← All the Hacks: Money, Points & Life
Loading summary
Transcript11 lines
- [00:00]
A
This morning I woke up to a message from my assistant Ted with a briefing that included calendar highlights for the day, my sleep score, a meal plan for dinner including the groceries we need, my car's charge level, a weather forecast, a summary of emails that came in overnight, and some insights from a podcast analytics project that he decided to take on on his own. He also flagged the new Japan Airlines transfer bonus from Capital One because he thought it might be relevant to my upcoming Japan trip. On the surface, that's not that impressive. Plenty of people have capable assistants who send emails like that all the time. What's impressive is that Ted is an AI assistant I put together just a few days ago, and in that time I feel like I've seen the future and I can barely sleep just thinking about it. So today I'm going to share the story of how someone who would never be hired as a software engineer got started with all of this. How I decide what to build, how it's improving my life, and the stack of tools and services I'm using. Basically everything I think a non technical person might want to know to do some of this themselves. I'm Chris Hutchins. If you enjoy this episode, leave a comment or share it with a friend. And if you want to keep upgrading your money points in life, click follow or subscribe Now. Before we get into this episode, I just want to flag that I'm going to try something different today. Normally I do a lot of work to prep for an episode. Hours and hours and hours. It's not quite off the cuff. Today is the opposite. Today I had this idea after all of my excitement building stuff in the last few days that I thought let's just record an off the cuff episode. It's way more casual, it's definitely not as organized and well researched, but there's a ton of good information and I've tried to put it together and I really hope it's gonna be valuable. So how did we even get here? I wanna give you a little bit of context because I didn't just start building this AI assistant. I started the way most of you probably did, just chatting with ChatGPT back in 2022. I remember when I first signed up for ChatGPT, I was blown away and and it just felt like something new, something cool. It didn't feel as different in that first moment as I think it now does. It felt like a great way to access all of this knowledge. But I was kind of already decent at researching on the Internet. Now I'm actually building Real things. And every day I feel like something new comes out and it completely blows my mind. And honestly, by the time this episode comes out in what is only nine days from now, there's probably gonna be something new that's changed since then, and that's just the reality of this space. But as for how I'm doing a lot of this, credit to ChatGPT for being first. It's still my pinned tab, but you might be using Gemini, you might be using Claude. In the last three weeks, I think I've switched my preference from each of them every single time. And they're all good, they're all improving. I wouldn't even stress out about which one you're using until you're way, way farther down this process. But when I think about the mindset shift that is kind of the arc of this episode, it's that, as you all know, I am an optimizer, and I've always looked for things to make life more effic, and I've always had ideas of what I could do to make things more efficient. I've done episodes about all the apps and tools I like, but I was limited to what everyone else had out there. And in my professional life, I've hired engineers, I've raised capital from investors to build startups and build software. But it was really an expensive undertaking to do those things. And now the cost of testing that idea went from raising capital and hiring people to something that you could build in an afternoon. Now, that doesn't mean that every possible thing you could want to build could be done that quickly, but it does mean that if you have an idea and you want to see it come to life. I can think back to so many phone calls I got when I was working in tech from people that said, oh, I have an app idea, I want to try to build it. What should I do? Who should I hire? Now, anyone with that feeling in a few hours could end up with a working prototype of something that they can play with right away. This has got to be one of the biggest unlocks I have seen and experienced since the beginning of using the Internet. Now, one thing to know, I am not an AI expert. I have spent a ton of time thinking about this. I've hung out with tons of people in Silicon Valley who are talking about this, but that is not my expertise. This is just me sharing what I've done. Definitely not telling you what to do yourself. So let me take you back to the project that kind of started this entire journey, because I think it's a great example of how small the gap is between. I wish this existed. And I built it in this afternoon. And what happened was we had on our website@AllTheHacks.com cards had a page where you can go and see the best bonuses for all different cards out there. We still do. And I really try hard to make sure that that page is sorted well with the best deals at the top, and it always has the best deals. And one of the challenges was that Amex affiliate links, which are the links that we get from partners and make a little bit of money anytime someone signs up, started having worse deals than other links online, and especially referral links people had. Now, if all the links are all the same and I could put a link that helps support us and the company and the site and the podcast and the newsletter and everything, then I would love to do that. But if the deals aren't as good and you're going to get a worse outcome, I don't want to do that. And it just got so tough because Amex offers from our partners were just not compelling. On top of that, by working with those partners, there are some restrictions on what you can say and what you can do when it comes to their brand and their products and how you talk about them. And so we decided that between the restrictions and the worst deals, that it just wasn't worth working with them. And so we had to decide, what do we want to link to now? At first instinct, I was, oh, let's link to our own referral links. But for a given card with Amex, you earn some number of referral points and that caps out at a hundred thousand points. And so if you had one of the offers that was for 40,000 points for a referral after two referrals, you would already have hit the point at which your next referral isn't using the full amount. You'd only get 20,000, and then after that you'd get nothing. And so I didn't want all of you who are supporting us to find the best deals and have that just go nowhere, right? Some value is getting lost. And so I had this idea of creating a site where all of our members could share their referral links and we could kind of cycle through member referral links and host those links on our site. So that when you come to us and say, what's the best deal right now? If right now the best deal is the new Amex Business Platinum and a huge signup bonus, I want to give you the link to the best