Full Course: The AI Stack We Actually Use for Prototyping, Strategy, and Personal OS (2026)

When I think about a personal OS, you can really think of this as like almost being an embodiment of you to some degree in markdown files. I was able to build prototype for my product way faster than the designer could make the Figma box. By the time the Figma mocks came up, I already like tested the prototype with some users and got some feedback. Doing this stuff is like way more fun than like writing some document. — What all three of us are doing is we basically created our own personal AI product that we're curating and iterating on. That in itself is like an incredible source of learning and building intuition. — AI is really helpful for me stay focused on something over multiple weeks. It's an accountability partner. — If I could summarize what I've learned about AI, it's all text files all the way down. — Welcome everyone. To kick things off for the new year, I'm excited to host my friends TA and Aman. They're both real AI PMs, not the ones that just add AI on their LinkedIn. They're the real deal. uh who have taught tens of thousands of people how to use AI. And today we're just going to do a very special episode where each of us will demo our favorite AI workflows to do product work from you know writing strategy documents to prototyping to using cloud skills. So welcome to you both. — Thanks Peter. Great to be here. — Awesome to be here man.

— Let me start by demoing like what all of us actually do. We all spend a lot of time you know writing strategy memos writing prds and these documents right in Google Docs. So I have a project here that I built in cloud that helps me write these documents and it has a bunch of files attached to it. One of the files is this template file that has my strategy document template. So basically it's just like a pretty simple template with the problem, the vision, uh the core product principles, the goals, solution, and what we're not going to do. And I also want to try to keep this um basically any kind of memos I write these days I try to keep to one page max because if you write longer than one page someone will just use AI to summarize it anyway. So I try to keep it to one page max. So what I do is I have this template uploaded to my project and I also do some deep re research. So let's assume that all three of us are working on a strategy memo for 2026 planning for Google AI studio. Okay, Google AI Studio is basically this product uh to that Google built to prototype apps and you know websites. Let's take a look at the research that I put together. And like this stuff usually isn't perfect. It's not going to give your full strategy for you obviously, but it's really good reference material for you to riff the AI on coming up with the strategy, right? Let's see. There's a big explosion in vibe coding. Uh and there's a big chasm between prototyping and deployment. And they started getting to some random stuff about AI governance initiatives from IBM that I don't think is super re relevant to this stuff. Yeah. Okay. So I basically just uh saved this thing uh to the pro project and now let's start a new conversation with uh AI. Let's just say uh using my strategy doc template and Google AI studio reference come up with your best one pager strategy memo for me to review. All right, let's go ahead and see what it comes up with. All right, here we go. Enterprise product teams can prototype AI features in hours but blah blah abandoned pies with production. Enterprise AI prototypes are born production ready. Okay. So, basically it's saying that uh once I prototype something in Google AI studio, it should be very easy to ship it or make it production ready, right? Through some other things. I mean, what do you guys think about this strategy? — Feels kind of vague um to me. Yeah, I guess like on first pass. It's like a it's very like high level kind of generic almost. Let's see. Um so, I guess like if I put myself in the shoes of like a PM on this team, — I'd probably start more with like what's the north star? what's the goal and vision for the team? Like what's our like two-year plan like high level? What does success really look like? — And then I might work backwards from there and then think more in like a quarterly incremental sort of what do we want to achieve in the next year towards that northstar? What's the metric we want to move? I probably wouldn't start directly jumping in with like SSO and enterprise and these types of like very specific niche features just based on — you know like just the fact that we kind of gravitated towards enterprise here a little bit. — Yeah, I totally agree. Normally I would have probably come up with an initial kernel of idea myself and then got AI to do research around that. Uh so maybe we skipped that step but why don't we actually think about that right now. I think there's two different types of vibe coding apps. One is to improve some production code and another one is to just like pro like PMs and designers and other people prototyping features to show to users. And I think where Google AI studio fits is probably more in the prototyping space. um because again they have like anti-gravity and you know whatever came from uh wind surf to to build the production side the IED right so let's assume that like you know we want to uh win the prototyping space so then the competitors are basically lovable replet um Figma make and some of these other things and I think what Google really uh Google's advantages is like the Gemini model is incredible at multimodal output I think the design in particular has become a lot better with Gemini 3. Like it used to be pretty crappy like purple AI slop, but now it's actually stuff that actually is pretty close to what the actual webites and stuff look like. So maybe um they have probably have the best model for this stuff and um they also can like um afford to kind of be the lowcost player, right? they don't have to they don't actually have to make any money theoretically because they're backed by Google but like you know something lovable has to eventually make a profit to survive. Um and of course they can also bundle everything in with uh like you know Google suite or like all the other stuff that enterprises already use for goo Google products. So I think like you know Logan and the team is in a pretty happy space here like they as long as they don't screw up like they're probably going to win a ton of share in the market. What do you guys think? I think that there's like an interesting note here as well. You know, it's kind of a product strategy thing that I've heard from a few interviews from like Boris from Cloud Code and um Cat a few other folks uh even the codeex uh PM that there's like the models are improving so fast that building product around it is really hard to predict. And so it kind of feels like