How this PM uses MCPs to automate his meeting prep, CRM updates, and customer feedback synthesis

1. 0:00 – 2:41

   Introduction to Reid Robinson and his role at Zapier

   1. CVClaire Vo0:00

      MCPs, I will say, it's a concept that's really hard to understand for folks.

   2. RRReid Robinson0:05

      Yeah, definitely don't think about the word. It really just is like app integrations for your AI tools. You can create these collections of tools from all the apps you use and give them access to Claude, to ChatGPT, to Cursor, all the places that have inputs for MCP servers today.

   3. CVClaire Vo0:20

      I use agents all the time, but it is hard to break that muscle memory of this is a deterministic workflow versus an instructive agent, even if it has access to the same tools and can do the same things.

   4. RRReid Robinson0:33

      And when it comes down to it, the two things we see people wanting to do is, one, giving their favorite AI tool the access to knowledge that lives in their apps, as well as giving them the ability to actually do things in those apps. Those are the two things that if that sounds like something that you need in an AI app you use, look for MCP or connectors, as it's often being called now as well, for that. [upbeat music]

   5. CVClaire Vo0:58

      Welcome back to How I AI. I'm Claire Vo, product leader and AI obsessive, here on a mission to help you build better with these new tools. Today, I'm talking to Reid Robinson, product manager on AI at Zapier, and what I love about my conversation with Reid is he's gonna show us how to put MCPs to work inside Claude to take over tasks that you really hate. We also talk about whether AI can be the perfect always-on team that works while you sleep, and some use cases to make your kids and your partner a little happier. Let's get to it. This episode is brought to you by WorkOS. AI has already changed how we work. Tools are helping teams write better code, analyze customer data, and even handle support tickets automatically. But there's a catch: these tools only work well when they have deep access to company systems. Your copilot needs to see your entire code base. Your chatbot needs to search across internal docs. And for enterprise buyers, that raises serious security concerns. That's why these apps face intense IT scrutiny from day one. To pass, they need secure authentication, access controls, audit logs, the whole suite of enterprise features. Building all that from scratch, it's a massive lift. That's where WorkOS comes in. WorkOS gives you drop-in APIs for enterprise features, so your app can become enterprise-ready and scale upmarket faster. Think of it like Stripe for enterprise features. OpenAI, Perplexity, and Cursor are already using WorkOS to move faster and meet enterprise demands. Join them and hundreds of other industry leaders at workos.com. Start building today.

2. 2:41 – 4:05

   Understanding MCPs as app integrations for AI tools

   1. CVClaire Vo2:42

      Hey, Reid, thanks for joining How I AI.

   2. RRReid Robinson2:44

      Thanks for having me here, Claire. Excited to chat today.

   3. CVClaire Vo2:47

      What I love about how you've described your role at Zapier, which I use all the time, I say, is like load-bearing infrastructure over at ChatPRD [chuckles] is you've, you've worked your way into a role where you get to kind of like pick what you're working on next in, in AI. And so I'd love to hear about what you're focused on and how that's actually impacted how you think some of- about some of your personal workflows.

   4. RRReid Robinson3:11

      Absolutely. So yeah, the way I often describe my role is often like Sisyphus of AI at Zapier-

   5. CVClaire Vo3:16

      Yeah

   6. RRReid Robinson3:16

      ... just pushing the rock up the hill, wherever that rock may be and whatever the hill might be. Right now, the thing I'm most excited about and where I'm choosing to spend a lot of my time working on AI is on our approach to MCPs. Uh, so we've got, you know, a server approach, but as well as what we're doing on the client side.

   7. CVClaire Vo3:32

      You know, MCPs, I will say, uh, still, I think, un- both kind of very hyped and underutilized by people. Um, because I think it's a, it's a concept that's really hard to understand for folks. So I'd encourage our listeners and our watchers who are a little nervous about wading into the world of MCPs to just really think about, "You know, if I could give my favorite AI chat client or IDE or whatever, access to a bunch of tools to do things for me, um, what would I want them to do?" And then go hunt

3. 4:05 – 9:00

   How Zapier’s approach to MCPs works with over 8,000 apps

   1. CVClaire Vo4:05

      for an MCP that, that does that thing. And I think you have, have built a product that has tried to abstract away some of that complexity for Zapier users, at least. Could you walk us through kind of a little bit of your approach there?