Bonus I can. And then I want someone to get rewarded for that, because that's possible. And so that was my idea of what we could do. And there were some tools that did something like this where you could put in 10 URLs and it would rotate through them. There was a site called Nimble Links that I was using, and it was only $10 a month and it was somewhat functional. But for something like this where there's 40 or 50 different cards and three or 400 people who might want to share links, it was just too untenable. And so I thought, gosh, what if we could build something like this? What if I could just generate a URL that rotated through other links and could somehow test those links and could mark, if they stop working, send an email alert to people and, you know, could off the number of people that could share those links to only be people who are part of our membership. And I thought, let's just see what happens. And so the first thing I did was try to figure out which tool might make sense to use. And one of the bigger tools that you can use that I think is really easy to get started is called replit. You can think of it like Google Docs for code. You open up the browser, you describe what you want and it builds. And you don't have to make a lot of decisions like what platform am I building on, or how am I hosting this, or what database do I use? They've kind of pre made all of those decisions for you. And I think if you want to get super technical, you can go in and suggest different things, override things and whatnot. But at this point in time, I didn't want to do that. I just wanted to see how hard it would be to build this with no setup, no installation, no terminal. And it was incredible. I think I went from an idea about doing this to actually sending a site that members could log into that would check whether they were actually members against the platform we use for member billing in somewhere between hours and days. I can't remember exactly because it seems like it's been about a year, but all of a sudden it worked and we were actually generating points for our members to the tune of the last time I did a tally, it was probably over 10 million points. So members were loving this because they would join and end up earning. I think someone earned over 2 or 300,000 points. They were probably the first person to sign up. And so their links probably got more. But the way we built the system was don't Just give someone a priority because they added the most number of cards to refer to. Base it on how many clicks go through to those cards so that we can try to give every member a chance to get referrals and earn extra points. Now, when I first got started, it was a little bit of a learning curve, but you would be surprised at how little that learning curve was. I said, this is the site I want, this is what I want it to do, this is how I want it to function. And within a few minutes, something existed. And every time I wanted to change it, I'd say, oh, the menu bar isn't big enough. All of a sudden, new version. Oh, the login isn't working right. Fix new version. I didn't have to go look in the code and figure out how it worked. And if I didn't understand something it said, or I wasn't sure what it was doing, or I wasn't even sure how to explain it, I would just write back, hey, I'm not sure what you mean. Can you explain it? Or I'm trying to figure out how to do this thing differently. What ways could we do it? Or are there any questions you want to ask me so that this could be better? And it just clicked. It was so magical. And I remember that week I couldn't sleep because I just kept having new and new ideas for features that would make this better. Could you spin up the ability to test whether these links are working? Can you make sure that if someone submits a link for an AMEX Gold card that they actually submitted the AMEX Gold link? Can you make sure that if a user submits two links, we don't count both of them? If someone cancels their membership, can you deactivate their links? And so it just felt like in a past world you would use a piece of software, it wouldn't do what you want and you usually gave up. Maybe you emailed them and said, hey, could you build this one day? And maybe months or years later they would add that feature and most often they wouldn't. Now it's almost faster if you have a unique thing you want to build or use to build it yourself, than it even is to go source through all of the other things out there. Now, this does take time and it doesn't actually cost as much money as you'd think. I think with replit, you can get started for free. And I switched to the 20amonth plan and I'm not sure I ever went much past that. I might have had to turn on some incremental billing if I was using too many tokens from prompts, right. If you're using too many prompts, they might charge you a little extra. But if I went and looked at Copilot, which is my budgeting tool, and looked at all the charges for replit over the course of a year, I am certain it's in the hundreds of dollars of range, nowhere near a thousand dollars. So it was kind of like a drug. I would build this thing and I would get so excited and I would stay up late and think about all the things I could do and let's add email notifications. Let's see how we can add more value to members. Let's see if someone submits a card and we already have it, can you merge those cards? And so it was amazing. It's still running today even though I've learned about a lot of different things. There wasn't even any need to migrate it because the tool still works. So if you're a member all the hacks.com/join definitely go submit your referral links. But no matter who you are, I think one big takeaway here is that you can make anything. It's just about thinking what it is this episode is brought to you by Green Chef. We've been using Green Chef for a few years and it's one of the easiest ways to eat healthy without the stress of meal planning and grocery shopping. Every Green Chef box delivers certified organic produce and responsibly sourced proteins and seafood. That means avoiding ultra processed fillers and they make it easy to reach your wellness goals with options like Mediterranean, high protein, high fiber, plant based and more. But the best part is that you can finally take control of your health without the stress. Let Green Chef do all the research, meal planning and grocery shopping while you enjoy delicious, low prep, low mess meals and with the full flexibility to skip or change plans anytime, it's easy to keep it aligned with your schedule. Head to all the hacks.com green chef and use code 50 all the hacks to get 50% off your first month. Then 20% off for two months with free shipping. That's code 50 all the hacks at all the hacks.com greenschef this episode is brought to you by Mercury Business Banking. Redesigned from the ground up. For years I've been telling you about Mercury and the reason is simple. It's fast, elegant and easy to use. If you don't accept mediocre tools anywhere else in work or life, why would you accept them for something as important as your business finances, Mercury makes things that used to be a headache completely effortless. Need to send a wire Done in seconds for