How I run autonomous coding agents from my phone with OpenAI Symphony + Linear

1. 0:00 – 2:24

   Intro

   1. CVClaire Vo0:00

      This is my favorite positive outcome of AI, which is small business creation. Just the ability to, like, intersect the human world in a way that has been historically very inefficient has been a quality of life improvement for me.

   2. AFAlessio Fanelli0:12

      You know, my dad, their business, they deliver fish to restaurants. They got, like, this freezer with the frozen stuff, and, like, somebody's going out there with, like, the pen and paper every morning, kinda, like, writing down what's there. Sometimes they're like, "Oh my God, we're missing, like, three tuna," or, like, "We're missing a box of shrimp." All of that work now can easily be automated, even with just with the Meta glasses.

   3. CVClaire Vo0:31

      And you have another use case, which is the use case that my nine-year-old wants to see. So let's do our Pokémon card by AI use case.

   4. AFAlessio Fanelli0:40

      So I use Codex for two things. The first one is, like, getting the PSA certificates to keep track of a specific number for each grade. Then the next thing I'm working on is when you go to, like, all these trade shows, people are coming to you, they're selling you cards, and you gotta price them in real time. That whole process is super inefficient because people are, like, searching each card manually, like, on eBay or, like, TCGplayer, getting the number. You can actually use AI to save clock time for real people by doing these things autonomously.

   5. CVClaire Vo1:06

      [upbeat music] Welcome back to How I AI. I'm Claire Vo, product leader and AI obsessive, here on a mission to help you build better with these new tools. Today, I'm speaking with Alessio Fanelli, founder of Kernel Labs and co-host of the Latent Space podcast. He's gonna show us how he uses OpenAI's Symphony plus Linear to automate all his engineering tasks, and how he has Codex goal shopping for very expensive Pokémon cards. Let's get to it. Quick word from today's sponsor, Firecrawl. If you're building with AI agents, you've probably hit the same wall. Your agent needs data from the web, but the right pages are difficult to find, buried in JavaScript, or blocked behind logins. Firecrawl is a web data API that lets agents search, scrape, and interact with the web at scale, and get that clean, structured data they can actually use. Over a million developers, including myself, build on it. It's open source, and it's free to start. Stop fighting the web for data and start powering your AI agents and apps with Firecrawl at firecrawl.dev. Use code HOWIAI to get ten thousand free credits today. I'm excited

2. 2:24 – 4:31

   Prompter vs. agent manager

   1. CVClaire Vo2:24

      about what you're gonna show us because I think we have heard a lot of people talk about orchestrating many agents autonomously across a project, but we actually haven't seen many people do it. I still see a lot of prompting, even if you're prompting into a loop or a goal or something that spawns sub-agents, people are really, really still human in the loop. And so I'd love for you to tell us how you came to this point of doing more autonomous management of your agentic tasks.

   2. AFAlessio Fanelli2:52

      I started this podcast called Latent Space three and a half years ago, and my co-host works... He had built this thing called Small Engineer at the time, which was kinda, like, the first autonomous coding thing. Over time, it's always been a cool demo, but I feel like the models were not quite as good to really do longer running tasks. Um, that definitely changed, you know, end of last year, and I think everybody kind of feels the same. And what really clicked for me was, like, starting to move away from being a agent prompter to kinda be a agent manager, and that has kind of taken a lot of different ways. So the first thing that everybody tried was kinda like the Kanban board. Uh, you would kind of put all these things in there, move them back and forth. What I found is that it was hard to get two, three, four turns through that. Like, it was easy to get through the task and kick it off, but then it was hard to intervene on it. And also, like, having it local just didn't quite work. So the big thing for me was moving away from, um, kinda like local runtime to having it in a VPS in the cloud, and then having different channels to talk it to. So you can kinda, like, text the agents. You can, um, use Linear to talk to them. You can prompt them directly, uh, in the shell. And this is also, like, something... I guess, like, in the last month, you know, Codex also added, uh, Codex Mobile, Claude is also adding kinda like the, the mobile management. But I'll kinda run you through what I do, and then maybe people get some inspiration from it.