that might be a core part of the strategy too is being really close to the model layer even though it's not maybe not reflective in the like first iteration of the product should probably be some part of our Nordstar or part of our strategy in some way too. — Got it. I also think one more thing is like um from a feature perspective like ma magic patterns some of these other uh startups they actually are like they have better products for prototyping than Google studio or even liable like uh they have features where you can prototype multiple variations at once and then you can like mix and match between uh different v variations I mean like you know when a designer prototypes they don't just make one thing right they make like five different variations and then they talk to the team so like I feel like a lot of that basic stuff is just mi missing. Okay, so let me kind of recap this for a little granola thing. So basically I think um one vision could be um like validate your idea in front of customers in 30 seconds or less or something like that like that's kind of like a vision and then um some of the pillars could be like make it easy to go from prototype to production explore you know whatever multiple variations at once or like you know be part of the design process. Um, and then you guys want to offer one more one or two more something about security maybe or — first I love what you're doing here and I'm really resisting the urge to prompt inject — liquor knowledge transfer. — Yeah, you should imagine that you're a really great PM. Yeah. — And the strategy should be entirely in pig Latin. — Exactly. Yeah. — I mean, yeah, this is just a sandbox. This is not like a real thing, but like the real thing of coming up with the strategy will probably require like 100 more iterations of this, but like you know this is kind of like a way to work with AI to do it. — This is brilliant. I love it. — All right, cool. Why don't we just stop the transcript? All right, I'm just going to generate the notes. Okay, so uh you know like in reality like what I do is yeah, I have a meeting with my product team and I use like Zoom AI or granola to take some notes and then uh yeah, let's go ahead and copy this. How do I copy stuff in granola? Copy text, right? Okay. So, uh I put some notes into uh Oopus and let's see. Okay. So, it's actually smart enough to put push back on some of our spitballing. Your research docs makes a strong case that prototype production gap is the opportunity and enterprise governance is the moat. Winning our prototyping first ear. Okay. So, maybe the long-term vision is to be the production pathway, but 2026 should be winning our prototyping first. I mean, that seems fair, right? like you know I think prototyping space is fairly competitive and if they can get a strong foothold in that space they can link up with the rest of their Google products and you know make the pro prototype to production thing happen. Um okay I mean we don't have to get into this but like um this is basically like this is basically how I work with now uh write these memos and docs. I don't go off into like a dark room and try to write stuff from scratch. I always just iterate with AI and um in this case I think we kind of had some mixed results but like the more context you give it the better like you know more meetings with the team and meeting notes or more re research or even just like going out on a walk and like kind of forming my own opinion about things and kind of feeding that into AI and going back and forth with with it is kind of how I write these docs. Now this episode is brought to you by Quarry. The biggest slowdown in product development isn't your code, it's your process. Specs, texts, updates, and keeping everyone in sync can all take time away from actual development. Corey is an AI agent that handles the busy work of product development and turns rough notes into dev ready specs, breaks work into tasks, and can even hand things off to coding agents like cursor or Devon with full context intact. Teams using Corey report 48% time saved on spec creation and scoping. Try Corey with your team for free today at corey. ai. That's k o r e y. ai. Now back to our episode. Let me actually do another demo now. So let's keep on

the Google AI studio theme. So um I think like in November I was like writing another one of these documents, right? And like getting a lot of comments in Google and like it's just like a huge pain in the ass. So then I um decided to just prototype start prototyping and um Google AI studio is actually pretty brilliant for prototyping. Um well one thing is actually a proof that my work so I can prototype you know and just prototype our products without having to worry about security but it's you know as I mentioned Gemini is actually a really good model to um build a UI. [snorts] So what I did was I basically um just as it to replicate the Google AI studio UI. Please reproduce this UI and make it interactive and it did a pretty decent job of replicating the Google AI studio UI as you can see here. So what I typically do is I save this UI as kind of a base temp template. So if I was on the AI studio team, this is the base template that I'm going to use to build like uh to prototype additional features. And now let's say we want to prototype a feature where um [snorts] I mean this is pretty uninspired, but like let's say we want to add a bunch of templates to Google AI Studio, like templates to make this stuff, right? So, I'm just going to I hope that I I hope the team never does this, but I'm just going to go copy this thing and be like, "Hey, can you add some pre-built templates to replace the discover and remix app ideas section? It should look like this uh like this image, but fit our existing theme. " Of course, it's Peter who builds the Frankenstein monster of all AI prototyping tools together. — Yeah, it's got all of them in one. Yeah, — you got to get inspired by your competitors. You know, there there's no shame in getting inspired. Okay, so it's going to go ahead and do this. And um I think another great thing about AI Studio is uh like once I have the base template, I can share it with you guys. Share app, right? Then you guys can iterate on it too. Let's see what it comes up with. But while we wait, um, do you guys have any preferred prototyping tools or — I'm still wrapping my mind around the the AI studio instead of inside of AI Studio. So, I'm on — Yeah, I mean, I wonder if you could just keep recursively sort of going through, you know. Um, yeah, I mean that that's I personally I love this, you know, kind of it very much feels very like no code, low code, very approachable. It's just prompting to get UI for iteration and for like this feels like true prototyping in the sense of like it really shouldn't be much work at all to get some UI that you can use to collaborate. So um — you know and then I I'm