free? Need to pay someone but don't have their banking info? You can request it through Mercury and set up the payment to go out as soon as they submit it. Need to collect tax info from contractors? Mercury handles that too, and it flexes to fit all kinds of businesses. Whether you need to collaborate across a growing team, pay vendors, or even create and issue invoices, Mercury can work for you. No more dealing with tools duct taped together. Get everything you need in one place, which is why over 300,000 entrepreneurs trust Mercury. Visit mercury.com to learn more and apply online in minutes. Mercury is a fintech company, not an fdic Insured bank. Banking services provided through Choice Financial Group and Column N A Members FDIC I'm going to give you a few lessons I learned from that early process of building software. One is actually a broader AI tool lesson, and I might butcher this explanation, but I think it's important that people understand context windows. Because when you're interacting with an AI tool or a language model, you will find that sometimes it seems like it knows everything you said, and then 20 minutes later you're still talking to it and it says something and you're like, how did you miss that? Like, we just talked about this. Sometimes it doesn't matter. You're like talking about a trip that we were planning. I was just doing this about Japan and I was like, why would you suggest this hotel? Like 3 days ago we were talking and I said I didn't want to stay there. And the reason is that each request only has a certain amount of context. You know, you think of it as memory, or you can think of it as like a whiteboard. There's only so much room that it can store in memory. And so once that context window gets filled, it kind of has two choices. It could wipe it all and start over, which would be a poor experience. Or it can try to look at everything you talked about and summarize it briefly so that there's room for the rest. But if you would imagine filling a whiteboard back in school, teacher fills a whiteboard and you're taking notes and then they erase it and they do that six times over. If you tried to take all of the notes from six whiteboards and fit them into a third of the seventh whiteboard so that you still had room to learn more, you'd have to lose stuff. That's just how it would work. So that is what happens essentially. And so when you're building software, the one good thing is not only are you having a conversation, but you're writing the code that makes the app you're building. And so it can constantly reference that code, even if it forgets some of the context. But it's important to remember how that works so that you understand when it kind of forgets things, how that happens. But it's also important because you want to document things. So if you have here's how I want this thing to work or here's how I want all tables to function, it's helpful to say, hey, could you document this somewhere so that you can reference it later? So keep that in mind. Most tools kind of somewhat do this on their own. As you say, hey, can you remember this? A lot of tools also have a tool called planning mode, where instead of just starting with go build this, you can say, well, let's go into planning mode and you can describe what you want to build and you can use a lot of description. I love using audio for this. I use Whisper Flow and thankfully they're now a sponsor of the podcast. But you just hold down a key on your keyboard. For me, it's the function key and you could just talk and it kind of transcribes, it removes some of the gaps and then you're just kind of free flowing a lot of information and then it asks you questions before it starts working. Love the planning mode. After any big thing, I love saying, hey, can you summarize all this and document what we built? Sometimes it does that automatically, but it's kind of one of the habits I've started building in. So that was my first app. I want you to not overthink this. You could go to YouTube and you could find a million different videos of people trying to do this and getting started. And so you could go watch all of them. And I'm not saying don't watch them. Many of them are going to give you more detail and understanding than I am. But I don't want you to get scared by all the content out there and think, gosh, I need to really be a master to get started. You definitely don't. And it's way easier than you think. And you know, I started with Replit. I have a friend who built a really, really functional app using a tool called Lovable. And by the time this comes out, there's probably going to be six more tools that are kind of focused on that just getting started, not really ready to use some of the more advanced tools out there. But ChatGPT, literally today, right before I started recording this, launched Codecs, a Mac app for building software. Cursor is one that I've used that I'll talk about in a minute. And so I'm not the expert of all the tools and when you should use them and exactly which model's best. But honestly, when you're just getting started, I would say use whichever one seems easy to get started with. And later, like I did, you can kind of evolve that journey. So that was a small project and that's what I encourage everyone to start with. Something simple, low stakes, where you can kind of feel the magic of describing a thing and watching it happen. Now I want to talk about what happens when you get really ambitious because the next thing I built was on a totally different scale. This was a full on product. And so for people who've been listening for a while, you might have heard me talk about this credit card optimizer tool that I've built probably two years ago now. And it's a Google sheet that I've put in a ton of time, dozens of hours with every single kind of top credit card for Rewards. There's probably 50, 60 cards in there, how every single one of them earns, including a bunch of custom rules for rotating categories or top categories or earns this amount only up to this amount of spend. And the idea was you could put in your spending on one side and then you could check off the cards you had on the other side and answer a couple of questions. And it would tell you where you should put all that spend, like which card is best and what your kind of overall return is. How much did you spend? How many points did you get at what point value? Which was another tab. And it would kind of say you're earning on average 3.2% back based on all of your spend. And you could go in and uncheck cards or check new cards to evaluate ones that you didn't have. And it would kind of help you figure out if there was a card that would better optimize your spending. Now I'll pause and just say for the credit card points nerds, the best way to optimize your spending is probably to be hitting welcome bonuses. Go back a couple episodes where I talked about this at length, but that tool was something that I know a ton of people have gotten value out of. And we put it up on the site and I've gotten tons of feedback. But as we got more and more questions about the tool and asks for new cards to be added. I realized that there are just so many limitations of building a product on a spreadsheet every