← Behind the Craft
Behind the Craft
How to Build for AI Agents and a Claude Code Second Brain in 25 Min | Ryan Wiggins
00:26:12
Loading summary
Transcript72 lines
- [00:00]
Ryan
Our job is being available wherever customers are. In the 1950s, that meant building a bank branch. In the 70s that meant building an ATM. In the 90s, that was a website. In 2010s, that was a mobile app. In the 2020s and beyond that is going to be APIs and set of services that you can connect with what I call the conversational interfaces of the future. I pulled this together and it was almost 5 million words from anything that was touching my surface area over the last five years and built that kind of as a knowledge base that then served as the foundation for everything. When I run a meeting, Claude tells me, hey, in this meeting you were doing this exact thing that is and your performance reviews.
- [00:33]
Peter
Of all the execs I talked to, this is the most impressive system at a high level I've seen. So yeah, so why don't we get right into it. Do you want to walk us through Mercury's MCP and what you can do with it?
- [00:45]
Ryan
Yeah, 100%. So Mercury's MCP is something we just recently launched. It's available in the Claude App store and it has a URL directly if you're taking in some different tools. But if you go into Claude, you go into the Connectors, you'll be able to find the Mercury app here. I already have it set up, so we're just going to jump right in. How Mercury MCP is useful is you can ask natural language questions about your bank. Something like, hey, Mercury, look at my spend over the past couple of months and see how I could save some money. I like to use whisper flow for some things like this. And so what's happening here in the background is this is connecting to our API and pulling the information that would answer this question and being able to take that and answer it pretty flexibly as well. MCP is a read only version of our API to kind of keep it safe. But we launched this about six months ago and we're really excited with how it's going so far.
- [01:39]
Peter
This is not your real account. Right.
- [01:42]
Ryan
This is a sandbox account for this demo. And so it's going to have real data. And one of the nice things about Mercury is we kind of have a great demo environment where we can send some things here so it will show some real data, but it is not my personal data.
- [01:53]
Peter
Okay, $500 per month, man, that's a lot, I swear to a lot of AI products, but I don't think I'm paying $500 a month.
- [01:59]
Ryan
You know, I think this is modeled after A and sometimes those startups are spending quite a bit. And you know, okay, so what we see here is, you know, Mercury is bringing in some of the finance data into Claude, and Claude's actually just launched a new update that has actually shown some of the data here in a pretty nice way. It has some tables, it has a bunch of different information. But using this, I can really get a good sense of where my money is going and where I could be saving some money here.
- [02:27]
Peter
So the API is read only assets, so I can't be like, hey, cancel my Netflix subscription or something. Let's do that. Right?
- [02:33]
Ryan
Yeah. So Mercury has an API that's broadly available and it matches exactly what you can do on the web application or the mobile app. And that is a read and write. But our MCP and the thing that is available in Claude and Anthropic is read only just to make sure that people have control and security over their finances.
- [02:50]
Peter
Got it. So what have businesses or people's favorite use cases been since you launched this mcp? Has adoption been great and know what's it been like?
- [02:58]
Ryan
One of the things is that we've been really surprised at how people have responded to this. You know, Mercury generally is building banking and what we call radically different banking for entrepreneurs and ambitious people of all kinds. MCP is just one of the ways that builders today are bringing in their finances to their different workflows. And so we've seen all sorts of people. One of my favorite examples is I was talking to an animation studio here in Los Angeles, uh, and there's a specific tax code. And when they connected their MCP into their finances and started asking questions, they were able to find some tax breaks that they didn't know about otherwise. That, to me, highlights the example of how MCP can be used. It's context about your business that's unique, novel. But then your insight as a business owner and someone who is operating your business, how those things come together becomes quite powerful. And we've heard a couple of. A couple of people have saved over $1,000. We have a running scoreboard of this number and it's getting quite high.
- [03:51]
Peter
Oh, already. Okay, so two questions is like, what are my top monthly expenses and where can I save? Another one is like, where can I save money on tax?