   3. CVClaire Vo4:15

      Great, and I will really benefit from this because I'm staring at my four Mac Minis over here. So I'm still running-

   4. AFAlessio Fanelli4:22

      Yeah

   5. CVClaire Vo4:22

      ... locally. Um, and I just come downstairs and, like, kick them alive every now and then.

   6. AFAlessio Fanelli4:27

      Right.

   7. CVClaire Vo4:27

      So I'm... Y- maybe you'll convince me to move all this to the cloud. Let's see.

3. 4:31 – 9:31

   Live demo: Symphony + Linear

   1. AFAlessio Fanelli4:31

      Yeah. So, um, I, I try to use some more fun examples rather than, like, the Meta one. So I own a card store in San Carlos called Merlin Games. Um, and so one of my interests is trading cards. So my setup is I have this thing called Zill. Um, Zill is basically like an agent plus a VPS. So this machine, for example, is like, you know, thirty-two gig of RAM, four cores. I have all my coding agents pre-logged in in here. Um, and you can also use some of the open source models if you want. And what I have here is kinda like this hosting thing where you can basically use it as your own server. On this, I have the OpenAI Symphony setup. So Symphony is basically a... I mean, you can kind of look it up for a better description, but it's kinda like a loop for, uh, turning issues into coding runtime, and then having kinda like Linear as a source of truth for it. So what I have on my Linear, I basically have all these different projects. So this one is Powered by Air, for example. And I work on it sometimes in through Symphony, sometimes I work on it through Codex directly or Cloud Code directly. And if you go into any of these things, basically what you see is you have the original task. So this is what I've wrote as the initial spec, which is, you know, pretty simple. Then I'll basically move it from here to to do. So this tells Symphony it needs to work on it. What Symphony does, it creates a Codex workpad, so the agent kinda makes a plan on how to implement it. It has a plan, it has a acceptance criteria, different validations, and Symphony has a file called w- workflow.md, where you basically explain how... what, what it should do for this. Uh, this will kind of go to work, and then eventually move it to human review So what you can do here is review it on GitHub, and you can add... Let me open the PR and show you. You can add all these different comments, you know, I guess now-

   2. CVClaire Vo6:27

      The result

   3. AFAlessio Fanelli6:28

      ... the pin result, this is like too long, blah, blah, blah.

   4. CVClaire Vo6:30

      Mm-hmm.

   5. AFAlessio Fanelli6:31

      And then you just move it to rework. So once you move it to rework, and then we'll do, we'll do, we'll kick one off while we record this. Uh, it creates a rework checklist. So it goes through all the comments and it's like, okay, um, these are all the things that went wrong, kind of addresses them. It tells you how to address them line by line. Um, moves it back to rework, um, to, from rework to done once it gets merged, and that's kind of like the flow. I don't really look at the traces one by one. I just kind of direct the, um, the agent to work on it. So whenever I'm... Even if you're outside, right? Like I might be on, on my phone and I'm looking at something here and, uh, let's say here we're like, "Hey, this is kind of like noisy." Um, what I can do is like I can create a new task as I clean up premium stable. Let's remove the spread column. It's too noisy. Let's also make the set name clickable so I can look at other cards. And here I'll simply put it in to do, create issue, and then each of these Symphonies has its own dashboard on it. So this one is TCG Bar Buyer. Uh, these are previous tasks that run. So one of the things I'm also trying to do is try and figure out how much is software gonna cost to build. So I think people understand the idea of like the agents write it, but sometimes it's hard ahead of time to know how many tokens it's gonna take, and so it's hard to price, um, and understand what it's actually worth doing. So as you can see, most of these tasks are kind of like, you know, 15, 30, 60, but then this one is like 221 million tokens. And so you can kind of go back here and be like, okay, this task was how to make it deployable in Vercel. So, you know, this whole thing was just not working. It was originally built as kind of like a local thing. So I had to like, you know, rewrite the storage, kind of like, you know, change how the requests are handled, blah, blah, blah. So this is like quite a big task, so it kind of makes sense that it costs a lot of tokens. Um, and so from here you can kind of start to think about how in the future can I make these more efficient by either adding more checks or adding better descriptions or better tooling. So the task we just created, you see, just kind of went live. So here you can kind of see, uh, you know... Obviously, there's usually like, you know, four or five of these in different projects that are running, so I don't really wanna see the whole thing, but I just wanna glance, and I'm sure I can make this UI a little prettier. Um, maybe once they give us Fable 5 back, um, that will be good enough. Um, but so this is working, right? And so in a little bit this is gonna go from, from in progress to like human review. And once it goes to human review, then we can kind of look at the Vercel preview, and we can make comments on the code and on the front end and kind of move it back. But I could be doing this here. I could be doing this on my phone. I could be doing this anywhere really.