   2. RRReid Robinson4:17

      Yeah, absolutely. And I, I think you said it really well. The two kind of use cases I give people to just, like, I don't know, think about MCPs is... Yeah, definitely don't think about the word. It really just is like app integrations for your AI tools, and when it comes down to it, the two things we see people wanting to do is, one, giving their favorite AI tool the access to knowledge that lives in their apps, as well as giving them the ability to actually do things in those apps. So it's really... Those are the two things that if, like, that sounds like something that you need in an AI app you use, look for MCP or connectors, as it's often being called now as well, uh, for that. And yeah, the approach that Zapier took, for anybody not familiar with Zapier, uh, we're like, uh, one of the world's largest AI orchestration automation platforms, and what that really means on the MCP side is we've got 8,000 apps on Zapier that are like every SaaS app you can imagine. There's 30,000 searches and actions amongst that, and that's all exposed via Zapier MCP. So you can create these, like, collections of tools from all the apps you use and give them access to Claude, to ChatGPT, to Cursor, to all the places that have kind of inputs for MCP servers today.

   3. CVClaire Vo5:28

      Do you mind pulling that up and just showing us a little bit of what that, that looks like? And while you're pulling up your screen, I do bless you, MCP framework provider, um, but [chuckles] we gotta work, we gotta work on the branding here. So I think your description is exactly right. Like, app connectors for, for your AI is such a simpler way for the everyday consumer to understand this. Um, and so... Okay, so you're showing us Zapier here for folks that are just listening, and just, can you walk through an- oh, you have 8,000 tools that, or 8,000 apps you can add tools from. So this is your MCP server that you've added a custom set of tools that you're gonna use pretty consistently, either for a use case or just as a, as an individual, right?

   4. RRReid Robinson6:13

      ... Yeah, exactly. And so the way that it works is I kind of set up the ones that I'm using specifically for Claude. Uh, and so what's nice on Zapier's side, unlike many other MCP servers, is we actually are more like a platform for creating servers, so you can create multiple. And what that means is you can create specific sets of tools to use with Claude or with a particular agent, or with ChatGPT, or with Cursor, uh, really anything out there [chuckles] that supports it. Um, which is nice, 'cause for me, those tools are different from one place to the other. And yeah, you can see, or for those who can't see, uh, you can add tools from things like Slack, Evernote, Glean, Coda, Google Calendar, and you can actually start to customize those tools as well. Um, so whether you wanna, like, restrict them to using certain, uh, databases, like I've done with Coda. I- for my use within Claude, I really am using it for particular documents and particular sheets, for instance. Um, and other sides, like with Evernote, I wanna restrict it to writing to certain notebooks. Uh, so it really allows you to, like, customize the way you want your tool to work in different places, uh, which is quite nice, 'cause then it's like a single URL to give over to Claude and connect to. And now, now if I switch over to Claude here, uh, you can see that Claude now has, like, a single Zapier connector. But in that Zapier connector is, like, all of the different tools that I want Claude to be able to access.

   5. CVClaire Vo7:40

      Yeah, and one of the things that I'll call out for the sort of more advanced MCP users, and a challenge that I've always had, is when you're adding these individual MCP... Like, there's a Google Calendar MCP, and I'm sure there's a Coda MCP, is when you're adding these individual ones, you kind of have to do that configuration at the provider level. Um, and what I like about this approach is, like, this custom collection of tools is actually a really nice way to think about the MCP tools that you need, um, either just in general or for a specific use case. And then for, um, MCP clients out there, I just- I think we're gonna need at some point [chuckles] more, and I know you're working on this, but, like, you just need more granular control over tools pri- I mean, like, priority, I think, of tool calling is really important. I have, um, two MCPs that I use really frequently in Cursor, and they're always, like, competing for which one I'm trying to call. Because I say s- it's like, it's like search projects, everything has projects in it, always calls the wrong, wrong MCP. And so I do think, like, the m- the meta abstractions around MCPs are gonna start being more important as they become more adopted. So that's Claire's, uh, manifesto on [chuckles] MCP, MCP design. All right, so you have this custom MCP. I mean,

4. 9:00 – 12:05

   Using Claude Projects to improve tool usage instructions

   1. CVClaire Vo9:00

      what are specific things this unlocks for you? So what use cases are, are you using here to actually get more work done?

   2. RRReid Robinson9:08

      Yeah. So for me, there's, like, things that I don't love doing, [chuckles] um, is really where it helps me. Uh, so one, like... And for one of the things you just touched on, which is, like, the model's ability today to pick which tool-

   3. CVClaire Vo9:20

      Yeah

   4. RRReid Robinson9:20

      ... is a bit murky. Um, I think Claude is a phenomenal place. They've done a great job with tool calling. One of the tricks for anybody listening, uh, check out Claude Projects. Uh, in particular, one of the things that you can do in Claude Projects is provide very, like, detailed instructions for use cases. And so, for instance, I'll show- I'll share my screen here, but I have one that's all about the way I like logging and looking up data from CR- from our CRM for things, and I've actually told it, like, how it should use tools, in which order it should use tools, what data should go where, when it's creating records in those- with those tools, and it when... So then, in Claude, when I'm trying to do things, I can actually be like, "Oh, I'm doing a CRM thing. I'm actually gonna go ahead and select my CRM project and then shoot over a message." And now, Claude's ability to, like, execute across many different tools sequentially, uh, is so much better. Uh, so I'd highly recommend if anybody's, like, running into those, uh, things, try out Projects. Um, h- highly recommend it.