- [04:00]
Ryan
Right, yeah, yeah, exactly. I would say most people really love asking like, how can I save money on spending? What could I be doing to optimize my spend? But one of the things about Mercury and our API is it offers all the information that you have on your account. So it even offers things like your ein, your business address. And so you could have this kind of source of record of your business identity that then can come into your workflow whenever you need it. How many things do you apply for where maybe you need your. It's hard to find that. And so we've tried to make that all available within the MCP and make a great experience as part of it. It's been really cool to see how people are creating their own dashboard and their own kind of workflow on it. And I think that's probably one of the best use cases for it. It's just what happens when you can bring your financial data wherever you are.
- [04:41]
Peter
All right, dude. So from one product person to another, let me ask you some hard questions, man. Okay, so a lot of people are like, oh, I got to get my product agent ready, I got to build an mcp. But I feel like building an MCP is like the very last step. Right. What are some steps that you took before even building this MCP to make it really easy to use?
- [05:00]
Ryan
Yeah, I mean, first, Mercury really kind of has taken the journey that a lot of banks have taken over the last 20 years, which is it starts with a great website themercury.com experience, and you can see it at demo.mercury.com we believe is one of the best banking products that's on the market. Our mobile app pairs super well with that. And what we're asking with MCPS and API is what is the 2020s interface going to be like? And we think it is going to be, you know, kind of portable in a much bigger way. And so, you know, this MCP is built on a huge foundation of the products that we built at Mercury. Things like really reliable depository accounts, treasury accounts, credit card that people love, user management and different spin controls and things like that. And so all that stuff then comes into our API where when we built this mcp, we thought a lot about the experience of it. One of the things I found using a bunch of different MCPUs was it's kind of hard to log in. Sometimes you actually have to go into a text file or a config file to set up the server and set up your credentials or whatever it may be. So we tried to make it like logging into the web app. We tried to make it an oauth experience. It's just one click, same password, use the multi factor authentication to have that security that you need on a bank account. But it's seamless once you've connected it within Claude. It just works like it's your web dashboard. And that's what we think a banking experience should be like.
- [06:18]
Peter
Got it. Okay. And of course, having like, really good API coverage for, like, core work workflows and, you know, the data flows and stuff. Right. Like, did you also think about, like, documentation that agents can read or, like, any stuff like that or.
- [06:31]
Ryan
Yeah. I mean, this is the journey we've kind of been on is we've been thinking about our API for a long time, really. Mercury launched with an API and it's been around for a couple of years, but over the last probably six months, we've seen a huge amount of adoption for people creating API tokens, requesting into customer feedback channels that they want to be able to bring their data elsewhere. And so what we've tried to do is have the API have high coverage, but then make it super useful wherever we go. So you can do things like OAuth or MCP. And so it starts with that foundation, but then it's really about where are customers and where can you meet them.
- [07:04]
Peter
Got it.
- [07:04]
Granola Host
This episode is brought to you by Granola. If you're in back to back meetings, you know how much work it is to take notes, live and clean them up afterwards. That's why I love Granola, the best AI meeting notes app in the market. Here's how I use it. Granola automatically takes notes during a meeting, and I can add my own notes too. After the meeting ends, I use a granola recipe to extract clear takeaways and next steps in the exact format that I want. Then I can just share notes directly in Slack with my colleagues or even get Granola to share their notes automatically. Honestly, of all the AI apps that I use, Granola is the one that saves me the most time. Try it now at Granola AI Peter and use the code Peter to sign up and get three months free. That's Granola AI Peter. Now back to our episode.
- [07:51]
Peter
All right, dude, well, let me ask you another hard question. So like I said, Mercury's app is like the most beautiful, the most beautiful user interface for a bank that I've used. And I don't know, does it feel kind of weird that now people are like, now I'm getting my Mercury data through Janky openclaw instead of using that app or the product, Maybe you actually see DAU going down a little bit because people are just using these integrations instead. So how do you think about. Are you worried about where people. Like a world where people actually stop going to mercury.com and just go talk to your AI agents instead?
- [08:25]
Ryan
Yeah, I mean, the short answer is no. I think that our job as a technology company that is building incredible banking software is being available wherever customers are and meeting that moment. And in the 1950s that meant building a bank branch. In the 70s that meant building an ATM. In the 90s that was a website. In the 2010s, that was a mobile app. In the 2020s and beyond that is going to be APIs and the set of services that you can connect with what I call the conversational interfaces of the future. I think that we're already in a platform shift and we need to show up in that way and we need to make it available. And you know, I was thinking about this for myself. I'm a consumer. I use Chase Charles Schwab and it is so hard to get my data out of those products and it makes me so angry as a consumer. I just really wanted to build something that I believe was good for the experience of managing my finances. And we try to do that every day at Mercury. And like, you know, I think it's actually great if people are saving time and able to do more, have to spend less time in our web dashboard. It means that they're able to spend time building what they want to build.