curious like going from this into real code and you know making it look like your own application that feels like the next leap that could be made here if you like hook it up to your codebase uh and get like a full endto-end prototyping in AI studio. Yeah, it's actually kind of like, you know, just working with my team, it's actually kind of funny because like I uh I was able to build prototype for my product way faster than the designer could make the Figma mocks for the product, which is kind of a interesting situation because like you know um by the time the Figma mocks came up, I already like test the prototype with some users and got some feedback and um it kind of makes me think that uh we should flip the process like we should do prototype first development. So build a prototype first because it's like really easy to do explore a bunch of different ideas and talk to our users and then once you get a good direction then make the PRD and the design so you can you know work out the edge cases and a bunch of other stuff. — Yeah. — Um I'm not I'm not sure how your team do this or if you guys do this. — Yeah, — I think in this I think it's still kind of early like this type of workflow that you're describing. What's interesting is um for me, you know, coming from the hardware world before this, this is that's exactly what you would do in hardware as well. Like you wouldn't spend all this time like building a final product. You build like really happy prototypes first and get feedback from users. I'm curious, Peter, if you've gotten any feedback from designers that you've worked with, you know, on the fact that, you know, we're using prototypes to communicate ideas that, you know, do people feel like this is something that people ask me a lot actually. It's like especially designers, like should I be worried about this? like how should I think about PMs coming up with designs? Um yeah, curious if you guys have thought. — I think the best thing about this thing is that uh it kind of democraticizes the process like it's like right now this thing is my prototype, right? But like you know Aman, if you are the designer, you can just take this thing and like build like make it better yourself and build your own version of this. So this is more like the team's prototype as opposed to any of our individual prototypes. And I think uh you know going back to the theme of you know any anyone can wear multiple hats now as a builder like this prototyping uh phase is like the clear um example of that like designer can do it like even my manager took my prototype and like started iterating on it uh and then uh you know like showing the pro showing something like this to the VP it depends on the stage of the pro product if you're like in the very beginning trying to define what the problem is you're working on maybe may not but like but like you know if you're on the solution phase and trying to figure out what to build like showing this stuff to the execs is like way more visceral and v visual like you can get a much better feel of the product than like you know something like this right so like a bunch of bullet points so that's why I think even if you write strategy documents uh I always try to include like some sort of prototype with it — I think this is like going to be the best thing to happen to the product design role and I think like the reason I think that is like think of the analogy to like a decade ago or like 15 years ago where like to be a product data analyst so much of your day was like really low-level queries or like each iteration of a PM having to like think about what they even wanted took so much of your time and then like mix panel amplitude and all those came out and then like a PM can go figure out what they even want you know before they come to you and like your job is like for really high stake stuff or like strategic or like the conversation's like way more far ahead by the time you're involved like way more baked I should say. Um — yeah, — you know what I mean? Like the feel like it's like it just helps us as PM come with like just better prepared for meetings with other people. — Yeah, I totally agree. And doing this stuff is like uh way more fun than like writing some document. I mean way more fun. So yeah, so I don't I don't think prototypes will completely replace clear thinking and writing thoughtful documents and concise and crisp documents, but I think it's definitely like another tool in your arsenal to kind of bring your stuff to life. So yeah. So that's my two demos. So let me uh stop sharing now. Which of you want to go next? — So I'm gonna start uh with my

initiatives uh database for a side project I'm working on. I'm showing a piece of software called Obsidian. And Obsidian is making like a resurgence because it's just like a really pretty way to visualize markdown files. It's like a really convenient, you know, human way to edit and read. Um, which means that like what you see here is one big markdown file actually and each of these links is a is another markdown file. Um, which means that I can open the same directory on my computer in cursor which then opens up a whole world of opportunities to play with it with AI. So what you're seeing in Obsidian just to prove what I'm saying, right? See this um folder structure? It's the same folder structure that you see here. And that allows me thanks to this uh free plugin to create a conbon. Um and so that whole little Trello looking thing you saw, right, is actually this file. This is the entire back end for what you saw. Every column is just this if you remember, you know, and like and each card is just one of these. Um you see these like double brackets, that's just like a link to like another file. That's what the purple ones were. And that's I keep that in this directory here. So there's like a ton of them in here. And so just to prove this again, if I say like learn enabling technologies like I'll just remove the word enabling. So right now it says, you know, and then I save it. And then this disappears here. So it's like literally the same thing. That's all there is. This is just like a nice way to view that. And that allows me to have a lot of fun. So I can tag uh I'm going to actually just tag the entire initiatives folder. I can put that into context in cursor. Uh, cursor is an ID, but I use it mostly actually not to code. I'm gonna put it in ask mode. Uh, and I can say something like, I need your help like taking a look at this. I need a second pair of eyes. Like, what do you think of this prioritization? Um, yeah. the order of operations and how I'm approaching this? And I'm even going to say, well, I want to give it even more context. So, I'm going to say, um, in light of opportunity assessment. There we go. This is like the more strategic overall uh document that I wrote that's all those initiatives kind of support. Um I can go even