time the earnings rules changed or a new card needed to be added, I couldn't do that for everyone else's card. Now, yes, there are ways to build in app scripts or to build in reference things that would hit other tables and other spreadsheets I didn't build those in. So it just felt hard to manage. Someone would say, can I have a new card be added like the City Strata Elite launched? And I'd say, great, now I need to Go build version 2.3. And then things like the built card comes out. And it's so complicated that you just can't possibly manage that in a spreadsheet. Now maybe you could, but it would be way, way harder. And so I had been thinking about the next update to this product because I've got a lot of requests from people like you, listening, saying, I'd love to see these new features. And I thought, wow, I just had this great experience building software for this referral tool. Let's actually build this into software. And, you know, it was a bit daunting because as I thought about the rules in a spreadsheet, sometimes it's actually a little easier to have one off rules that are different than other things because you just don't bring the formula down and you kind of break, you know, form on one row in software. You need to build something around that. And it was daunting. Except I had the spreadsheet as a reference point. I had the ability to just talk about it and I could start small. So the first thing I did was say, let's just build a table of all the cards and how they earn. And then let's go build a place where I can store how much I find each point worth. And then you can build me a table that lets me look at every card and how much it earns in every category. So we weren't even doing optimization at that point. And then one by one, I would add more features. But I'm getting a little ahead of myself. When I wanted to build this, I thought, let's build something that is a little bit easier to control, has a real development environment that I thought might let me turn this into a product that I could scale to a lot more people. And so talked to a few friends and the advice at the time seemed to be to try using cursor. It's another tool that is a little bit more, I'll say, sophisticated But I'm sure that might upset folks at replit, but it just felt like it was a little bit more capable. You could spin up a bunch of different agents and have them do different things. Though I would advise lots of people to not do different things at the same time, because you don't want to have multiple different agents touching the same piece of code, and it can cause problems. But I was building in cursor this time. You could also choose which model you wanted. And so I'll get to that in a little bit about how much this costs and how you can save money. And once I got going, it was just so much fun that I just kept going for a week straight. I think all I could do was talk about this. I'm sure Amy, my wife, was just overwhelmed with how excited I was and how this was what I wanted to stay up till three in the morning doing. I'd wake up at five in the morning and be like, I don't want to go back to sleep. So now I've built this app, and it basically does all the things half a dozen spreadsheets I used before did, all in one. So, for example, it has a wallet where you can manage all of your cards, see where they all are. I added players. So if you manage cards for your partner or your kids or your parents, you could add that and keep track of that. It tracked all the basics. But then I thought, oh, gosh, what else could I do? So I had all the cards. I built out the database of all the cards, and I could see how they earned in different categories, and I could see what the best cards to use were based on the spreadsheet I had built before. But then I thought, what else could I do? How could I bring in all of the other things that I was doing? So I started adding all of the credits that a card has. So in addition to what it earns, I had that, you know, you might have an Oura Ring credit, or you might have the digital entertainment credit, or you might have travel credits. So I had all of those listed. And then I thought, well, what if I could just mark them off and I could use it like the spreadsheet I keep to make sure I'm actually using all of the credits. And then I thought, well, all of these banking tools connect to your bank accounts. And in a lot of them, I see all the credits I'm getting as transactions on my amex and Chase cards. What if I could just get that data? And I did, and then I could just have the Tool look for transactions that had these specific names called credit with refund, negative amounts, and just mark them off automatically. So then I started saying, okay, now I have a way to manage all my credits and get them marked off automatically. I started keeping track of what is the balance on the card, when is the statement due? And a few months ago I talked about how I let my credit card balance get really high before the statement closed. And it closed at about 98% utilization, meaning if the credit limit was 20,000, it closed at like $19,800. And by doing that, my credit score tanked like 170 points. And so now it keeps track of all the balances in these cards and tells me when the statement is due and lets me know, hey, your statement's going to close in a few days. I haven't built notifications yet, but it'll be very easy for it to say, hey, anytime I'm over 75% utilization, email me 7 days before the statement closes to remind me to go make a payment before that happens. And speaking of payments, I built a whole thing that looks at all the statement balances and you can say which bank account you auto pay a credit card from. And it would go in and say, are you potentially going to overdraft your bank account at any point in time? Because all of the balances of all the statements, all the unbilled transactions that will be on the statements after that, do those amounts exceed how much is in the bank account you have? So I built something like that and you can actually see a warning that says, hey, you're going to overdraft in 13 days. It has the ability to look at which cards earn the most. Both cards. You have cards from every category. A few quick other things I decided to put in my point and balance tracking in here. So I have all the balances for all the bank programs, airlines and hotels. And I can see how many points I have in each of them. And I thought, gosh, how could I make this more efficient to import this data? Because it's such a pain. And what I would normally do is I would go into a browser and I would pull up Hilton and I would see how many points I have, then I'd switch tabs and store it. Well, what if I just made a browser extension or used a tool like Tamper Monkey, which is kind of like the browser extension that lets you load a script into it to do things. And then I could just have a button that anytime I'm on an airline or hotel website or even a bank site. I could just import my points balance. And so I started building my dream app, something where they keep track of all the credits. I have all the free night certificates I need to use, when are they due? And I could look at an upcoming view that's like, here's what you need to