- [09:26]
Peter
So, so how do you like, how do you like measure success for like a product that mcp like, because you know, like a user facing product is like, you know, dau and like retention and stuff like that. Right. So, so, so I guess it's like how often the agent calls your MCP or.
- [09:40]
Ryan
Yeah, totally. I mean, you know, it's funnels all the way down. Is, is maybe one way of putting it, you know, at the core it's our people using it when they discover it, are they hitting errors, are they going through once they set it up, did they have multiple engagements with it and then over time did they retain, do they expand and do they keep using it? And what we found is once people set it up, they really love using it. It's kind of exactly the use case you described where it becomes almost a week a daily, weekly or monthly workflow for them. And that's something I think we're really excited about because it's something that maybe we wouldn't have built ourselves, but actually people can build on their own. So yeah, yeah, it's like way more
- [10:14]
Peter
convenient, man, because now I can get the open cloud to push stuff to me from Mercury before I have to like remember to go to go log in and check my stuff. But now like every week it pushes stuff to me. So like, and I can give it, I can tell you to give me a pep talk. And the other thing is, like, it has my Mercury data. It has my YouTube data.
- [10:30]
Ryan
Yeah, 100%. I love that.
- [10:32]
Peter
Okay, so, okay, one last question about the mcp. So there's been a bunch of tweets out there about saying like, oh, you know, MCPs are dead. Is like, there's like too much bloat and too much context in there. Let's just build APIs and CLI tools and all that kind of crap. Like, how do you think about, how do you think about this?
- [10:48]
Ryan
Yeah, I mean, there's a lot of noise and like the world's moving pretty quick. And you know, one of the things that I think is, yes, inevitably I use mcps that clog up a ton of context. We are going to build a Mercury cli and this is actually something we're going to release in the next couple of weeks. And so breaking news here with you, Peter. The Mercury CLI is coming. We're really excited about that. But one of the awesome ways of thinking about MCP is right now it is hard when someone is using one of the consumer products, ChatGPT and Anthropic, to use the services like Mercury. In that experience, APIs and connecting to APIs directly is super hard. How those platforms have chosen to integrate is a first party app ecosystem which you set up and work directly with them, but then also the custom mcps. And I think that sort of app layer and distribution layer is so critical in this platform shift. If you look at like Apple, they don't, they don't allow third party apps. MCP is the third party apps of this ecosystem and I think going to make a big difference in how new products can come to market and open up what is possible for different companies.
- [11:47]
Peter
Interesting. Okay, so you think about MCP is maybe a little bit more accessible to the mass market and then the CLI is maybe for like the dev, the devs or like people who really want to optimize your context.
- [11:57]
Ryan
Yeah. And particularly because OpenAI and anthropic have gone in on the MCP ecosystem and they create apps that are built on the MCP structure. And so by them choosing that method, you know, it's kind of become a distribution service and the common one. And so it's not going away anytime soon until something can replace that. But what we see is kind of at the power user, at the premium Maybe not even the premium side. The, you know, deep Builder, they definitely want tools that are much more flexible, like CLI skills, different things like that. And I think that's something we're definitely going to meet them with.
- [12:25]
Peter
Got it. Yeah. I feel like MCP is the only thing that OpenAI Anthropic can agree on.
- [12:30]
Ryan
Yeah, no, they seem to agree on each other's product, but after each other.
- [12:35]
Peter
Yeah, yeah, they seem to. Okay. All right, well, why don't we shift
- [12:39]
Ryan
gears a little bit?
- [12:40]
Peter
Let us talk about how you use AI in your life as a product exec. Do you want to talk about how you just use these different tools day to day or maybe show us some work workflows?
- [12:51]
Ryan
Yeah, totally. All right, so maybe just quick summary on my role. I lead a product team at Mercury, about 20 different products that cover what we call our software side of the experience. It is how do people discover Mercury, how do they use it every day, how do they expand? I've been here about five years and I also work on the data team. I've worked on growth here at Mercury, and so I have a ton of context, a ton of information, but when I'm making decisions day to day, it is quite hard to access. And so I've been using cloud code to build out what I call my second brain. And so I'd love to walk you through it, if that's okay.