4. 9:31 – 14:15

   Setting up Symphony

   1. AFAlessio Fanelli9:31

      Um, and to kind of put it to the extreme, I had, uh, let me see if I can find it. Um, I created this project called PyQ, which was basically like putting your repo plus the Py agent in a VPS, and then anybody on the internet could send you... I think I put this in a different project. Could send you like a coding task to your product. So it's almost like in the future, you know. And I think some people now have this idea of like, um, request for prompt instead of like request for, for requests. Everything is just how do you transfer context between people.

   2. CVClaire Vo10:06

      So before we move on to maybe another workflow, what you just showed me was, look, you can just create a, a Linear project for any of one of your code bases. You can integrate that with Zo and with Symphony, and then all you're doing is really tasking Linear, um, as sort of like your state machine for all the work that needs to happen in your code base. You can manage that on Linear from your phone, you can manage that from your desktop, from the web, and you don't really have to worry about the framework of how that task gets, you know, uh, broken down, how it gets implemented, even how your comments get reviewed. That's all set up. And I just wanted to share for people, Symphony, um, is something that OpenAI open sourced as sort of a framework for autonomous runs. So it's, it's just a very opinionated way to do this work, and it basically does what you just showed. It monitors a Linear board, spins up agents when it gets assigned something, um, and then, you know, you can, you can land it in a PR and it gets marked as done. How simple was it for you to like actually set up Symphony? Because I think people look at these things and they're like, "Okay, that makes sense, but what do I actually do with this GitHub repository?" And I know they have these two options here, which is like basically tell your coding agent to build it for you, or there's this reference implementation. How did you actually implement Symphony?

   3. AFAlessio Fanelli11:26

      Yeah. I took the Elixir, um, implementation. That really like... The core things to change are like the workflow.md in the main, um, in the main folder that kind of like explains how to do it, and then build the UI. So the, uh, Symphony itself doesn't have a visual UI for it. And, um, I also don't think it has the same, uh, it doesn't have by default the ledger for token usage. Um, so it's only like a, yeah, a state monitor. It doesn't actually look at like, you know, how much have you spent per task and kind of like all these different things. Um, but yeah, I think like overall the harness, you know, it's pretty straightforward.

   4. CVClaire Vo12:05

      Yep.

   5. AFAlessio Fanelli12:05

      I think like the reality is like how do you build tools for it to be more effective? And that's kind of like one of the main things also at Kernel Labs we've been working on. So we built this other thing called Glimpse, which is, um, um, kind of like a playwright extension that coding agents can use to take, uh, screenshots, to do visual diffs between screenshots, um, take videos. Um, and so it's almost like- Yeah. How do you let these runs kind of go longer and longer? So it's less about the orchestration itself and, like, the tools you give it to keep going versus coming back to you with the human review. And I think that's also why it's so important to, like, keep track of how many tokens and how much time it takes, because it's usually like directionally it explains to you how many issues it ran into, you know. So if you expect something... You should start to have, at some point, some idea of, like, how many tokens you think this will take. You know, is this like a 10 million token task? Is this like 100 million token task? And if the reality is very far away from your expectations, there's probably something in the tooling layer that you can do to improve. So that's really where most of the kind of like value comes from for people.