further and I can say ask me like one question at a time. Let's work through this together like be a partner through this and then we'll kind of work through it and I would love to hear your recommendations and thoughts and then I can select the model. Opus is it's great personality. What I love about cursor, by the way, you can see the reasoning. You can see exactly the tool calls. Like you just see it working. Um, every AI tool out there like to some degree is a little bit more or less transparent. Like cursor is like the most transparent. — Yeah, this is for your personal OS or is it for like a — this is a product I'm working on? This is like the real instance uh real initiatives. Bit of a mess. Um, this would be any product or any hat I wear as a PM. I'll have like my own like initiatives uh conbon initiatives database. That's kind of like for me to see the big picture. Uh and then cool now. But now what's cool is like I can have a conversation about it. You know this I can is all context for uh cloud opus. So — um it's you know I can keep going with this and say okay first question what specific outcome are you hope it validates? I want to understand it's like really poking me right it's like really challenging me which is great. Man that's a pretty deep question. Um I no I really hope that it it's a real value milestone as a customer, not just a technical milestone, right? I can keep going and like have this conversation and like Yeah. But that that's just like a one way I use it very frequently is just like especially if I got distracted and I need to like get back in context, right? — Yeah. What's really cool about this if it like gives me a bunch of suggestions at some point, I can actually put it back into agent mode and I can say, you know, agent mode makes so it can actually edit the markdown file and I can say something like let's say we had a longer conversation. I can say, hey, can okay, great. Can you implement those suggestions and, you know, rearrange the cards according to the prioritization we decided, right? Um, and then if I hit that, then it'll actually make the changes and they'll show up here as well. So there's like it's so many opportunities when you're like because it's all just a markdown file, a text file on my computer. — Do you feel like uh like I also interviewed Teresa and she has like a pretty fancy setup too, right, with her Obsidian and like I think one thing that kind of holds me back from doing this is like the OCD part of me like just wants to make sure like every MD file is like super clean and crisp and like just like relying on cloud to write all this stuff is just like I have no idea what the hell is actually in all these empty files, you know? Did you hear that uh problem or — so? That's actually why I personally really prefer cursor. It's like a very personal subjective preference. — Um so just we'll use a quick example. It's like uh can you reverse the order in each column in so I'll just tag this and have it take action and then I I'll show you this is I can't Peter I'm the same as you. Like I need to see the changes. I want to go over be in the loop. Um like cursor feels a lot more like pair programming or like pairing with AI and cloud code is a lot more like it goes off does things reports only the result and that's like could be pretty overwhelming. — Yeah. Sometimes it's the right tool but not for thinking and stuff and strategizing. I want to like be you know partnering with it. So like what's cool with cursor uh if it finishes at some point is it'll actually show me the diff and then I can accept or reject each change or edit the change on my own. — Yeah, that that makes a lot more sense. Uh yeah, like if you just change like a bunch of MD files without like the human supervising it, I'm just worried it'll turn like the whole repo will turn to crap at some point, — right? — Exactly. It just completely like misalign with — Yeah. So here, this is like my favorite. This is like the cure to my OCD. — Um I can it I can like Oh, no, wait. Don't delete this. Like keep this, right? Uh actually, no. This whole thing This is all a really bad change. Let's undo the whole thing, right? Uh, and I can go back and revert everything and I can just give it more specific instructions or I can just make the change myself. I personally like to your point I actually spend a lot of time in ask mode and don't let it make the changes and I have a conversation with it and then I make the changes myself. — Interesting. That that's a really good point. Like yeah whenever I v cursor I always just accept all but like I think it's important to actually review the diffs. Yeah. — Yeah. I mean I'm the same when it's coding but like when I'm when it's something that I can actually understand right then when it's plain English like this then yeah I review it. — Um — cool. — Yeah. The other thing I really like to do so this uh Obsidian Combound is kind of like for me as a PM it's like the order of like initiatives not development tasks. Uh so what I do is

actually I keep my development tasks um in a separate I've just like over the years converged on this and I keep doing this with AI. So I use linear for tracking development tasks and what I can do is still from cursor I can still like have a conversation and consult with it as a thinking partner about this. So what I'll do is I'll say uh this is a situation I had recently so I'll kind of like replay it is I just had a usability session where it was even unplanned and it was like we installed it on another computer that was like really similar setup to mine. and I thought it would go really smoothly. Uh, and as it was actually a complete disaster, a complete mess, like all these other so many frictions that I did not experience on my own computer, which like just highlighted for me how much more work there is. I'm a little shaken up by that. It's like, you know, not expected at all. And uh, I'm actually wondering how that might affect the um, rollout plan and like the, you know, what I need to do to release um, each milestone. Um, can you use the linear MCP? Uh, look at the MVP projects and yeah, like what do you think of the blockers that are holding back each phase? Do you think I'm being more too conservative, too leaning into risk too much? There we go. So, and yeah, it should I think I have the linear MCP enabled. Yeah, I do. I have it here. Oh, no. Interesting. That's It says it's erroring, but it's working. So you use linear to manage your own tickets for your for cyber cyod project or — yeah it just started like really simple and doing everything in the same combat and I was like getting confused because it was like different types of things. I realized that like even when it's just me I — I need to keep like product initiatives and development tasks diff separate it just helps me and so — Got it. See? Yeah. I try to keep it really simple and