- [13:22]
Peter
Yeah, I love see it.
- [13:23]
Granola Host
Yeah.
- [13:24]
Ryan
All right, so maybe first I'll show a diagram because then I can show you how it actually works. So at its core, I built what I call the context layer or the knowledge base structure. This is essentially a download of all the information at Mercury from my five years at Mercury. The company strategy docs, every spec that's ever been written, every query that's ever been run, I pulled this together and it was almost 5 million words. I think I cut some to 4 for this just to make sure I didn't share anything too sensitive. But I tried to get all the information that has ever been created from anything that was touching my surface area over the last five years and built that kind of as a knowledge base that is stored on a local file system that then served as the foundation for everything that I built. On top of that, which is kind of a workflow where every morning I open up terminal, I go to Claude code and it references memory that has been built and is updated every day, it references that knowledge base using QMD and Toby's kind of local indexing solution that he has built out. And then it uses Claude hooks to inject that context into every question I ask. So I ask a question like how is activation trending? I'm not just asking that question, I'm asking that question with all the knowledge and history of Mercury going into that question. And it's actually injected into, into each request and it has made each request much, much better. You know then that kind of system plugs in to a set of different things, skills that I built out which are kind of patterns like an analysis or building like a small little app for myself or learning kind of coming back and feeding back into the memory system, MCP integrations. I plug in all the kind of core tools I have and then I have some things in workflows that are multi agent teams. I can send in like a full analysis to go understand all the data, go think about all the problems, go run an analysis, go kind of discuss it and then come up with a report. And that is kind of the system that I call my second brain or mission control. And it really goes into everything that I'm doing as I use it. I'm in quad code kind of all day, all the time. And so this is kind of the system that I have and I could give you a kind of a preview of it if you want to see it.
- [15:25]
Peter
Well, this is like I think of all the execs I've talked to, this is the most impressive system at a high level I've seen because most execs are just in meetings all day, man, they don't have time to build this stuff. So.
- [15:34]
Ryan
Yeah, well one of the interesting things that I have done as part of this is I've tried to make it so that I can pay more attention at meetings. So I have notion transcribe it and that feeds into the system so that every day, at the start of every day I get a brief of what's my calendar, what's a linear GitHub, Slack, what's the kind of meetings that I have and at the end of the day it summarizes those meetings, gives me the action points and other things that I have. And so I've kind of really built this into my like workflow where I start the day, I end the day and then like I start the week, I end the week using this. It has been so surprising for like the knowledge and context that it has unlocked for me all sorts of things.
- [16:11]
Peter
Yeah, Jimmy, Cloud, Cloud dude, how do you do some of this stuff?
- [16:13]
Ryan
Okay, so what I have here is my cloud code and kind of the entry point to most of my workflows. So I would start with a question like this. Hey, knowledge base, I'm here with Peter. I want to operate in safe mode for this conversation, but I want to show it how we would use some of the knowledge base that we have to be able to answer questions. One of the things Peter just asked me about is how is the MCP doing? Could you answer this question and try not to reveal anything sensitive?
- [16:38]
Peter
Yeah, I guess, because all that information is locally. It's like searched super fast too, right?
- [16:42]
Ryan
Exactly. And the qmd, the nice part about it is it's indexing. And so when it searches for something like MCP product traction, growth, it's not searching for those terms, it's searching for the concepts of those terms. So it returns adjacent stuff. Let me actually show you this and I'll ask you to black this out because I can show you what it looks like, but this stuff will be sensitive. So it references strategic context, which is a specific doc. It references the growth product team check ins that I have. Team charter. Yeah, different person's onboarding doc. So it's able to pull from about 20 different docs. So that context is inserted into the actual query that I run. And so when I'm running something like this, it is just again, totally enhanced and yeah, yeah. So here we have an answer on our kind of knowledge base. I asked a question, it brought in a couple of different pieces of context from the knowledge base and then gives me an answer. And you know, if I wanted to go run analysis, I could actually trigger it from here. I have it hooked up to MetaBase and RBI tool Omni to be able to run those. And so this is kind of a system that not only has knowledge, but it also has tools to be able to go take actions and run a full analysis. Get back to me insights on that.