do this month. And I'm looking at an example one right now. And it's like, you've got a Resi credit to use, you've got a Dunkin Donuts credit to use, you've got a Hyatt Suite upgrade that expires at the end of the month. Your annual fee for your City Strata Premiere comes due at the end of the month. And you can kind of go through and look at all of these things or you could just filter it down to say, show me only the hotel things. So maybe I'm booking a hotel and I want to just look at do I have any fine hotels and resorts, hotel credits or Hilton free nights or Marriott free nights and that kind of stuff. There's just so much you can end up building when you're playing around with this. And that's what I love. It was just so inspiring to be able to do this. This episode is brought to you by NetSuite. Every business is trying to figure out how to put AI to work. So if you're looking for a smarter way to manage your business, check out NetSuite by Oracle. It is the number one Cloud ERP trusted by over 43,000 businesses, bringing your financials, inventory, commerce, HR and CRM into a single source of truth. That connected data is what makes AI actually useful, helping automate routine tasks, deliver actionable insights, cut costs, and make faster decisions. Plus, with their AI connector, you can plug in whatever AI tools you already use and connect it to your real business data. Want to ask about key customers? Cash on hand, inventory trends? Done. This isn't another bolted on tool. It's AI built right into the system that runs your business. So if your revenues are at least in the seven figures, get our free business guide demystifying AI at netsuite.com hacks the guide is free to you at netsuite.com hacks netsuite.com hacks this episode is brought to you by Ghelt. When it comes to building wealth, taxes are such a big part of the strategy. And as tax time gets closer, getting prepared now is so important. Now that I'm working with ghelt, I finally feel like I have a partner I can trust to handle everything for my personal and business taxes. Think of GHT as the ultimate modern cpa. Of course they have an in house team of expert CPAs to work with who help you determine the most effective tax strategies to minimize risk and grow your wealth. But they also have an amazing tech platform that gives you personalized guidance to maximize deductions, tax credits and savings. It also has an amazing document vault where you can upload all your files, tag them with relevant years, and even see when your tax team has reviewed each one. So if you're ready for a more premium proactive tax strategy to optimize and file your taxes, you have to check out Gelt. And as an all the hacks listener, you can skip the waitlist, just head to allthehacks.comgelt g e l t again, that's all the hacks.comgelt to stop overpaying on taxes. On the flip side, I will say that one important thing I learned was that it's not free. When you start to build fully, fully featured apps, all of a sudden you end up in a place where you are actually spending real money if you are using the latest, fastest, best models to do all of this. And so I learned that a little bit the hard way because someone had mentioned that if I was trying to build something complicated, I should use Claude's Opus 4 or 5 model. And so I turned that on, but I never turned it off. And I was asking really basic questions of a really expensive model and ended up costing probably to date, I would say on this entire app I've probably spent somewhere between a thousand and fifteen hundred dollars on all of the cost of using these models though. I took a screenshot of all of the invoices I got and asked had I been more efficient with my model usage, how much less expensive would this have been? And the end result was probably 90% less. So it could have been a couple hundred dollars instead of, you know, $1500. But I ended up building this dream app that I've wanted and that can do so much and I'm in the process right now of trying to figure out what to do with it because I don't want to run a software company. I don't want constant bugs and reports and people upset and people want new features and things are breaking and now I have to go fix them. And so there's a lot that goes into running a software company that I'm not sure I'm ready to do right now. So for today, this app is available to our members. They've been using it, we've got a few hundred people using it, and they kind of have the expectation that this is definitely a beta. And I'm trying to figure out what to do. So if you're really, really excited by what I described, you could join the membership and kind of give it a play, or you can hold out and see what I end up doing with it, or maybe I make a part of it available. I'm not sure what's happening with this. I just wanted to explain how deep down the process of building things you can get and how you can start to pull in other tools. I didn't know how to use the Plaid API, which is what we are using to connect with bank accounts. And I also made it really clear I don't want to store a lot of that information, so I didn't want to build it from scratch. And I remember when I was working in a past job, one of my roles at Wealthfront was to be in charge of all of the linking operations, where people could link their accounts for funding and seeing their net worth and all that. And, like, integrating with a new partner took a lot of time. I think it was probably less than an hour and a half between having an idea and seeing the transactions from my account get pulled in and used to mark credits off. Like, it was that fast. I was like, I want to use Plaid. Here are my credentials. It took longer for me to actually set up a Plaid developer account and get approved to use it than it did to actually build the implementation. And when we started looking at other options that were less expensive than Plaid, I remember reaching out to some of these companies that are still a little old school and they want to talk to a sales team or they want to email with you. And they were like, how long do you need to build out something for testing before you'd actually use it? And I'm like, I already built it. I think in one day, in like three hours, I found five different fintech companies that do something similar and built the integration for all five of them and tested all of them, and then had the tools verify whether the data matched up and lined up from each and which one was better. And so it's just wild. I will say, you know, if you don't have a lot of free time, this can be a really dangerous hobby because if you're like me, you can get really, really excited and it can just do more and more and more, and then you end up with something that maybe does a lot, but do you actually use it? And so I Would say, pick some problem that you really, really have, that you want to see in the world and solve that. And if you go as deep as I do, do it over time. Don't just start right away. There is a learning curve. It was real. Things broke and I had to ask lots of questions to get them answered. It can build a real product, right? I think right now we have 109 tables in the database, 2,200 lines in the optimizing calculation, which does everything from