calling out which is it's not just one way with these MCPs so you again showed hey I can take this content that I have in cursor I can push it to Confluence.

21:02But then you can actually do that round trip of context back into what you're working on in Cursor and say, great, add the URL to that page in my doc and update everything. And so it's just a very nice new user interface. This used to have to be buttons and complex API mappings. And now you have this language model that can figure out all these perimeters for you based on very high level context, do the work for you, kind of self heal on issues and go back and interface with other tools as well. So I think it's a really nice flow. So other than kind of creating PRDs and pushing them into Notion and or Confluence and how product manager-y of you that they have to go both places.

21:51They're both great tools. Yeah. Yeah. What else are you doing with these MCPs in terms of project management, project workflow, collaboration? We love to share our PRDs across the company to see what comments. This is the thread where I'm now gathering. Like I said, it looks like people have been commenting on the PRD. Please go through the comment one by one and let's see how we'll respond. So what the MCP actually does is it reads the PRD. It sees the exact comments. So I mean, I could use my own and read the comments myself and respond to them here in confluence but that's no fun um so we'll have cursor read the comments and then it actually does a breakdown this is this i found quite hilarious um it it organized the comments into high priority medium priority and then um other different types of clarifications and then actually wrote suggested responses um a lot of which sounded of decent.

22:49So I started saying like, yep, that one's okay. You can respond to that one. And since the MCP is authenticated as Dennis, as me, it appears to my other product managers at Chime that I am responding to their comments. Okay. So this is true behind the veil stuff. I have not seen this before. And this is spoken from somebody who has thought a lot about AI-generated PRDs. So just to clarify for people who are not watching, Dennis writes his PRDs in AI. He pushes them into Notion or to Confluence with AI. He then waits a little bit, reads the comments with AI. AI generates comments back and you review them and you either approve or not and then they get posted as you right so here's i love this i love it yes that's exactly the flow um and i am i'm in the loop right like one of the key things about building with ai is that you want to insert the human at the appropriate point um where they make a lot they're adding the most value right so gathering the comments that's not value add but reading them and understanding them and responding, like I can provide perspective there.

24:09Yeah, this is really interesting because I do feel like people are really good at that push part of creating an asset and pushing it into another system, but I haven't seen somebody sort of close the loop on the next step and iterate forward. I also think this shows that, you know, you're a busy person and honestly, people's, you know, The comments need you to think about them and agree with the response or not, but you don't have to type with your human fingers every single response. And you can also prioritize what you want to focus on. And I think that's really, really valuable as well. Exactly.

24:55I mean, ruthless prioritization is what good product management is all about. I love it. This episode is brought to you by Brex. If you're listening to the show, you already know AI is changing how we work in real practical ways. Brex is bringing that same power to finance. Brex is the intelligent finance platform built for founders with autonomous agents running in the background. Your finance stack basically runs itself. Cards are issues, expenses are filed, and fraud is stopped in real time without you having to think about it. Add Brex's banking solution with a high-yield treasury account and you've got a system that helps you spend smarter, move faster, and scale with confidence.

25:41One in three startups in the US already runs on Brex. You can too at brex.com slash howiai. All right, so what's next after you're gathering comments? Your PRD's great. All of your compatriots are loving the PRD that you've created and commented on. Typically, the next step in a workflow is to ticket this out into JIRA. So read the PRD and create an epic in the TI project. So I created a task automation is our JIRA project that we're going to be using. I'm going to make sure this is green, and then we'll watch it work now again. So what I really like about this is that when I'm doing this, right, it's reading the PRDs.

26:30I don't know what all of my other fellow product managers are like, but when Dennis creates tickets in JIRA, I could probably do a better job. But when Dennis's cursor creates the tickets, Dennis's cursor reads the PRD that I've spent a lot of time doing and then splits the effective tickets. The story tickets in particular are very, very well described. Yeah. So I love doing this style of ticket creation because it really, it does what you wish you could do if you had infinite amounts of time. It's your words. It's your PRD. It's your stories. And it's putting them where they should go, which is in the actual story tickets that then go to the engineers.

27:12And then they can, you know, you can always just say read the PRD, but they have to read the whole PRD. So let's create story tickets. Let's get this ready when it's ready. that are all, let's create story tickets that are all associated with the parent epic for this feature. And again, this is sort of a conversational flow. I've in my personal cursor for product management cursor roles, I'll say things like remember to always associate story tickets with their parent epic. I did notice that when I was doing this at first and it didn't say associated, then it would just create orphaned tickets without association.

27:56So this is the flow, right? It's handling a lot of the busy work, right, of that I don't get a lot of value from doing. Well, and I want to go back to what you were saying, which is this is one of those tasks, I say, that is such a toil reducer for product managers where you really hit cognitive fatigue on translating the same content for different audiences. And like PMs out there, we feel you, we know what you do. You take your customer notes and you turn them into a PRD and then your engineers don't want to read the PRD. So you make a one pager and then the one pager has to turn into an email for your boss.

28:35And the email for your boss needs to be three bullet points for your CEO. And like, you just do all this translation and then you get to like JIRA tickets or support documentation and you're like, I can no longer do a good job here. I have reached the limit of my cognitive interest in this task. And so you get lazy and you like kind of like do the epic name and the ticket names and then you say link to PRD in the in the description and you push that cognitive load on another person in your team. And so I think one of these tasks that are kind of like administrative, low, you know, low incremental value is a really good thing to offload something to, you know, to Claude or AI with.

29:24Completely. A lot of this like housekeeping rigmarole, like these tools are so good at. And then what's, as you're saying this, like what's another thing that product managers constantly do, which is that same exact shape? We update status. So how might I use cursor to update status? So here's our epic. Let's pretend I'm one of the engineers and I pick up one of these tickets. I comment. And look at the ticket. It has story. It has gherkin. has acceptance criteria, man. I built an agent or two that does this in a couple other platforms than Jira. And it's just like, everybody's like, oh, I got a good ticket.

30:11I got definition of done. It's organized. So this is a better job than I would do. I'll tell you that. I mean, Cursor is a much better product manager than I ever was. And now we're going to make it done, right? So we're going to do a few tickets in here. Once again, what is a status report other than using a tool and doing a job. So let's, once again, it's time for status reporting. Let's write a status report and describe everything that has happened since 9.24 or for this epic. Give it this epic here. And what it'll do is it will look at, it'll write JQL and essentially write a status report, right?