- [17:54]
Peter
What is kind of like the proudest skill that you built that you can share with us? Is there something that's not super sensitive?
- [18:01]
Ryan
Yeah, I mean, I think one of the things that we built internally at Mercury kind of based on the system, was an automated data analyst. I have spent five years at Mercury answering all sorts of questions about how is growth going, how many people applied yesterday, how many sales leads have converted using this product. And we built a, you know, what I would call an AI agent that can answer about 80 to 90% of the questions that most XFN teams are asking at Mercury. And it really kind of started from me being able to prototype it locally, build the confidence that it's actually accurate enough and then go kind of peel that off and ship it internally to the company. And it's something like that is really, really unlocking a bunch of speed and capacity. So that's probably number one, I would say that's like a pretty cool use. Case number two is it's really been surprising how good of a coach it is. You know, I get a performance review every six months and sometimes it is, you know, high level themes or kind of abstractions. But when I run a meeting and it tells me one of the things I'm working on is that maybe I jump to the solution too quick. I don't probe enough about the question or work with my team enough, it tells me, hey, in this meeting you were doing this exact thing that is in your performance review. So it can kind of keep me honest at a much faster frequency than any other system can. And it's quite interesting just how that has actually changed my behavior because I know that there's actually like a system I'm going to have to be accountable to that is shaping my behavior. And I can tell you my manager and my people partner are so happy about this. It is definitely making the feedback land.
- [19:30]
Peter
Oh, hang on, let's talk a little more how that works. So you have like granola or something, take the meeting notes and then you fit it into the cloud clock code and it just gives you feedback after every meeting automatically.
- [19:41]
Ryan
So at the end of the day I essentially ask it, I am prompting it. And so it has the ability to pull the information from slack, from linear, from notion transcriptions or any other docs that I've created. And then I can ask it to summarize things. And usually that's a conversation takes about 15 minutes, but it's like, hey, did I have anything left over today that I need to take action on? Is there anything that showed up today that I should be thinking about for my performance? And like, you know, I go back and forth with it and it finds all sorts of things and man, it's, it's a great accountability measure is the way I would put that.
- [20:13]
Peter
Yeah, that sounds great, man. Okay, so you have it hooked up to Google Docs and everything. So you can just pull all the stuff that you did during that day, right? Is that it works. Got it. Okay. Got it. Well, yeah, I need that too. I get feedback that I move too fast too. It's like you move too fast, you got to keep bringing people along. And then I always forget about it. I always move fast. So someone to remind Me?
- [20:35]
Ryan
Yeah.
- [20:35]
Peter
Well, speaking of stats, dude, did you want to show us Mercury Insights? I think it's more user facing, but yeah.
- [20:44]
Ryan
So one of the things that we've been thinking about is how are our customers, the startups of America, adopting AI tools and what's kind of changing as the ecosystem changes. I noticed about six months ago that my behavior really started trending towards using anthropic cloud code and kind of building my workflows on that. And we did this analysis and looked at for new startups which Mercury gets a great signal on because we serve one in three startups in the US we looked at what tools are they choosing first, is it OpenAI or Anthropic? And you know, for the past kind of three, four years that has been OpenAI but recently that really shifted to anthropic. And it is both matching the kind of internal sentiment that we've had at Mercury, but also what I hear when I talk to customers, also what we see in the data here at scale.
- [21:29]
Peter
So this is the first model that they choose, right?
- [21:31]
Ryan
First model that, yeah, startups choose and for each kind of cohort, each quarterly
- [21:35]
Peter
cohort, yeah, I definitely see a shift, but I also feel like there's no real loyalty to any of stuff. A lot of people are saying Codex is better now and I'm sure a lot of people are switching to Codex now, switching back and forth.
- [21:47]
Ryan
Yeah, yeah. I mean it's fascinating, right? This is what competition looks like and it's so exciting. I mean there are going to be many missteps. We went in and got ChatGPT Enterprise and then we found there was a huge kind of demand for Claude as well. And it turns out that these products compete within companies themselves. And so one of the things that I'm really looking for is are these products able to enter workflows? For my example of my knowledge base, what I call my kind of second brain, I can't go to Codex. I can't easily flip that over. Maybe I could, I've tried but like it is not an easy flip to switch. I kind of have it now configured, set up the way I like. It kind of has trained on my response and the memory and kind of system that is building on that seems like it is locking me in. But you know, maybe OpenAI, maybe Codux is going to be able to push that further. What I'm really curious about is when a company chooses a product like this, when they choose to use open AI or Anthropic first, that model or that decision probably Makes a bunch of other decisions in the company. The next license that gets chosen, the enterprise chat model that gets chosen. One kind of choice at the early stage of the startup makes a lot of other choices and I think that we're going to see that play out over the, in the next 612 months.