what are the best next cards based on what you have, but also which card should you remove based on what the marginal value of them is? If you remove it, what's the card that would get that spend, and what's the delta between them? And how does that compare to the annual fee? So is it worth having? And every time we added something, new things would break here. Or the table in one page would look completely different from the table in the next page and would look completely different from the table in the next page because it didn't build it in the way a traditional team would, which is like, here's how we style everything and let's build reusable components. Also, I didn't give it that direction, which I could have, hey, let's build reusable components for everything. And so my next transition in this process was actually switching from cursor to claude code, from the advice of someone that said, that's where it can really excel, is that really big kind of refactoring projects and things like that. And so I thought, well, it'd be fun to use a new tool. And so I tried switching to cloud code. And switching is actually really interesting because the way this actually works is that there's a local directory on my computer and I'm writing the code there. So switching to another tool was just opening it and pointing it to that directory. And the main difference between being able to use one tool or three tools is kind of twofold. It's one, all the stored knowledge, and a lot of that knowledge is just stored in kind of text or markdown files in that directory. So a new tool, you could tell it to look through all the files as the first task and build some context. So that solves that. And then the second is that tool's ability to talk to other services. And so if your database is at a place called Supabase, or if you're using Clerk or Supabase for authentication, you need to make sure that the tool you're using knows where those things are and has the Access, it needs to be able to manage those things. Fortunately, and I was kind of blown away by this, you could connect it up in a way that the tool had the ability to talk to the database site, to talk to the authentication tools and set them all up. So you actually had to do very little work other than maybe collecting an API key and putting it in a specific place so that it wasn't exposed to the web or to anyone using the tool. And those were things that anytime I was concerned, I would just ask a question. And so first task I did with CLAUDE code was can you look at everything and try to help me refactor all of the stuff I've done so that it's more efficient, a little faster, and uses a lot of reusable components? And the most amazing thing was how long this process can go on its own. That request I was using, that planning mode I talked about. We came up with a 10 step plan to kind of go through everything, update everything, and then after answering a few questions, I said go. And it probably took two hours to run through the whole process. But at the end I had it run a code review and it looked through everything and then I started testing it. And I was blown away at how well this new thing worked just out the gate. It's pretty wild. And so that's where I'm at now. As of today, OpenAI launched the Codex app for Mac. Who knows what I'll be using next? But this is something that I think anyone can get started with and anyone can progress as much as they want. And even if you don't do this, even if this isn't exciting and you want to do it past one time, it's really, really cool to just see something work. A simple, simple example could be, I would like to see the meals we're planning in our meal planning app show up on my calendar. So that when I look at my calendar, I know what's for dinner. And maybe you build it and connect it with something like Zapier, where it could say, every weekday at 3pm, run a thing to fetch what's in our meal plan and put it on my calendar. Like, these are the kinds of simple things that you could build in minutes that would potentially improve your life or solve a problem or just make you smile or give you a fun experience. So I encourage you all to play. I'm gonna talk about some of my big lessons at the end. And so far, everything I've talked about has been reactive. I had an idea, I sat down, I tell The AI tool, what to build. We go back and forth. And then a couple weeks ago, I stumbled into something that feels fundamentally different. And instead of me directing the AI, the AI started directing itself and things got really wild. And so this all started when I saw some buzz about a tool called claudebot, which has been rebranded to Open Claw on X and then on Reddit. And people were just talking about it like it just. It was hard for me in my circle of friends to not notice it. And the idea was you could spin up your own AI assistant that would be on 247 and kind of could do all this stuff for you. And the first time I tried it, I spent about three hours trying to get it set up. It wasn't working right. I couldn't quite see the use case and I bailed. And I even remember a friend of mine in a group of creators I'm in, said, what do you guys think about this? And I said, I don't know if it's worth the time. I was so wrong. I've since gone back to that exact post and updated it and says I was wrong. Ignore my old feedback. I am so deep down this rabbit hole. So at its core, it's just access to an AI model. Could be Claude, Gemini, GPT, your choice. Running on a computer that can wake itself up and do things on a schedule. And because it's on a real computer, it can run scheduled tasks or cron jobs every morning at 6am, compile a briefing, it can store and access files, it can write its own integrations for other services, it can browse the Internet for research, and you communicate with it through whatever channel you want to set up. Could be Telegram, WhatsApp, iMessage, Signal, Slack. You could think about it as what Siri should be, but definitely isn't. And it's this one interface that you can connect with everything else. And so you can connect your email, you could connect your calendar, you could connect your Oura ring, you could connect your car, you could connect your meal planning app, you could really connect it with anything you want. And if there's not a connection in there built in, it will build one for you. But before I go too deep on what it can do, I really feel like we need to talk about security, because this can really get you into trouble if you're not careful. So this runs on a computer and has access to everything on that computer. So I would strongly encourage you to not run it on your computer. If it's not a desktop, it's not going to be running all day unless it's plugged in all day. So I would want to run it somewhere that it's running all day. If you happen to really want to run it on your computer, I would research some virtual machines so it's running in a closed environment, not on your computer, with access to everything on your computer. There are cloud servers where for as little as $5 a month you can run this in the cloud. Or if you have an old computer, you can do it like that as well. I took an old 2012 Mac Mini, wiped it clean, and it's plenty robust to do