then I ran into So, okay, you can see actually like how many calls is I learned a lot about MCP just by using it like this. Yeah, you can see like that it's like it's a lot of calls. It's like very, you know, I don't know, clunky is the word, but it's not like this like smooth thing like before it was just reading a markdown file in context, right? — Yeah. That's a big complaint — people have with MCP, right? It just loads too much into context, you know? — Yeah. And that too. Yeah. Both in the setup and also along the way, right? It's like the experience is just like all this like reading. So now it's finally got to the same context. It finally pulled it out and like navigated my linear and it's given me uh yeah like a really harsh uh judgment that I need probably needed to hear, right? Um and I can have a conversation about it. I can say hold on hold wait okay let's like talk one question at a time. Let's think through it like you know let's align. Don't just judge me right away but you know it's this is an amazing starting point to have a conversation just like before. Okay. So, just to recap, so what did I actually do? It looked at all your linear tickets. — Yeah. So, what I had in linear, what I have is for this project, I kind of created a roll out plan like a staircase of like, okay, wait, if I wanted to just like, if I want to just give it to like three patient friendly friends, you know, like uh over Zoom to like just get going before I, you know, share this out like what do I need to do before I even give it to three patient friends? Like what are the things that are holding it back? And then if I want to give it to like you know the alumni of my course as like my beta group um you know for now I only think that only actually this doesn't make any sense but it's just an example right and if I want to like announce it on LinkedIn I should probably do these two things before that as a and so this is kind of like uh for me how I've organized the tasks and the priority and like and then I had cursor read that and so it now has everything here — as context in this conversation. — Got it. Okay. — Yeah. it like reproduced it to kind of like prove this to me, right? Like showed me the same thing. Bonus points. I love probably spent too much time doing this, but like I love reading the tool calls and reading the reasoning between them. But yeah, — got this. This is pretty cool, man. Yeah, I haven't thought about using like I always thought of Linear as kind of like a tool to collaborate with other people. I haven't thought about using it for like my own stuff. So maybe I should try that, too. — It's somewhere in the middle because you it's a really good platform for collaborating with other agents like coding agents in the cloud. They're a platform that's really leaned into that and like they get seats in linear that don't cost money. It's it's a cool new world. It's like it's the future. Yeah. — Onboard your agent teammates basically in one platform. Yeah. — Got it. That's great, man. Because I don't actually like working with other people. I don't like working with — Yeah. — They love working with linear MCP. — All right. A man, last but not least. — All right. So, I'm gonna uh kind of lean into a little bit of what Tal just showed as well and just maybe I'm going to start with a really quick highle diagram of just like what I'm about to kind of jump into from a workflow perspective. So, let me go ahead and share my entire screen here so we can all be looking at the same thing. So

you know, I know uh you had Trist on as well who showed kind of uh her personal OS system. And I think that what's super interesting about this like cloud code, coding agent, obsidian type of world is just that [clears throat] the the range of tasks that you can give uh these agents and so much of what you can do is basically determined by how you've customized your system. So, I'm going to kind of show uh sort of my own sort of version of this personal operating system, which consists of me just kind of like typing in unstructured thoughts like what should I work on today? And really the agent using a set of tools and skills that I've kind of vibe coded to help it figure out what it needs to what it thinks I should do and pull the right context to actually help me get those things done. So just to kind of maybe show the stack at a high level here, I'm going to show all of this in cloud code. Uh but as Tal showed, he uses cursor. Really most of these tools right now are largely uh not super differentiated. The main differentiations actually come from kind of UI UX of the experience and maybe a little bit of like the wrapping and uh the sort of like edges of the tech itself. Like you can definitely feel differences the more that you get into using the tools, but if you feel like you're, you know, you might be behind because you're not trying Cloud Code or you're not trying cursor, really it doesn't really matter that much as it's is just forming an opinion of what the difference of these tools are from firsthand use. uh you know that you can use them already today for coding tasks, but I'm going to try and show a couple more tasks that you can use these tools for like analysis like what Peter showed for AI Studio, some writing, brainstorming like what Tal just showed, and a couple more like ideas on top of that. Um we're going to talk about basically three parts of the stack here. There's the like action part of the stack, which is skills, sub aents, MCPS or APIs. again kind of fancy terminology for basically like running some code to do something. Uh those are going to actually interact with this concept of tasks which we're going to also look at in the same obsidian sort of flow as well as in linear and all of that actually interacts. You can see that there's a back and forth with basically this concept of memory or knowledge which is a little bit interchangeable at the moment in just this part of the stack. Memory is like think of this as your preferences or the agent's ability to recall or remember things that you've asked it to remember. And then knowledge I've defined here as your context. Things like your backlog, transcripts, samples that you've given it, or even goals. And so we're basically going to construct this end-to-end stack for a personal operating system, interact with a few MCPS along the way, and hopefully get some cool stuff done. So let's go ahead and start from scratch here. I'm going to open up Claude. And actually, what you can see at the top is I've got this system here called backlog MD. This is just a file that has just a couple of random notes in it. And I'm going to open it up in another window, too. You can see that it's open here in Obsidian. So, I also like to use Obsidian to help me view these markdown files a little bit easier. This is the same repo that I just opened in Cursor. It's open in Obsidian. You