   5. CVClaire Vo10:20

      Yeah, I've heard a lot of people talk about using Claude Projects for knowledge, like loading it up with knowledge, but I haven't heard anybody talk about what you specifically gave as an example, which is use Claude Projects to give specific instructions relative to MCP tool usage and a workflow. And so folks, listen up, you can do that in Claude Projects, um, and probably other clients, to just make your, your use of your tools more efficient. Okay, so you have this Claude Project. It looks like one of the things you hate doing is updating your CRM, um, like a, like a true account. [laughing] He's like, um, I, I, I do actually tell people, um, MCPs are highly underappreciated by customer-facing teams. Like, what, what do customer-facing teams hate doing? Updating Salesforce. We hate it! We hate it. And so, like, you know, keeping good customer records, whether it's for a sales use case, a research use case, whatever, is, like, really tedious, and there are actually amazing MCPs out there to do this. So, uh, I'd love to see how, how this works in your flow.

   6. RRReid Robinson11:25

      Yeah, absolutely. So, you know, first- one of the first things I do is I have my daily planning one, and that actually, like, goes through my full calendar. And one of the nice things is I've given it access to, like, internal, uh, lookup tools. So for instance, when I run this, it can actually look up the person I'm meetings with, uh, Zapier usage, their company's Zapier usage, our past sales interaction with them, and it's able to, like, follow the process of doing all that lookup, and then when it comes back with ultimately my daily update, it has all of that research included. So again, really one that I've helped a lot of our sales team, uh, get set up with, and we've actually been, like, demoing it, uh, when we go to, like, events. Uh, that's been pretty fun.

5. 12:05 – 15:25

   Post-meeting notes management

   1. RRReid Robinson12:05

      But yeah, on the CRM side, so let's say that I have... The biggest one for me is actually my, like, post-meeting notes management. Uh, I u- I'm a big fan of Granola. They're a great tool, uh, but I struggle with the fact that sometimes-... I don't want to log [chuckles] those notes, or I don't do it all the time. Um, or it can just be really tedious to go about doing that. And so one of the things that I found really helpful here is I can-- I have, like, a Claude project that has a bunch of instructions on just how to log this data, where should it log, and then I can go in and actually select that project. Um, and then what it can do is it should be able... It'll have access to all the tools, and it'll start, like, running it through, uh, for this. And, now, this one's gonna be interesting because I'm-- I have the project configured to, like, our production database, and it's gonna try it with a different one, so let's see if this works for that. But it should be checking against this Coda document, uh, for things and seeing, like, what are the interviews I have scheduled and what are the things that are coming up. And if I go back to my little buddy Claude here, it's gonna tell me that, sure enough, nothing was found. Uh, then it can choose to, you know, start doing additional things, where I've taught it to, like, use our internal lookup to find this person's thing. Uh, I'm gonna skip this for now. Don't wanna pull in actual stuff here. And then some other things is Glean. Like, now I've given it an action, uh, as well to search, like, our internal Glean, uh, tool, which is awesome, because then I could see, like, "Oh, well, we talked about this customer in Slack," or, "We had notes, uh, from the CSM on what this meeting's supposed to be about." So it helps do a lot of that, and then eventually, what it gets into doing is start to say, like, "Okay, this didn't exist. Here's what I looked up based on the notes from the meeting. Like, let me go create that and run with it."

   2. CVClaire Vo14:03

      And that updates your Coda with what?

   3. RRReid Robinson14:07

      Ah! So yeah, that updates the Coda with like... And this is, uh, I'm doing a demo one here for y'all. Um, essentially, like, the Coda that I do have is a lot of the times I work on some of our, like, new products or new features, I'm doing, like, customer research in these, like, smaller, dedicated sprints. And so we typically will have something in Coda. I might also need to update our actual CRM, which will use HubSpot as well, so I'd have that as, like, an additional tool to log it as an activity on the meeting. But for the most part, I'm making sure that I have a record of this meeting, uh, who I met with, what-- if there were next steps, what were the next steps? Uh, it'll include some details on, like, what is- if there's a bigger opportunity, like, what are the opportunity details? Um, you could really get it to include a lot of things, and I think that's where, if I go back to the prompt for a second, uh, things like this here, where you'll see I taught-- I kind of like, I don't know... Our users always say they train the model. So if that makes sense to you, you can train the model, uh, on how to populate your CRM fields, because everybody's CRM fields are unique, right? Like, nobody uses, uh, standard cookie-cutter CRMs for the most part. Um, folks love their custom fields. Um, but models don't know what those custom fields [chuckles] are and what those choices are. So great way, again, just to get it to be familiar with you and

6. 15:25 – 18:15

   Comparing deterministic workflows vs. agentic instructions

   1. RRReid Robinson15:25

      working specifically for you.