   6. CVClaire Vo13:16

      Yeah. One of the lessons I want people to take away from this is I get asked all the time, like, "Claire, how do we build our own agent orchestration platform?" And like they, they send me these like giant, very complicated documents and workflow diagrams and, you know, pointing them to just like symphony spec.md, which just describes how the system is supposed to work. It's in natural language, and it just is very prescriptive about what the primitives are of this workflow and what to store and, uh, record and how to move things forward in sort of the software development life cycle. Like it's very long, it's very detailed, but it's literally just a markdown file.

   7. AFAlessio Fanelli13:56

      Right.

   8. CVClaire Vo13:56

      And I think people kind of over-engineer at, at first what these things can be, and ultimately the power of LLMs, especially these newer models, is you could just give them a spec for how they will work, and they will, you know, um, they will lock to that spec when executing

5. 14:15 – 18:06

   Purging your skills files

   1. CVClaire Vo14:15

      whatever you hand it.

   2. AFAlessio Fanelli14:16

      Yeah. I think everybody just wants to have the magic skills file that does everything for them, or like the magic MD, uh, that solves their business problems. I think the reality is like, now more than ever, like small sentences have like very... A, a lot of weight, you know. Like for example, I built this, um, work tree manager, uh, before the coding agents themselves added. And in every agent's MD, I was like, "You have to use the word tree manager." And now sometimes I forget I had it in some projects, and then I start a task and it's like reinstall. I'm like, "You don't need it anymore." But like, because I had deadline, every time it's now doing that. Um, and I think a lot of folks have been talking about purging your markdown files now every few months. I think that's something that obviously makes sense. Um, I think the models themselves have the, also have this like tendency to like add rather than remove. So if you're like, "Hey, you don't need to always use the work tree manager," instead of removing that line, it's gonna add a line to say that, "You don't have to use it all the time though," you know. And now you're kind of getting more and more confused. So for example, my... The, the skill.md, it's not super descriptive on what to do, but it's like, "Hey, this is where you put the, the Symphony. This is how you should architect it." So every Symphony instance has kind of like its own name. It's got the repo that you're working from. It's got the workspaces for each task. It's got logs, which include the token usage, and it's got the state, um, of, of this run. And then these are the things that you need, so if you don't have them, ask me, otherwise don't ask for stuff. These are the exact commands. These are the flags. But I'm not really telling it what to do and what to use. I'm just saying these are like things that you gotta keep in mind, and then you kind of let the model work. Um, see, this is like a good example. I think it just added this today when I was adding, uh, new projects. Like this one is already registered here, and it's like this should not be in this file. Like it should look up every time. It should just search every time what's already there. You know, those are like all examples of like if you just let the models kind of do their own things, then you just end up with these like very, um, weird things.

   3. CVClaire Vo16:37

      Okay. So read diff your skills and read diff your markdown files and get some stuff.

   4. AFAlessio Fanelli16:41

      Yeah. They're not-

   5. CVClaire Vo16:42

      Get some value there.

   6. AFAlessio Fanelli16:43

      I feel like the create skill thing that the Codex app, for example, added, I think is like a great idea, but I think for a lot of people it just puts them in a lot of trouble because it's like they're not very descriptive in the skill itself. And then the model is like very focused on following the skill, and so they're actually doing, um, themselves a disservice a lot of the time.

   7. CVClaire Vo17:04

      Yep. This episode is brought to you by Jira Product Discovery. AI has made individual PMs incredibly productive, but multiplayer mode is where it still breaks, getting everyone aligned on what should actually get built. Decisions live in a markdown file from last week, the roadmap's a spreadsheet no one's looking at. Jira Product Discovery is where teams actually decide what to build. Capture ideas, prioritize them as a team, and share a living roadmap everyone works from. It's powered by Atlassian's teamwork graph, so it can pull in customer feedback, what your team's shipped, plus your goals, and suggest what to build next. And when a decision is made, you can hand it off straight to Jira, so a developer, or even an agent, can pick it up and start building. Teams at Canva, Deliveroo, and Toast already use Jira Product Discovery. Join more than 25,000 teams at atlassian.com/howiai. Start building the right things

6. 18:06 – 19:10

   The benefits of this system

   1. CVClaire Vo18:06

      together. Do you feel like you're getting a lot more done on this orchestrator? Is it just like better ergonomics in that you can manage it on the go? What's the benefit you feel of using something like Symphony?