can see it has the same text here. So, a couple of things I just was like writing out while we were actually uh talking is I want to help summarize my uh Peter, you know, the Peter's episode here into like a blog post of like learnings that I might get and maybe want to share with others and then maybe work on a new AI product sense f post from our lesson one of the workshop that we ran a few weeks back. Uh and I that's kind of high priority for me. What's interesting is that I use Obsidian sort of as like my notes app now because they actually have a mobile app. So this is synced with my phone. So I can just type in stuff from my phone and it will sync with the cloud and show up in this Obsidian backlog too. So I've actually started moving more and more of my workflow to mobile uh kind of interestingly enough. Um so that's just sort of like the highle kind of note of Obsidian is this like easy way to interact with text. Uh, and you know, we I'm going to try and add some structure and organization here to make it a little bit easier to understand. All of this directory and the skills that I'm going to show as well, they're all open source. So, I'll share a repo uh with Peter to share out. So, you can actually pull this down and try it out yourself. And there are some tutorials in there for you to kind of try yourself as well. So, let's go ahead and start here with um actually just start with what should I work on today and see what happens. What's interesting is like I have some workflows built in here, but we're going to uncover what actually is going to happen. And what's interesting when you're working with these coding agents is they're pretty non-deterministic. Uh so, but you can see here that the first thing it does when I give it when I ask it something like what should I work on today? Uh is it actually uses this skill called morning planning. So, let's go ahead and just hit yes. And it's like a daily planning routine. It's a skill that I've kind of built into this personal operating system. Uh, all of this, by the way, is just in the readme file. So, if you want to open the readme, see what other capabilities are there, there's a lot. You can definitely check it out yourself, but just going to kind of go through a workflow of like what should I work on today. And you can see it's doing a few things here. One is it's finding meetings since my last sync. I'll kind of explain a little bit of how it's doing that, like finding meetings. Two, it's listing out tasks using this MCP that I've kind of built to help me list out tasks. And then it actually gives me the summary of meetings that it's found and asks me should I sync these to my knowledge. It gives me my focus for the day. So it's actually pulling some tasks from uh the things that I've basically been wanting to do. Uh and maybe some quick wins that it suggests I should do as well as finding housekeeping and kind of finding duplicate tasks here and asking me what should I work on first. So let's go ahead and kind of just go level one level deeper like how it's actually doing this. Um, so the first thing is it's pulling, you know, it it's actually saying like here's some new meetings that I found. This is actually using granola as well. So I use granola for, you know, basically transcribing. You can see I met with Ben a couple days ago. Uh, I have my personal workflow OS workflow uh, optimization. Uh, I have uh, the day one from uh, the boot camp that we ran a few weeks back. So these are all just transcripts from Granola. And I'm going to go ahead and show how I'm doing that in the first place. So Granola, interestingly enough, does not, as far as I know, have an MCP. But what's really fun is that there's people that are trying to hack together versions of quote unquote MCPs. It's funny, MCP has become this like catch-all term for basically anything that an agent might be hooked up to. So this is really just like a Python script that looks at your local Granola cache. So every time you're recording something with Granola, it stores those files locally, the notes and the transcripts. And what this is doing is creating a user kind of an agentfriendly way to wrap that cache that the local files that it's storing and access them with cursor or cloud in this MCP format. And what I've done is I've kind of taken that off of the shelf repo of this granola MCP and I've turned it into a skill where it when I say, you know, every day when I'm like, what should I work on today? It actually goes back and checks which new meetings or recordings have been added to that cache and suggests adding them to my knowledge bank here. So that's actually all it does is it runs it like a job. Uh so it runs it when I ask it to. So it can add that context into that knowledge folder and use that those transcripts for further tasks. Uh kind of like as a first step of like a granola MCP. — And uh to install the MCP you just like got cursor to install it basically like — Exactly. Yeah. — GitHub. — Yeah. And I can kind of show really quick workflow of like what that looks like as well. And what's what you know what [clears throat] we can kind of do and I'll maybe show the first couple of steps here. But when you whenever you just want cursor to install an MCP or like you know take a look at some type of code where it looks like it might be useful. There's a lot of really interesting packages and libraries getting shared around for cloud code and your agent harnesses. I usually just copy paste it in and I'm like uh how should I use this? this granola MCP? And I'm not going to actually implement it, but when you go into cloud, you can do shift tab and go into plan mode. So we can just get a plan back from cloud. And I know that, you know, because I've done this implementation once already, that it's going to say, oh, this is just going to hook up to my local granola cache. And it will just work out of the box. But what's interesting is like then you can ask on top of it hey I actually want to build a workflow on top of this which is like run every day automatically or run when I ask you to uh you know summarize what I should do today. So in my case I built a skill for basically use this MCP when I say what should I work on today. — Great. Yeah. Um so super straightforward. Uh, nothing really. Um, you know, I would say it's a and we can even take a closer look at that skill as well, just so we can understand it a little bit better. And so I'm going to open it up here, just so we can kind of understand what's going on in my morning planning skill. So this is um actually kind of useful I think for a lot of folks that may not have seen this one is that in your directory now, if you're building sub agents or skills, these get stored in this. cloud folder. So this doesn't, you know, usually this gets added to a git ignore. It's sort of like a hidden folder. And underneath that, you'll see that there's