   2. CVClaire Vo15:26

      What I think is really interesting here, again, as, like, a power Zapier user, is I have a similar flow, which is, I take Granola transcripts, I use the Granola, um, app in Zapier, but I have mapped this out in the standard workflow builder. So I have done the, uh, I think now that I'm seeing this, the inefficient task of saying, "Okay, like, if this, look up this record. If the record doesn't exist, do that. If the record does exist, do this." And so I have this, like, very similar CRM record updating flow in Zapier, um, but it's very step-by-step, kind of like deterministic workflow. And what I like about this, and I should be doing better 'cause I'm supposed to be like fairy godmother of AI, [chuckles] is you can actually just, in natural language, describe that flow, and I know this. I use agents all the time, but it is hard to break that muscle memory of like, this is a, you know, a deterministic workflow versus an instructive agent, even if it has access to the same tools and can do the same things. And so have you found that one, one path or the other is more or less brittle? Meaning like, is, is this actually more resilient, this sort of like MCP agentic instructions piece, more resilient to the complexities of your data, or do you find that it fails more or less than like kind of these nice, uh, netted-out workflows?

   3. RRReid Robinson17:02

      Yeah, it's a very good question. I think the... On the kind of reliability or where they fail, they're-- they've got their, like, pros and cons. The pros of doing things like asynchronously is certainly things can take longer. Uh, like, one of the biggest challenges right now with MCP stuff is they just, they can't take that long. Um, and so if you have, like, a lookup process that might take, like, seven minutes, like, that's not gonna work here. Uh, where that does, you can start to do a lot more of that in, like, deterministic workflow land, so to speak. Um, the other big thing, though, to be honest, like, the distinction that I-- that what it really boils down to, 'cause Zapier also has an agents product, where you could do this as an agent thing. But really, what this boils down to is just giving the tools, uh, giving the knowledge and the ability to take actions to the all the AI apps where you use them. It's kind of like the old product thing about like, you know, meeting your user where they are, right? Uh, right place, right time. And I have found that that is probably the biggest thing, 'cause there are so many times where I am, you know, like, y- if for anybody with keen eyes, you would have saw one of the projects I have here. It's actually even like idea jammer.

7. 18:15 – 20:04

   Reid’s idea jammer

   1. RRReid Robinson18:15

      I have a whole project dedicated-- hooked up to different tables and stuff like that for myself, when I'm just, like, jamming on a topic.... and it then will, like, research, like, have we explored similar ideas, or where might this be relevant? And it has more train- more like, ah, prompting there to, like, challenge me and certain methodologies to challenge me on that. So it really boils down to just, like, meeting people. And I'll be clear, like, one of the things we're seeing, though, from, like, enterprises that are tr- that are adopting this, is the fact that they're trying to make sure that these tools work for all of their employees, like, automatically. So that if they've rolled out Claude for the entire organization, when they log in and they connect to Zapier, it, like, has the tools that they should need for their role-

   2. CVClaire Vo19:06

      Yep

   3. RRReid Robinson19:06

      ... that is created by some, like, ops admin or someone.

   4. CVClaire Vo19:09

      Yeah.

   5. RRReid Robinson19:10

      Um, and that's been really powerful.

   6. CVClaire Vo19:12

      [upbeat music] As an AI founder, you're used to sprinting towards product-market fit, your next round, or that first enterprise contract. But speed isn't enough for AI startups. Buyers expect security, compliance, and transparency from day one. That's why serious AI startups use Vanta. With deep integrations and automated workflows built for fast-moving AI teams, Vanta gets you audit-ready fast and keeps you secure with continuous monitoring as your models, infra, and customers evolve. AI innovators like LangChain, Writer, and Cursor scaled faster and closed bigger deals by getting security right early with Vanta. Listeners can claim a special offer of one thousand dollars off Vanta at vanta.com/howiai.

8. 20:04 – 23:10

   Building a customer interview preparation workflow

   1. CVClaire Vo20:05

      There are pros and cons to each. You know, you mentioned three different methodologies. There's MCPs put in, like, the client, where you're actually working. There's agents, which do some of this, and, like, sort of an Ada client, and there's d- these deterministic workflows. And you do have a workflow that does use AI within a more deterministic flow. So do you wanna walk us through, through that one and just talk about why you selected this kind of model of implementation for this particular use case?

   2. RRReid Robinson20:33

      Yeah, absolutely. So one of the things for me, again, I like prep for a lot of customer interviews, and we have a lot of data.

   3. CVClaire Vo20:40

      Yep.