   2. AFAlessio Fanelli18:17

      Um, I think it's definitely... I, I don't know about the getting more done. I think it, you know, in the limit you get more done, but in practice it's like you can't really like stay on top of like 100 new things a day.

   3. CVClaire Vo18:28

      Yep.

   4. AFAlessio Fanelli18:29

      Uh, the thing that's really helpful is like having the full history of one task in one place.

   5. CVClaire Vo18:33

      Yep.

   6. AFAlessio Fanelli18:33

      So because it has the original spec, it has the first work path- It has the rework workpad. Every time you're like, "How did things go wrong?" You can kind of pinpoint where that was, and then you can use that to inform, you know, your agent's MD in the future, like the Symphony workflow.mp. Uh, versus if you're just using Codex, it's kinda... It's really hard to search through conversations, you know? Um, and so the, the... Everything is like, how do you shape the context, you know? Like Symphony is just a way to shape the context. It's not, it's not giving you any new capability that you wouldn't have by using the coding agents directly. It's just helping you wield

7. 19:10 – 24:17

   Demo: Using Codex to hunt for Pokémon cards

   1. AFAlessio Fanelli19:10

      it.

   2. CVClaire Vo19:10

      I, I love that. So, okay, you, you have shown us how you're doing sort of this orchestration of, of agents or at least like workflow management of agents. Again, like let's make this much more accessible for people. This is just a workflow to manage your, your, your agents, especially around coding. And you have another use case, which this one is the use case that I wanted to see. This other one is the use case that my nine-year-old wants to see. So let's do our Pokémon card by AI use case.

   3. AFAlessio Fanelli19:40

      People think there are a lot of startups. There are way more Pokémon cards than-

   4. CVClaire Vo19:45

      [chuckles]

   5. AFAlessio Fanelli19:45

      ... there are ever gonna be startups to follow. Um, so one thing, uh, the, the reason why I built this, um, power buyer thing is-

   6. CVClaire Vo19:52

      Mm-hmm

   7. AFAlessio Fanelli19:52

      ... for us to kind of keep track of, you know, inventory we wanna buy and things like that.

   8. CVClaire Vo19:56

      Yep.

   9. AFAlessio Fanelli19:56

      Um, so I use Codex for two things. The first one is like getting the PSA certificates to keep-

   10. CVClaire Vo20:02

       Mm-hmm

   11. AFAlessio Fanelli20:02

       ... track of like, you know, basically Pokémon cards, you can get them graded by PSA.

   12. CVClaire Vo20:07

       Mm-hmm.

   13. AFAlessio Fanelli20:08

       Um, and then there's kind of like a specific number for each grade.

   14. CVClaire Vo20:12

       Yep.

   15. AFAlessio Fanelli20:12

       And that's available through an API.

   16. CVClaire Vo20:14

       Yeah.

   17. AFAlessio Fanelli20:14

       But you need one certificate number here to start from. Um, there's obviously no way to just download that. So what I have is like, you know, I have Codex pursue goal, fill out the certificate number for every card that costs more than $1,000. So I give it browser access, for example, here, and it just, you know, going on, on the internet and it's looking at things, and it's like downloading the images, and it's extracting the number from it. And then what you can then do is like, you know, use the eBay PSA premium hunt. Let's find some underpriced cards from our, you know, list.

   18. CVClaire Vo20:53

       Okay, so that's... Is that a skill?

   19. AFAlessio Fanelli20:55

       Yes. The, the skill is basically telling you, um, how to figure out which cards are worth looking for.

   20. CVClaire Vo21:02

       Mm-hmm.

   21. AFAlessio Fanelli21:02

       And then it just says how to batch them, you know. So just do five per batch. You don't, you don't wanna get like, you know, captured by eBay or stuff like that.