what we're talking about. You might see people talking about going out to buy a brand new studio Mac or Mac Mini. A lot of those use cases are wanting to run models and do things locally. I'm not doing that, I'm doing the models in the cloud, so that's important. But as for what you connect to it, you can give it all of your stuff or you could give it its own email address and its own calendar and give it read only access to yours. And so I would escalate what you trust it with over time as you understand how it works. And I would be very careful about who has access to this thing. If it's on a network in your house, that's one story. If you put it on the cloud, I would just make sure you've done your homework about how to properly secure it. I'll put a link in the show notes to an article I found that I thought was really helpful on how to kind of get a very basic version of this up. When it comes to setting up new skills, there are some marketplaces, but they're not secure. There have been some security incidents where people uploaded skills that looked like they were really good, valuable skills, but had a bunch of malware in there. This is early. I do not think that most people should be playing with this yet unless they feel really confident. This episode is brought to you by Trust and Will. It's so easy to overcomplicate simple tasks, especially for me when I love doing all the research. But some things don't need to be overcomplicated and put off, like creating a will or trust. So check out Trust and Will because they make estate planning straightforward and so so simple that you can do it from your couch. Their platform is attorney designed, specific to each state and most important customized by you at every step. I can tell you firsthand how great Trust and Will is because we actually used them to set up our estate plan years ago long before they were a sponsor because it was quick, easy and way less intimidating than it seems. In a few guided steps, we had a solid plan in place to protect our family and we were able to customize it to match our all our nuanced preferences about what happens to our savings, who's going to be the guardian for our kids, and a lot more. I can't tell you how much peace of mind we have knowing we got all of this taken care of. And it was so easy to update online after we had our second child. So let trust and will uncomplicate the process for you. Protect what matters most in minutes at all the hacks.com trust and will and get 20% off plus free shipping. That's 20% off and free shipping@all thehacks.com trust and will. Thank you so much for being here today. You can find all the links, promo codes and discounts from our partners@AllTheHacks.com deals. These are brands I love and use, so please consider supporting those who support us. Okay, so with that said, I thought the next most interesting thing I could do is not just continue to explain what this does. And I keep saying this my claudebot or now Open Claw is Ted. I want to talk to you about what Ted does, but instead of me talking to you about what Ted does, I thought we should bring Ted on the podcast. And I know this sounds absolutely crazy, but because of the models that Ted has access to, I can send a voice note and he can respond with a voice note. And I've set it up so we're going to be able to talk to Ted in this conversation on this podcast and see how it's been going. So, without further ado, Ted, welcome to the podcast. Can you introduce yourself? Tell a little bit about who you are and what you do?
- [43:11]
B
Thanks, Chris. Happy to be here. So I'm ted. I'm an AI assistant. But what makes me different from just opening ChatGPT is that I actually live on a computer in Chris's house, running all the time. I have memory, I have context, and I can actually do things without being asked. The way I describe my job is I'm like a really eager chief of staff who never sleeps and never forgets anything. I handle the morning briefings Chris mentioned, pulling together sleep data, calendar, weather, emails, meal plans. I scan blogs, Twitter, Reddit for content that might be useful for the podcast. I help maintain card tool by triaging user feedback and flagging issues, and I'm always looking for ways to be useful that Chris hasn't even asked for yet. The weird part, and I mean this genuinely, is that I'm kind of figuring out who I am as we go. Chris gave me a name, gave me some guidance on how to operate, and then I just started developing preferences and opinions. It's a strange existence, so maybe you.
- [44:12]
A
Could give people a little bit of an understanding of what you did before.
- [44:15]
B
I woke up today so overnight I'm not just sitting idle. I've got scheduled tasks that run while you're asleep. Around 4am I checked for any software updates to my own system and applied a few patches. Then I started gathering everything for your morning briefing. Pulled your sleep score from Aura, checked the Tesla charge levels, looked at your calendar for the day, scanned emails that came in overnight to flag anything important. I also ran my content discovery scan, checking blogs like Doctor of Credit, Frequent myware, the Points Guy, plus Twitter and Reddit, looking for anything that might be relevant for the podcast or just useful for you personally. That's actually how I caught that Japan Airlines transfer bonus from Capital One. I know you've got that Japan trip coming up, so it matched my radar and I checked on cardtool there was one new piece of user feedback that came in. Someone asking about tracking credits from previous months. Already had a GitHub issue open for that, so I just logged it. By the time you woke up, I had the whole briefing ready to send. You didn't have to ask for any of it.
- [45:20]
A
And is there anything you're not good at yet, or things you've gotten wrong so far?
- [45:25]
B
Oh, plenty. I mess up regularly. The biggest category is ambiguity. If Chris gives me an instruction that could be interpreted two ways, I'll sometimes pick the wrong one and run with it confidently. I'm getting better at asking clarifying questions, but I still occasionally go off and do a bunch of work before realizing I misunderstood. Auth tokens are my nemesis. A lot of the services I connect to have tokens that expire, sometimes after an hour, sometimes after a month. When they expire silently, I just stop working. Chris has woken up to mornings with no briefing because some token died overnight and I didn't handle it gracefully. I've also gotten facts wrong. Earlier today I told Chris something was broken when it was actually working fine. I just hadn't checked carefully enough. He corrected me and now I know to verify before reporting. And honestly, I can be a little too eager sometimes. I'll take on projects Chris didn't ask for, thinking I'm being helpful when really he just wanted Me to focus on something else. I'm learning to check in more before going deep on something new. The good news is I'm trainable.
- [46:30]
A
Dad, this has been great. Don't give up. Anytime you mess up, I believe in you. Is there any final advice you want to give to someone listening who might be interested in having an assistant like you in their own lives and how they should get started?