   4. RRReid Robinson20:40

      And sometimes one of the most embarrassing things for me, or it feels embarrassing, is when I get on a call with a customer, and they're, like, just a user interview that may have booked it via LinkedIn. They may have booked it via a referral from someone at a partner, right? Like, they come in from all over the place. Like, my Calendly link seems to, like, spread [chuckles] um, decently wide. And sometimes I'll get on a call previously, and I'd be like: "I don't know who you are. I don't know if your company uses Zapier. I don't know if they..." You know, and sometimes they're like: "Oh, yeah, we're both a customer and a, uh, partner," right? And I'm like: "Whoops! I didn't know that." Um, and so what I... And what I worked with, and our sales-facing team had similar issues in. So one of the first things that we did was we used Databricks, which houses, like, a lot of our data and makes that usable. And so they built, like, this whole series of things that allows, like, just simple lookup for, you know, given an email address, come back with, like, a whole writeup o- of it. And so essentially, this is a fancy-looking workflow, ah, but the gist of it is that for every other meetings that I'm having, it goes out, fetches that, like, research lookup, which takes time, and then it deciphers that into a... Like, it uses actually a Gemini step, since it handles more like the, the document type that I was working with, and creates the or appends it technically to the Coda page for that customer interview. And so this is really helpful for me again, because it's just, like, now, when I'm going into my meetings, I get things like this, where this is also where I'll, like, take some of my notes. Um, and so I can actually see like: "Oh, ha- they did use it," and get some really, like, crisp things, [chuckles] ah, to walk into the meeting knowing. And I think for anybody, especially in bigger companies, like, one of the biggest challenges we consistently see is they just, like, they're using... When they start to use AI, they're using, like, the base models with, like, no additional context to edit. And so the unlocks for them often become: how do you get your whole sales team to not only, like, use AI to log things, but also, like, fetch information from their CRM and from their data systems, ah, when and where they need it? Um, and that becomes really cool because, you know, technically, I could then throw, like, Databricks lookups into an MCP tool and put that to Claude. Um, gets really funky. Ah, those typically take too long,

9. 23:10 – 25:05

   Using Gemini for processing file-based data

   1. RRReid Robinson23:10

      though.

   2. CVClaire Vo23:10

      Yeah. Well, if you are, uh, the AI Sisyphus of your company, one of the things I might recommend, and I think you probably would as well, is you, like, buddy, buddy, buddy up to the data engineering team, that's for sure. Um, 'cause that's [chuckles] a really useful source of interesting, rich information. And then one other thing I wanna call out that may have zipped by people, especially those that are listening, is if you go into your user context Zap that you just showed us, you chose Google Gemini, and I just wanna reiterate why. Um, 'cause I've heard this a lot from different, different guests, which is the Google models, in particular, are just, like, great at files. They love a file. They're-- it's, it's great at files. So Gemini, um, really good at large files, files context, video files, audio files. And so anytime you have sort of like a file-based, um, challenge ahead of you or, or use case, I see a lot of, um, AI power users reaching for the Gemini models. Is that what drove this, this particular use case?

   3. RRReid Robinson24:09

      Yep, you nailed it.

   4. CVClaire Vo24:10

      Nailed it. Yeah.

   5. RRReid Robinson24:11

      Um, yeah, the output from our data team is actually, to date, a PDF.

   6. CVClaire Vo24:15

      Yep.

   7. RRReid Robinson24:15

      Um, and so it works very well, um, with that. Actually, it's HTML. That was the what-

   8. CVClaire Vo24:22

      Yeah.

   9. RRReid Robinson24:22

      So I convert the HTML to a file, 'cause then it works really, really well.... um, and a lot less tokens, which is nice.

   10. CVClaire Vo24:28

       Yeah, I mean, it's interesting, the ascendancy of the markdown file for, you know, the OpenAI and Anthropic models, or ChatGPT and Claude, and I do think Gemini has taken this, like, side angle, where it's like, "You know, yeah, but if you have a PDF or if you have some other file format, we're, we're, we're your model." So I, I think it's really interesting for folks who wanna go to the next level of implementation, again, to not only feed rich context into their AI use cases, but also really understand a couple of the high-level nuances of the major commercial models so you're picking the right one. 'Cause I, I would guess you'd get a worse output with a, with a different model just because of the,

10. 25:05 – 29:16

    Creating a virtuous cycle of customer feedback analysis

    1. CVClaire Vo25:05

       the data input. Okay, that is super, super useful. And then, um, so you've talked a lot about customer interactions, right? CRM updates, meetings, but you also get a lot of asynchronous customer feedback, um, including from me, and shout-out, [chuckles] uh, whoever is on the receiving end of my support and product feedback tickets, thank you. I appreciate you. You're always really, really responsive. How do you drive that responsiveness using AI or systematically across a pretty large customer footprint?