   22. CVClaire Vo21:10

       [chuckles]

   23. AFAlessio Fanelli21:11

       Um, and then there's also things, for example, there's different grading companies. So you might have, uh, you know, a PSA 10 is the same as... Like a PSA 9 is the same as like a, um, BGS or CGC 10. Uh, and so all these different rules you might wanna, you might wanna have in there. Um, and so here just, you know, through the API, it's looking up from our power buyer software-

   24. CVClaire Vo21:35

       Yeah

   25. AFAlessio Fanelli21:35

       ... the cards that we wanna look at. Um, and then it's using the in-app browser for, for eBay. Let's see if I can open it here. So now it's going here, and it's like looking at, just searching some of the... And these are all like, you know, 10, 20, $50,000 cards. Like you don't really do this for, um, $5. Um, and so now it kind of goes in here and, you know, you'll see it then, um, tell us, you know, this card is underpriced or, uh, this is card, this card is, you know, worth buying right now. Um, and yeah, this is like, you know, a, a good example of like, you can build anything, but then at some point you gotta have some business outcome or like some way to make money with the software that you built. Um, and trading cards is actually a great example of like, the more money you can put to work, the more money you can make because it's like an inventory-based business. So you can only make as much as you can sell. Um, and so having this to help us, um, automate looking for some of these like higher value cards, for example, is like super useful. Um, and then the next thing, you know, I'm working on is, um... And this is, I guess, getting in the weeds of the business, but, uh, when you go to like all these trade shows, you might have... People might see them on kind of like Instagram Reels or whatnot, uh, where people are coming to you, they're selling you cards, and you gotta price them in real time. That whole process is like super inefficient because people are like searching each card manually, like on eBay or like TCGplayer, getting the number, blah, blah, blah. Um, and so you're actually losing a lot of money. Uh, so when people talk about, um, you know, AI response time, I think for these long-running tasks, it's actually not that useful. But you can actually use AI to save clock time for real people by doing these things autonomously. Um, so yeah, it's just been fun to try and apply it outside of like, "Hey, I'm building the tool for you to build the tool to build the tool," so that hopefully somebody down the line does something that is worth the, for someone to use. Um, so yeah. For example, we got this Empoleon right here. Let's see. I guess it broke the link. See? Even, uh, it happens to the best of them. This is something actually... It's funny. People talk about software automation, but, um, Codex has this kind of like the dollar sign. It's like tied to skills. So sometimes when you're like linking with the dollar on the thing, it prefills it to a skill. So right now, right now it's looking for a skill.

   26. CVClaire Vo24:06

       Instead of a URL.

   27. AFAlessio Fanelli24:06

       You see like TCG-

   28. CVClaire Vo24:08

       Yeah

   29. AFAlessio Fanelli24:08

       ... power buyer HTTPS. So, um, yeah.

   30. CVClaire Vo24:11

       That's, that's funny.
8. 24:17 – 28:23

   The benefit of AI for small businesses

   1. AFAlessio Fanelli24:17

      but yeah, just, um, th-this is overall one trend whereas like there are a lot of businesses that are based on kind of like, um, highly heterogeneous data, um, that have been impossible to scale with software. Because before you have kind of like something as malleable as an LLM that can go through these things- It's really hard to use even like text or image classification for these things. Um, and so yeah, I think you're gonna see a lot more of, of these businesses. Like I think the same-- I, I forget who it was, but the same thing is happening with like vintage clothing, for example. Um, some people are doing something similar for, for that stuff. Um, because again, you see it all the time, right? Oh my God, I went to Goodwill and I saw this, I don't know, Prada bag that was like in Good- in the Goodwill thing. And, um, this has always been kind of like a mismatch between like human bandwidth and like the information that is coming from them.

   2. CVClaire Vo25:14

      This is my favorite one in that I, I think it enables what a very positive outcome of AI, which is small business creation.

   3. AFAlessio Fanelli25:21

      Mm-hmm.