a new skills directory. When you ask an agent to build a skill, if we ask it to say turn this into a skill, this granola mcdp, it will build a skill in this folder. And a skill is really just a collection of markdown files. So let's go ahead and open this just to see what it looks like. Um, so I'm going to go and I'm just want to open a preview so that we can render this markdown and take a look at it really quick. So you can see this is my morning plan skill and I've just vibe coded it. Like this is just a markdown file that's like a set of instructions that claude code can follow based on sort of its decision criteria if it wants to open up you know a scale or not. And so it says check new meetings in granola. If there are meetings found, then show all of those to the user. Then get tasks and goals. List these tasks. Uh read the goals file. Uh and I'll kind of share what that is in a sec as well. And then present the morning overview. And so that's really what my morning sort of steps are here. Um so it's like, you know, basically show a user here's an example. And again, this is actually Claude building a skill for itself. Mhm. The other kind of interesting thing is the more context that you give Claude in a workflow, the better it's going to be at being able to actually be useful for you. So what you can do in this personal OS type of system is, you know, and really even if you're just think of uh like when I think about a personal OS

this is a productivity use case, but you can really think of this as like almost being an embodiment of you to some degree in markdown files. And so what I've done is like I can actually just add context and I actually do this in a lot of different places. Even when I'm working on code or if I'm a product manager, I want to encode my style of product management, what my quarterly goals are, what my big picture vision is, my roles and responsibilities on any initiative I'm working on, right? I want to bring my entire self to that project. And so I can set up a doc that says here's context on me. Here's what my job is. Uh here's what my goals are. my next sort of like 12-month horizon, maybe five star, fiveyear northstar horizon, what's my quarterly goals? Uh what are the metrics I'm trying to move? And so now whenever I ask uh Claude or I ask the agent to help me work on something, it has the context of those goals to help me align with whatever tasks I need to work on. So when it suggests things like, "Hey, here's what I should work on today," it's actually aligned with my goals. Uh, so it's not it's you know it and I've actually noticed when I've used it kind of on multiple days on end, it'll sometimes say, "Hey, it looks like you're drifting. You're working on projects that you that you're finding interesting, but are they really aligned with the goals that you care about and it'll kind of try to nudge me back on track with the things that I told it were important to me as well. " — How does it know to look at your goals file? You have some instructions. — Yeah, exactly. So, that was actually in the uh in the skill um in the morning planning skill. So it's uh you can see it checks uh it's a workflow basically to check for meetings but then get my tasks and get my goals as well that I have already — and the goals MD did you write that yourself or your cloud wrote it for you? — Yeah. So I uh wrote that in the sense that when you do when you run the setup script here it uh for the personal operating system it'll just ask you a few primer questions like basically the questions that you saw there of like what are your goals for the next few months? And so you can just add, you know, a little bit more information there, uh, and it'll populate that markdown file. So it just writes directly to the markdown file. But you can make it whatever you want. It's really just the concept of having a place for you to put information that kind of gets the agent to be aligned with what you care about. Um, and can kind of you can point the agent back to that file whenever you're like not sure what you should be doing. Like if you're a PM, another way to look at it is like you're building an agent of yourself to some degree. And so the Claude agent should always reference the like Aman agent or the Peter agent of like what would Peter do in this situation? What would Aman care about here? I'm trying to add more information around that. — Yeah. — Uh let me ask you one last question. How did you because there's like a lot of files here. Like what what's the step one to get started with this prop this whole thing? — Yeah. How about we do that? Uh actually let's uh let's zoom out a little bit here. So what I like to try to do whenever I'm working with any new directory actually is I just open Claude in the directory here. So I'm just in the same directory and I'll ask it like how should I get started here and I'll just kind of let Claude basically get the context from the readme and from all of the files here uh to try to understand what it should do. Um, and yeah, I think the idea is like basically like, you know, you actually want to minimize as many files as you need to have, but allow Claude to kind of like explain what's going on. So I can say, okay, here's what I should do. Start with my morning capture backlog, set my goals, start there. And so that's really, you know, if it's like a fresh directory, you can just ask what should I actually get started with and let kind of Claude guide you through the onboarding of the repo. — And did you say you made this repo public? — Yeah. So it's exactly so it's on GitHub uh here as well and you can see from the readme there's a quick start of you basically just clone the repo down again you can copy paste this into cloud uh literally copy paste the URL and then run this setup script which is just this file or yeah it's really just running this we can kind of show that and that is what kind of sets up the uh that's what sets up the golds um so it actually just runs through a setup process for getting your personal OS set up. — Awesome. — This is great. Yeah, we'll definitely put the link to that in the description. — My goal here is really to kind of build the lightest weight way for you to get started with trying to get organization around those directories I showed earlier. So like your knowledge, your tasks, and enough examples and MCPs like kind of useful MCPS for Claude to kind of get started here in the first place. So, why don't we just stop there and then let's do a wrap-up. — I have I could go on for this uh for like hours. — We usually go for this just like Yeah, exactly. We've got — Well, uh I'll tease. We've got evals in here. We've got uh linear in here. So, if you want more, just check out the repo. There's lots of really uh great tutorials in the repo itself for you to kind of pick up and try it out yourself. So, it's all very self-s served. Um, yeah. You