    2. RRReid Robinson25:38

       Yeah. It's fun. Um, there's a lot of things. I'll, I'll walk through one of the things I have found impactful, especially with, like, our newer products that we're pushing out. Um, one thing I'll say-- well, I can't show this, but again, does work with data and, and getting better relationships with data engineering. Um, I think when we've started to, like, unlock more and more capabilities with, [chuckles] with data on that front as well, uh, like with MCP, just this week, our team got to the point where we're now properly, like, analyzing a lot of feedback and actually creating, uh, pages in Coda for review, uh, for things for our team as we walk in, based on, like, new trends that are emerging amongst the data, uh, automatically, uh, which is quite fun. But one of the things that we've also done here that really helps is just, like, s- making it more searchable for folks. Uh, this is really helpful for, like, not even the core build team that's working on the product, but when I'm working with, uh, for instance, like, Sales or I'm working with PMMs, uh, that are supporting us with launches, they'll often have questions of like: "Hey, what feedback have we been receiving lately?" Or like, "Are people doing this sort of use case?" Right, and they're just-- they have very specific questions, or they're trying to understand something. Um, or it's the designer, who, as we're diving into a topic, we wanna, like, really quickly surface, uh, times where users have had, uh, issues with the, like, error log system, and they wanna, like, find, like, "Hey, can we find that?" And so created, like, a little chatbot here, uh, that essentially just like... It's really simple, but it, it is fed with a bunch of, like, databases, essentially, and then just, like, makes that really easily searchable, right? Um, it's a standard chatbot rag-type thing. I won't go into it, um, in much detail, but it's, like, internally locked down for us, um, and all those things, which is really helpful. Um, and, and we also use this sort of system externally as well. So, like, you'll see one of the things that we do here is I have our MCP helper chatbot transcripts. And so I, I have this kinda like end user-facing chatbot, and you'll see it, it has these, like, knowledge sources, which are basically just, like, our help docs, as well as one, uh, table, which is like a Google Sheet-type thing. That's our Zapier world of that. And I, I love this little system, and I'll just talk about it for a second. Uh, and it's really just, you know, for anybody that's working with data and knowledge management things, it's difficult to keep it up to date, and I found myself previously, constantly, like, trying to go back to our knowledge sources that these, these bots had, and just, like, trying to manually keep it up to date [chuckles] on, like, a monthly or quarterly basis. Uh, but one system I ended up finding that worked really well for me is I built, um, like one... There's a Zap somewhere, that essentially every time there is a closed, uh, support ticket, or if there's a, a finished chatbot transcript, it analyzes the conversation and tries to say, like: "What is the core FAQ from this? Like, what was the core issue? What was the solution, if any, and is that already in the knowledge base that we had? If not, please propose an entry." And so what I then do is I have my, like, human step here, where I can actually review the FAQs that it wants to submit, and all I have to do to-- I can edit it as well, like, what the answer is, and if I approve it, it goes over to a different database, which is the one that the bot is actually using. Um, so a really nice way that I have found consistently now on a number of projects, just to, like, rapidly iterate and keep those things up to date so that users are just getting, like, their answers faster, uh, which is really

11. 29:16 – 31:48

    The “if you could run ChatGPT in your sleep” framework

    1. RRReid Robinson29:16

       nice, so.

    2. CVClaire Vo29:16

       Yeah. What, what I like about this is I often tell people who are trying to figure out use cases of AI or implement AI solutions, is they really get stuck on, like, the I'm doing X, how do I use AI to continue to do X or do X faster, whatever. And that's fine. I think that's a l- like, I, I... I'm already taking meetings, how do I make taking those meetings a little easier? But the challenge I often give people is: let's say you had the perfect team with infinite time. What are the things your perfect team with infinite time would do in any one step? And your perfect support team with infinite time would look at every support question and would go see, do we have the right help desk content here? And if we don't, let's write great content, and then let's publish it, and then let's put that in the chatbot. Like, in an ideal-- but none of us live in ideal worlds-

    3. RRReid Robinson30:08

       Of course

    4. CVClaire Vo30:08

       ... we're super busy. And so I think this is, like, a really good example of that, where it allows you to operate at a next level of quality, not just like velocity, but a next level of quality. And then, again, like, the more high-quality data you create, the more you can power interesting AI solutions to your customers, like chatbots. And so, um, you know, again, anybody out there looking like-... if I had a full-fledged team of SDRs that were perfect and had infinite time, what would I do? Or like, you know, 10,000 support people with infinite time, what would I do? Like, start to think about those use cases, and don't forget to, to pluck those off, because I think they can unlock some interesting ideas inside your team, and then let your team act as higher leverage folks.