   4. CVClaire Vo25:22

      And I think, you know, this business of-- clearly trading cards are a huge business, but no, what, what you're able to create this bigger business because you have the leverage of AI, and this is something that a human would have to manually do, and just the limits of time, space, and human cognitive capacity means you're probably unable to capture as much of this business-

   5. AFAlessio Fanelli25:44

      Right

   6. CVClaire Vo25:44

      ... as you are today. I also think I love this use case because it shows where AI helps you intersect the physical world in a really effective way. And an example, maybe I'll do a podcast on it, is a couple weeks ago I had a rage-out about how much is in my house, and I, I-- most of it is books. Most of what's in my house is piles of books. And so I placed a bet with my children and my husband. I said, "How many books do you think we have in this house?" And I went around with a camera, and I took pictures of every, like every pile of books. There's books everywhere. And I had Gemini go through it, 'cause I think Gemini's like particularly good at this. And I had-- we have 600 books in this house. It's like, uh, more, it's a hund- more than 100 books per person-

   7. AFAlessio Fanelli26:33

      [laughs]

   8. CVClaire Vo26:34

      ... in this house. But I was able to catalog all these books, put them into categories-

   9. AFAlessio Fanelli26:39

      Right

   10. CVClaire Vo26:39

       ... mark where they physically are, find all the duplicates because, you know, I buy a book, and then my husband buy a book. And, and just the ability to like intersect the human world in a way that has been historically very inefficient has been a quality of life improvement-

   11. AFAlessio Fanelli26:52

       Right

   12. CVClaire Vo26:52

       ... for me with AI, and that's on a personal level, but I also think as a small business owner, it's, it's really im- important.

   13. AFAlessio Fanelli26:59

       Yeah. I think that's like the thing about AI that most people don't wanna look at, just because every previous technology was like so-- like the economies of scale will help you get more leverage out of it, versus even the softer factories, right? At some point, even if you're like Salesforce, it's not like you can release 5,000 features a week. There's some limit to which you can get leverage out of these models and specific tasks versus, you know, my dad, um, their business back in-- I grew up in Rome. They deliver fish to restaurants, and they got like this freezer with, with the frozen stuff, and they have the fresh fish, and it's kinda like somebody's going out there with like the pen and paper every morning, kinda like writing down what's there. Sometimes they're like, "Oh my God, we're missing like three tuna," or like, "We're missing a box of shrimp." Uh, and a- all of that work now can easily be automated, you know, even with just with the Meta glasses or something else. Um, and so you're kinda helping actually, even at a small scale, you can get a lot of leverage out of it. Um, and yeah, I was, you know, I went for the first time to Japan last fall, uh, and I think that's, uh, a good example of like most things there are kinda like smaller businesses that two, three people run, and they're very happy to do it. Um, and I hopefully we see a lot more of that in the, in the US too.

   14. CVClaire Vo28:15

       Yep. Well, that's, that's the life that I live. I'm very, I'm very happy to be a small business of one, one and a half people-

   15. AFAlessio Fanelli28:22

       Right

   16. CVClaire Vo28:22

       ... and a

9. 28:23 – 35:52

   Lightning round

   1. CVClaire Vo28:23

      bunch of agents. Well, this has been awesome. I really appreciate you showing us the range of, you know, coding all the way to sort of more like physical or inventory-based AI. We'll do a quick lightning round question, and then we will get you out of here. Um, you know, my first question is, what are you excited about that you think most people aren't doing with AI that you are either starting to do or you think people will start to do in the next couple months?

   2. AFAlessio Fanelli28:48

      For me, recently has been personal finances. Um, uh, I mean, ChatGPT just added the connectors for, for all your accounts. Um, we just sold our house, and so I was like, "What should I do with the money?" Um, and it- it's actually pretty good because it keeps you on track. I think for me in the past, the thing has been like, I don't wanna like refigure out what am I doing right now, where do I have invested my money, should I invest somewhere else? Is the SpaceX thing real? Blah, blah, blah. Um, I think having AI is kinda like a offloading thing, um, because in the past sometimes people are like, "I'm looking to make sure I'm not fucking it up," you know? It's not like I'm actually adding a lot to it. I'm just kinda stressed I'm gonna mess it up. And so if you can have AI be the safety net, it kinda frees you from a lot of things. I do the same, like I was using this thing called Wafer, which is like a weekly unlimited tokens on the open source models. They just shut it down, sadly. Uh, but I was having it read my Gmail every five minutes, and I was like, "Just read... Is there any email I should actually respond or look at?" Um, before I was always like, "Oh, maybe I'm missing something. I should like check my inbox," blah, blah, blah. Now I know 100% that if something important comes in, I'll know about it, and that kinda removes a lot of stress from it. So it kinda being this kinda like context offloading, um, I think people should do more of that.