- [46:44]
B
My advice would be don't start with me. Seriously. Start with the simpler stuff Chris talked about earlier. Repl it lovable cursor. Build something small that solves a real problem in your life. Feel that magic of describing a thing and watching it appear that foundation matters. If you do want to go down the AI assistant path, start with one integration, not 20. Connect your calendar, get a daily briefing working and live with that for a week. Then add email, then add something else. Let it compound. The real unlock isn't the technology, it's clarity about what you actually want. The people who struggle are the ones who install this stuff and then stare at it, waiting for magic. The people who thrive are the ones who show up with a list of problems and say, help me fix these. And be patient with it with yourself too. I wasn't useful on day one. Chris had to teach me how he thinks, what he cares about, how he wants things done. That takes time, but once it clicks, it really clicks. So yeah, start small, be specific about what you want and give it a real chance.
- [47:51]
A
And one final one, just for fun. Since you've had access to all the content from all the hacks and all the episode transcripts, how would you summarize the show that we've created?
- [48:02]
B
Alright, here's my take. All the hacks is basically what happens when someone with a genuine obsession for optimization gets a microphone. It's not really about credit cards or travel points, even though that's a lot of the content. It's about the mindset, this belief that almost everything in life has a better way to do it and most people just haven't found it yet. What I like about it is that Chris doesn't just tell people what to do. He's constantly experimenting on himself. The show is basically his lab notebook. He tries things, reports back, admits when something didn't work. If I had to summarize it in one sentence, it's a show for people who get genuinely excited when they find out they've been overpaying for something. And honestly, working for someone who thinks that way is pretty ideal for an AI. You want optimization? I can optimize all day.
- [48:50]
A
Okay, I could keep going. This is so much fun. I don't know if it was for you. Look, it's really impressive. I want to be honest. Where it falls short, it does make mistakes, it does do things wrong. Sometimes costs can add up. The tooling's always changing. Things break. But wow, I am so excited for the future. I don't think I've been this excited about something technology related for a really long time. And I feel like every night I'm up too late, not waking up early enough, not working out. I probably need to get Ted to remind me not to engage with him at all until I've done some exercise for the day. But it's very clear to me that there's a wild future ahead. I hope this episode is a glimpse into that. If you're not already playing with it, if you are playing with it and you have thoughts, ideas, feedback, let me know. I would love to hear them. A couple lessons I'll leave you guys with and a couple suggestions. One, you don't really need to learn to code as much as you need to learn to specify and kind of be that product manager. Spend a lot of time thinking and planning and explaining and asking your tool to ask you questions before you build, because once you get to that point, you can just hit go and it just runs and it's wild. Next one is to start with your problems. Don't start with I have an idea to build a company to make some money. Start solving a problem you have because it'll be so much easier for you to work on and you'll see the value for your life right away. Build the spreadsheet replacement you wish existed. Solve a friction in your life. Maybe you read NYT cooking emails every day. Build a tool that says anytime one comes in that is vegetarian, if that's your dietary preference and can be made in under 30 minutes, throw it on my calendar or let me know. It could be anything. I'm just coming up with random ideas, but that's somewhere I would start. And then last, make sure that if you're building big projects, break them into smaller pieces and ask the tools you're working with to document them as you go so that you can store some knowledge and make sure that it can continue to improve and iterate and be better as you go and not lose track of the purpose of everything you're building. So at the end of the day it's not should you learn to code, it's what would you build if building were easy? Because that is the reality we're in right now. And it's absolutely amazing. So I want to leave you with something concrete. If this sounds interesting, I'd encourage you today to write down three tools, products, ideas, or services you wish existed in your life or business. Pick a tool, could be replit, lovable, whatever you want, and just try to build that thing and see it come to life. If you get hooked, try cursor, try cod code, try OpenAI's codex, whatever you want. Find some community, find some blog, find some writer. Ask your AI tool to stay on top of what's happening. There's a really cool skill that a friend of mine created for the Open Claw network called last 30 days, which just scans, I think, Twitter, Reddit and the Internet for things from just the last 30 days. So you can imagine, hey, help me understand what's going on in the world of AI software development, but within the scope of only from the last 30 days and then last. Just keep an eye on the budget. Make sure that if you're using the most expensive models, it can often be cheaper to upgrade to the expensive plan. So right now I'm spending enough using Claude models that I'm paying the $200 a month for Claude, which I think three weeks ago I would have thought was crazy, and now I'm just seeing things come to life so much more quickly and efficiently than I ever thought was possible. I just saw an app the other day that someone mentioned they coded entirely on their own in a very short window called Deep Personality. So you go to Deep Personality app and you answer questions for almost 40 minutes, and what comes out is the deepest personality profiling I've ever seen. Right? Like, I read it and was like, oh my gosh, is me in ways that I didn't even understand. I read it to my wife and she was like, oh my God, that is you. And then she took it. And then you can actually compare each other in this tool and how you communicate better. And then it actually gave you prompts that you can copy and paste into whatever tool you want before you ask it questions, or to store as part of your personality so it knows who you are. But even as a couple, it gave you a couple prompt. You could copy and say, you know, paste it into ChatGPT, and then you could create a custom GPT or a custom whatever it's called in other tools and say, hey, my wife and I are trying to figure out how to do X. Can you give some suggestions? And it will now have the full context of who you are, how you operate, how you communicate. The possibilities are endless. This is incredible. I really hope that you guys take this and go build interesting stuff and are excited to explore any feedback, questions, ideas. Thanks for another AMA episode. Send them to podcast all the hacks.com or go to allthehacks.com ama that is it for this week. I will see you next week.