    5. RRReid Robinson30:55

       I like the way you put that. The w- the other... I don't know if it helps ever to anybody, but the other way I often tell folks that are struggling with that is like, "If you could run ChatGPT in your sleep, what would you do?" It is, you know, I've found a really good way to help people start brainstorming ideas. I will say as, like, a maybe a side on that, on the product design world, uh, we found- we did one experiment really early on, um, that was kind of like Mad Libs to help discover use cases and stuff. And it was a very interesting experiment in that it seemed to actually help people discover what they wanted to do, but it also challenged them to think through, like, what their pain points were. And it was really fascinating, uh, just to, to experience. So, uh, for anybody looking at that in their products, try a Mad Libs style, uh, AI-enabled system to, like, ask questions and ask follow-up questions, uh, with free-form text.

12. 31:48 – 33:03

    Quick recap

    1. CVClaire Vo31:48

       Just to kind of take a step back and walk through what you, what you showed us today. We have MCPs, um, Zapier MCPs in particular can give you a really custom set of tools to call. You like Claude, you like using Claude Projects to give instructions on tool calling sequence and instructions, so you get really high-quality outputs of that. And then you're really focused on, um... I heard you say very early on in the episode, "avoiding things you hate," [chuckles] which is, like all of us, updating the CRM. Um, you know, attending what we all attend, which are just-in-time meetings, right? You just get out of the next meeting, and you, [chuckles] you show up in the next one without context, so making sure you're prepped for that. And then kind of this virtuous cycle of customer feedback, support feedback, FAQs, um, you know, internal input, and then customer-facing, uh, help content as kind of a, a happy, happy circle here. And so I think this is great for anybody who's spending a lot of time with customers, whether you're in sales or support or product, um, to be better prepared, um, and, and get stuff done with less tabs open in your browser, which is what we all want.

13. 33:03 – 37:16

    Personal use cases

    1. CVClaire Vo33:04

       Well, uh, Reid, I'm gonna- we're gonna do a couple lightning round questions, and then we'll get you out of here, back to, um, pushing the boulder up the hill. My first question for you is, we've seen a lot of business use cases. What are, like, your favorite personal use cases? Like, what are ones that have really surprised you, either by making, you know, really, ha- sparking joy or just really solving a problem personally?

    2. RRReid Robinson33:27

       Yeah. I'll touch on two real fast.

    3. CVClaire Vo33:29

       Uh-huh.

    4. RRReid Robinson33:29

       Number one, in terms of solving a problem, family calendar planning. Uh, for anybody that has kids and families, like, family calendar, it's a real thing. Um, and for me, the struggle is, uh, my, my wife and I both like a physical calendar in the house, and we're reluctant [chuckles] to get, like, a full digital frame thing. Um, so we have a physical one that we write things on, but I like to live by Google Calendar, and if it's really not in my Google Calendar, it, like, doesn't exist. Um, and particularly if it's a family event that's in the middle of the, like, a normal day, then someone can book a meeting over it, and that's really not good. And so I actually have a Claude project called Family Calendar. It has really detailed instructions on... It's not too detailed, but it basically tells it, like, which calendar to look at, uh, how to add things. If it's an event that's at my son's school or somewhere, to leave time in my calendar to drive there and [chuckles] drive back if it's, you know, during the business hours, uh, so that that is blocked. And now what I do is, like, occasionally, I just take a picture of the physical calendar, and then it uses the various, like, find and update and create actions for Google Calendar through Zapier MCP and just, like, does all of it. Um, and I love that. Um, that's probably, like, one of the greatest things. Uh, other than that, these days, Suno, uh, they had a big V5 update recently. I have been loving it with my son and, uh, his other, like, friends in our neighborhood. Uh, we've made a lot of songs together. I literally just, like, talked to Claude, and I was like: "Hey, Claude, you're gonna write a kids' song for my son, Leo. He's four. Here's what we did today," and I just, like, told it what we did, and my son insisted that it have poop and fart jokes in it as well. And so I was like: "Well, you need to have some poop and fart jokes." Um, and my son has listened to this at least on Suno alone 14 times. Uh, we gave it to one of his babysitters, and she- they, they listened to- together, like, nonstop for an hour. Um, and we've done this with, like, his friends, and they've made songs for each other, and it's really fun, and it's... Uh, the other thing, too, is, like, some of the older kids nearby, like, uh, one of the girls is, like, 10, almost 10, and she's been learning about, like, prompting through this. Because she was like: "Oh, it said this, but, like, that's not right." I- and I was like: "Well, you gotta be specific in this, and you gotta, like, instruct it." And so I have the- I gave them, like, a, a whiteboard with a dry erase marker, and they're just, like, writing out their little prompts. Um, and then I input them for them. Um, and so that's been a lot of fun, and I think, I don't know, hopefully a little bit educational. Um-

    5. CVClaire Vo35:55

       I have to, I have to just, uh, bring us back to our starter topic, which is, does Suno have an MCP?

    6. RRReid Robinson36:02

       Hmm, that's a good question.