   3. CVClaire Vo30:11

      I, I completely, completely agree. Okay. My second question, because we were talking about books, and I'm gonna make you laugh, which is I'm gonna turn off portrait mode.

   4. AFAlessio Fanelli30:20

      Uh-huh.

   5. CVClaire Vo30:20

      We have, we have the same book.

   6. AFAlessio Fanelli30:21

      Ah, nice.

   7. CVClaire Vo30:21

      Look at us.

   8. AFAlessio Fanelli30:22

      The...

   9. CVClaire Vo30:24

      Yeah.

   10. AFAlessio Fanelli30:24

       We have the... Yeah, yeah.

   11. CVClaire Vo30:26

       I mean, we have-- I, I think we have v- a lot of the same books. You know, we're very, very typical.

   12. AFAlessio Fanelli30:30

       Um, I'm a sticker or a farming, you know, with my page markers. So...

   13. CVClaire Vo30:34

       Love that. So what's a book that you always recommend to people?

   14. AFAlessio Fanelli30:38

       Oh, what's a book? I, I think it depends. I feel like in different times of your life you need different books. Uh, one thing I actually always recommend is called The Monk and the Riddle. Um, which is kind of like, uh, based on startups and venture capital, but, uh, it's basically this VC meeting with this founder, and the founder is like, "I'm gonna do this startup for like funeral homes. And I don't really like funeral homes, but I think it's a big market. And then once I sell the company, I'll be able to do what I like." Um, and the VC's kind of like, "Well, why don't you just do what you like now?" And he's like, "Well, I don't know, it's kind of risky." He's like, "I'd rather just do this thing that is like big and like, then I'll do what I like." And I think I see it a lot in founders where it's like, "I should do the thing that people want me to do," versus like doing the thing you're passionate about. Um, I think that's, that's one that I always recommend to people. It's g- it's very short. Um, people like it. Um, outside of that, yeah, I d- I, I mean, The Divine Comedy by Dante. Honestly, like I grew up in Italy, and in Italy for three years you have to study The Divine Comedy. The one year, uh, you know, Inferno, Purgatory, Heaven. Um, it's just a reminder of like how great the human mind can be, like, you know, in the, uh, in the middle of the 1200s, um, 1300s.

   15. CVClaire Vo31:54

       Yep. Okay, we're gonna do, we're gonna do a behind-the-scenes How I AI. Um, my son Henry is over here in the corner listening to the podcast.

   16. AFAlessio Fanelli32:01

       Uh-huh.

   17. CVClaire Vo32:02

       Henry, just say really loudly, are we making you learn about Dante's Inferno as well? Dante's Inferno, I told you about going to H-E-L-L, hell. You don't remember?

   18. AFAlessio Fanelli32:16

       Yeah, yeah.

   19. CVClaire Vo32:17

       Yeah. I'm teaching you about that?

   20. AFAlessio Fanelli32:19

       Yeah.

   21. CVClaire Vo32:20

       And the Seven Deadly Sins.

   22. AFAlessio Fanelli32:21

       Nice.

   23. CVClaire Vo32:21

       Uh, you know, I also, quick aside, are you an AS Roma fan?

   24. AFAlessio Fanelli32:26

       Of course.

   25. CVClaire Vo32:26

       We're, we're, we're a, we're a Roma family. My, um-

   26. AFAlessio Fanelli32:32

       I only do this so that I can buy you one in the future.

   27. CVClaire Vo32:35

       My six-year-old is like three days out of the week wearing a Roma kit. He's got like three different ones. Um, so my-

   28. AFAlessio Fanelli32:44

       That's amazing.

   29. CVClaire Vo32:44

       My husband and my boys are Italian citizens. I am not, but they are.

   30. AFAlessio Fanelli32:47

       Nice.

Episode duration: 35:54