don't even — Yeah. I just want to have like a AI ammon that will do whatever I want it to do. So, can you just build that for us? — Exactly. — Yeah. — Peter, I want you to use it so I can get your goals in there and I'll have Peter guiding me uh whenever I'm like going off the rails. Yeah. — I mean, I I can already like, you know, your if your goal is to become VP of product, it's very easy. Just update your link thing and just add VP to — Dude, how did I not think of that? You know, — I know, right? — Yeah. Let's kind of recap each of our takeaways of the demos that we just shared, right? So, um I'll start with mine. So, I I'm still in kind of like, you know, the corporate PM environment, right? So, like I think where AI has helped me the most is number one just writing these me memos and PRDS using my templates, helping me do research on whatever topic I'm doing and then uh kind of going back and forth with it both myself and also with my team through like tools like granola. So just writing manuals has been super helpful. Um and the second thing that I shared is just using it to prototype first. So like it's and this has actually been a pretty recent phenomenon for my team where we decided to just like duplicate whatever UI we have and do a bunch of prototyping, share it inside a team, share with users, share with execs, get some initial buying and then go off and write the document and like you know make the design and make the product real. So yeah, so those three things have been super helpful in my corporate PM life. Uh Tal do you want to go next? — Sure. Uh for me I was using AI as like a thinking partner for a while and I think a lot of people listening and you Peter and Aman you know and then what I started doing realizing is like what if I had my like more my productivity side like my personal operations. So uh that's the initiatives level and then I started well like I work you know I can experiment with like linear um for development tasks and then can I use AI as a thinking partner for those things as well and that and I have not just what I had before which is like product strategy customers company context for each conversation but now I also have like what order I think I should go in my prioritization like you know um my rollout stages and like what I'm gaining on each rollout you know and so on and that's just like another starting point for conversation and helping me think more clearly. — Yeah, I think I find I use AI a lot in my day-to-day for really two things. One is uh and both of them are actually things I'm bad at or don't enjoy actually like each of those are one in those two buckets. Um I'm not really good at uh like I find I bounce from projects a lot. I don't know if other people suffer from this too, but I definitely jump from, you know, and maybe it's like the PM life of like every 30 minutes you're context switching. So, I find AI is really helpful for me to stay on track in my day and help me stay focused on something over multiple weeks. So, it kind of just realigns me and make sure that I actually am capturing all of the right context, even my life context of like, hey, you said you were going to do this thing and it kind of holds me accountable, like it's an accountability partner, which I find personally very useful. And then the second thing is I actually really hate um writing emails and uh like kind of like the mundane tasks of like communication. And so I kind of treat it like a personalized autocomplete because it has all of the context of the things I'm working on, how I would respond to something. Uh one of the skills we have set up is like develop your own writing style. So I can actually write in my own writing style. Uh and it just does that. I don't have to tell it to do that. And so building these like workflows so that I can just do way more like write a proposal, write an email, and respond to someone else all in like five minutes in parallel is like been super helpful for me personally to just do more things

that uh I don't necessarily enjoy doing, but just be able to get them done, the things you have to get done in your day-to-day. There's kind of like a meta lesson from the three of us here. So Peter what you showed you had a clawed project with you know you very carefully designed that context in that project whether it's me and I'm you know doing essentially the same thing right with just like a different skin different environment I'm on adding like a few a bit more structure um he has a local MCP server he runs he's just like also the same concept but like what all three of us are doing is we basically created our own personal AI product that we're curating and iterating on And like every day we're like, "Oh, this could be better. " Like we're both the user and the PM for each of these products that we just screen shared. And like that in itself in addition to helping us, is like an incredible source of learning and building intuition is like so much I've learned just from like curating and iterating on my own thing. — Yeah. So if someone says MCP is all you need, you know, do you actually believe it or like eval? Well, have you actually done one for yourself? Like I think just being actually hands-on to solve problems in your daily life is more valuable than reading about it on a LinkedIn post. So like I think there's a lot of talk on X about like uh people building personal apps for like workouts and like all kinds of random stuff playing for kids, but I think the best personal app you can build is just something that understands you that you can kind of expand, you know, and I think everyone should do this. Yeah. — And by the way, none like almost no vibe coding. there's like you know like very little like when I'm on vibe coded he was v coding a markdown file that was like expressing how he wanted it to like yeah you know that like so much of the plumbing is already there like it's like a cloud project it's cursor it's cloud code like — it's we just built like really impressive AI products that really mean a lot and have like at least one highly retained user for each of us right like without vioding tons of you know HTML CSS whatever — it's like cloud projects or chat GPT projects but can way more capable because it's hooked up to more stuff and it's personalized to you for your workflows. So, — got it. Okay. So, basically uh learn how to manage the context and just like use MD files and use text. It's just text B basically. — Yeah. If I could summarize what I've learned about AI, like Aman's been in this field like way longer than me, but I'm more of a newcomer, but like it's if I could summarize it, it's all text files all the way down. — Definitely. Yeah. — Cool. I can think of a good amount of that. Yeah. All right, guys. Well, um, it's great to chat, uh, and hopefully all have a great 2026. — Yeah. Awesome. Thanks, Peter. — Thank you, Peter.