- [22:54]
Peter
That's true, that's true. Yeah. The more context you give it, the less you like to churn. And also there's a whole ecosystem. Right, you can get cowork and all this other stuff too.
- [23:01]
Ryan
Yeah, exactly. And they're both moving super quick to extend that lock in for sure.
- [23:05]
Peter
You know you built a second brain with cloud code and how have you noticed how you and the product team, like how you guys are being products, like how has it changed since all these tools became available internally and like, you know, now you're kind of like age impaled, like.
- [23:19]
Ryan
Yeah.
- [23:20]
Peter
How has a lot of stuff changed?
- [23:21]
Ryan
Yeah, yeah, totally. I mean it's changed a ton and I would say Mercury has like significantly accelerated in the last six months. But it's maybe kind of a continuum that we've been on. There's really kind of like three things that we have anchored on that we find are the big levers for kind of EPD teams and what we're building for our customers. First is prototyping. We made a disposable front end. You can kind of see our demo, demo.mercury.com and what we've made is that that can be pulled by any PM designer, quickly edited. And you don't have to fully update the backend, you aren't editing the production site and so it's helpful for sharing ideas, early concepts, getting it in front of customers. And what I found is that's collapsed the time it takes for us to get to a good idea. And so you know, if you can build disposable prototypes super easily, that has kind of replaced specs and you know the kind of long write ups where you're conveying something, a meeting where you're trying to get everyone on the same page. You can just show them that and we have a bunch of people at Mercury doing that. 2 I maybe already mentioned this one a little bit, but an AI data analyst for everyone. I've been building a data team for the last five years and I've never been able to get on top of the demand that we've had for data. There's so many people that just want to know basic questions about the business. Now we have an AI data analyst and the thing that's become useful is that we can do analysis on that. What are people asking? What are the common things? And then we can go make sure the data infrastructure is set up to be able to answer those questions. And we actually found our sales team was asking something that didn't quite fit. That then helps us go improve the system. And so I think the system of self improvement is a big thing as well.
- [24:43]
Peter
Yeah, yeah. Like ever since I got access to the code base as a PM at work for like small bug fixes in the UI stuff, like when I asked engineers like, can you help me with this? It was like, hey, go, go do it yourself. Yeah, do it yourself.
- [24:55]
Ryan
You know, I mean, the other part I found is that I'm not waiting on someone to give me information. And I think being unblocked is one of the most important things in this environment. And like, you know, my ability to actually go ask what does the code do and have that conversation so that when I'm coming to the engineer and they're I'm treating their time with value, I think that's something that's also been really appreciated.
- [25:13]
Peter
Yeah. I hope all this stuff combined will shift the curve for BNPM to like, just way more fun because prototyping is way more fun than writing docs. Shipping stuff is way more fun than just writing a bunch of tickets. You know, like, I hope we can just like shift the curve slowly to more fun stuff.
- [25:27]
Ryan
Yeah. What I always say is there's going to be need to be someone that is keeping a group of people pointed in the same direction and making sure that we actually ship something. The job of a PM isn't going away, but it is changing. But it changed when we built on cloud. It changed when we built for AI and for cloud code and different products like that. The job is always changing, but, you know, getting the thing done is ultimately what the job of a PM is. Cool.
- [25:49]
Peter
All right, Ryan, well, thanks so much for your time, man. Where can people find you?
- [25:53]
Ryan
Yeah, I'm on Twitter. That's probably the best place Iwigs if you want to follow me. But otherwise on Mercury. Yeah.
- [26:01]
Peter
And if you have a Mercury account, definitely install the MCP and set up my OpenCL and hopefully things stay safe.
- [26:09]
Ryan
Yeah.
- [26:09]
Peter
Cool. All right, thanks so much, man.
- [26:11]
Ryan
Thank you.