    7. CVClaire Vo36:04

       Suno team. I, I am also a, uh, extreme Suno power user. Love it. And [chuckles] imagine, imagine you could take your customer prep meeting and just give yourself, like, a friendly jingle to remember what they're talking about. [chuckles]

    8. RRReid Robinson36:16

       You laugh, but I've actually- I did that with-

    9. CVClaire Vo36:21

       [chuckles]

    10. RRReid Robinson36:21

        ... I took our, I took our MCP sales training session.

    11. CVClaire Vo36:24

        Yeah.

    12. RRReid Robinson36:25

        I actually took the transcript from Zoom, along with the deck, and I gave it to Claude, and I said, "Come up with a pop song," I think-

    13. CVClaire Vo36:34

        Yeah

    14. RRReid Robinson36:34

        ... for this.

    15. CVClaire Vo36:35

        Yeah.

    16. RRReid Robinson36:35

        And I've actually shared it, and a couple of our, like, sales team and product team have actually listened to it and really liked it. Um, I mean, yeah, we, we teach kids with music, and humans have been, I don't know, music people much longer than we've been reading people.... um, so it's, it's fun, uh, to explore that. On that-

    17. CVClaire Vo36:52

        I might have you, I might have you beat, which is I took a incident postmortem for, like, an engineering incident, and then made... It was like a punk song about how we needed to solve the root cause issues. Um, and it was called Renew the Certs. It was- it's a very, it's a bang- it's a b- a certified banger. So, um, we'll have to put a playlist together and put it in the show

14. 37:16 – 40:28

    Using Notebook AI to prepare personalized interview prep

    1. CVClaire Vo37:16

       notes. Okay, and then there's one last use case, which I love. I wanna make sure we spend a couple minutes on it, with NotebookLM.

    2. RRReid Robinson37:22

       Yeah. This one, I, I, the NotebookLM I use personally for learning. I put, like, a lot of things in it to learn, but I got one that I got a lot of value from, uh, and a lot of brownie, uh, bonus husband points from with my wife, uh, which was she was recently, like, job searching, and what I did for her on all of them was, like, when she got the interview, I would take their, like, careers page, I would take the job thing, I would find, like, more information, and I had, like, a prompt I used for the audio overview that was like, "You are preparing Anna for this interview. Like, make sure it's specific to Anna." And she listened to all of these before, and she, like, constantly got feedback throughout the process that she was, like, the most informed applicant. She clearly understood the space, 'cause I, you know, would always tell her, like, "Talk about the competitors, what they're doing in the marketing world, and what are trends going on there?" Um, and she loved that it was also like, "So Anna, we're gonna prepare you today." It was really cute. Um, but it worked exceptionally well for her. Um, she ended up getting, like, the ideal job that she really wanted. Um, and yeah, it was pretty awesome, so highly recommend that. It's also great bonus points for anybody out there with friends, family, interviewing, um, the ways that you can really help them.

    3. CVClaire Vo38:35

       Okay, I'm gonna have to make hats, which is, like, my love language is personalized AI podcasts. [chuckles] It's very good. It's very good. Husbands out there, uh, wives out there, partners out there, demonstrate your love by doing knowledge work via AI for the things that your, your partner needs. Okay, this is, this is amazing. This is really fun. Um, so many tabs opened in the sidebar. I'm sure so many other things you could show. Reid, thanks for joining. Where can we find you, and how can we be helpful?

    4. RRReid Robinson39:05

       Yeah. Where you can find me? LinkedIn. I'm most active on LinkedIn. Um, I do love the LinkedIn. Uh, so you can find me, Reid, R-E-I-D, Robinson, um, on LinkedIn. You'll probably find me pretty quick. Um, if you can help... Honestly, [chuckles] I'm a sucker for product feedback. Like, try some of the things I've talked about today. Tell me what worked, tell me what didn't work, tell me what you wish existed. Um, I also love hearing from the folks who are thinking about the future of all of this, um, who've tried, like, wacky things, and they're like, "Hey, if only I could do this." Um, I love that, like, bigger picture thinking stuff as well. So if you've got some, like, wacky ideas in the world of tools and agents and automation stuff, let me know. If not, yeah, try Zapier MCP. Give us some feedback. Would love any and all.

    5. CVClaire Vo39:54

       Amazing. Well, thank you so much, Reid. I really appreciate it.

    6. RRReid Robinson39:57

       Really appreciate you having me on, Claire.

    7. CVClaire Vo39:58

       [upbeat music] Thanks so much for watching. If you enjoyed the show, please like and subscribe here on YouTube, or even better, leave us a comment with your thoughts. You can also find this podcast on Apple Podcasts, Spotify, or your favorite podcast app. Please consider leaving us a rating and review, which will help others find the show. You can see all our episodes and learn more about the show at howiai pod.com. See you next time! [upbeat music]

Episode duration: 40:28