Full Workshop: Setting Yourself Up for Success — Jason Liu, OpenAI Codex

Talk transcript

1. 0:00

   [upbeat music] All right, let's just, uh, kick things off. How many people here already saw the keynote that I gave?

2. 0:19

   Okay, not everyone. That's good. This talk is effectively gonna be a stretched version of what I had given in the main stage, except two things. One, I want to give you some time to try to set things up yourself, you know, Wi-Fi gods permitting.

3. 0:33

   And then also, two, be a little more interactive. I, I had to be very high level when I was talking about what I use Codex for, but here, if you have any questions, we have, like, seventy minutes.

4. 0:42

   If you have any questions, just raise your hand and we can start answering some of these things, especially because not a lot of the workflows have been really well documented.

5. 0:48

   And so if you, if you're very curious on how things work, I'm really happy to, uh, answer any questions we have here.

6. 0:55

   Um, yep, I'm Jason. I work at OpenAI. I don't really know what my job is anymore. We do a lot of things. Um, clearly the slides are already in the wrong order as well.

7. 1:05

   But generally, you know, I've done things like doing a lot of prototyping work and just writing lots of code and just having a goal run for two days to build a game or some kind of web application, right?

8. 1:16

   We've also looked at things like running evals and hill climbing things. Uh, I also use Computer Use to edit iMovies and make little videos. I do partnerships and education and operations by taking my meeting notes, turning them into documents, working with other vendors, and also working with different foundations and programs to get funding.

9. 1:36

   And all of this work is done effectively in the Codex app. Uh, I don't know if you can tell, but right now this slide is being served on localhost in the in-app browser of the Codex application.

10. 1:46

    And so anytime I find something I don't like, I might just hit the annotate tool, give a comment, and have Codex clean up these slides. Um,

11. 1:56

    I'm not the biggest token, token maxer. Uh, I, I think I'm doing all right. I see some folks doing like, you know, a couple, couple billion every day. Uh, the goal of this talk isn't just to like waste all your tokens, but really of-- help you avoid wasting your tokens by telling you what has actually worked, and

12. 2:11

    in particular, sort of the tricks I use to make these things productive.

13. 2:16

    Um, and again, the goal, like I said earlier today, was to catch you up on what's changed in the Codex app, uh, give you some time to set things up.

14. 2:23

    So if you have time right now and you, you haven't downloaded Codex, just go ahead and do that. And then we can go a little bit deeper into setting things up.

15. 2:29

    And so I've actually prepared a little monorepo that you can use to clone in, get all the skills that you need, get all the setup that you need, and go from there.

16. 2:39

    So let's go a little bit deeper. Um, if you're new, feel free to set things up. If you're pretty experienced, just like chill out, try some of the things I'm talking about and then, you know, I think every fifteen minutes we'll have some time for questions, and we can go into the more like what feels like AI

17. 2:54

    psychosis, but maybe actually works kind of, uh, domain of using these systems, right? A lot of the work and knowledge work now, because the coding is solved, because a lot of this operations work is solved, is really just understanding what you can do, right?

18. 3:07

    In a world without AI, maybe I have ten teammates. Each teammate is working on one thing, so I need to have like ten things I'm keeping track of. Now we live in a world where like everyone I'm working with has ten projects.

19. 3:20

    I now have to keep track of like two hundred things, and I don't know what's important. There's definitely a Slack message I've missed. There's probably some email I've missed somewhere by some foundation, and Codex helps me organize all of this stuff.

20. 3:31

    And so again, the things I really want you to take away from this workshop is the fact that compaction works really, really well. Like, I have threads now that are like five weeks old that have, you know, four hundred sub-agents in them, and they generally just know what they need to do.

21. 3:48

    They know what their job is. I also want you to become really comfortable with talking to your comp-computer. Uh, earlier today I said that Tony Stark is not texting Jarvis, right?

22. 3:58

    And there's really no future when text input is the thing that matters.

23. 4:04

    I basically use a foot pedal. So I like, I, I have a button that is transcribe and a button that says enter. And so I'll just come by my desk with my hands behind my back and I just go like, [foot thudding] you know, "Fix this, make this change.

24. 4:14

    Also, like message this guy on Slack." [foot thudding] And then I just go back to, you know, talking to my coworkers and trying to figure out what is like the human side of actually working at OpenAI rather than just like monitoring Slack all day.

25. 4:26

    Uh, App Shots is my favorite feature of all time. It's like very satisfying. If any of you just like are on Codex right now, just press the Command button side by side.

26. 4:36

    You're gonna get this real nice animation, or you're gonna get a modal to tell you to install Computer Use. Just do that. It's amazing. Uh, invest in your personal memory, right?

27. 4:47

    I, at this point, when someone asks me what I'm doing, I don't even have any idea. I kind of have to like look at my threads and look at the conversations to figure out how much I've delegated away and how much has been automated.

28. 4:57

    And if you can then invest not only in skills for yourself, but also plugins for your entire team, you can become the superhero that actually sort of augments the rest of your company.

29. 5:07

    It's one thing to say, "Oh man, like I can use all these tokens and look how many tokens I'm using." But actually, if you're rewarded by how often the plugins you've built are being used by your teammates, that's a huge win, right?

30. 5:19

    How are we doing like implementation? Like one of the most popular skills is just the like finalize the Codex app skill. And anyone who makes a pull request basically triggers the skill before review.

31. 5:29

    And basically everyone at the company uses it, and it's always been able to find things that I've done wrong, whereas against a style guide. Um, one of the skills I have is just like reviewing docs, and it's basically just copying the pull requests of, uh, well, the, the PR reviews of our friend Charlie over here.

32. 5:46

    And so I just have like a review my code like Charlie based off the past year of like feedback he's given on pull requests. Review my code like Dominic.

33. 5:53

    And these things are incredibly valuable. And then lastly, once you get more comfortable with all those first four things, your pinned threads with automations, these things that wake up these threads over time, they're gonna feel like teammates.

34. 6:08

    And more interestingly, now that threads can talk to each other, so every thread has the ability to list other pinned threads, has the ability to rename threads, and has the ability to send messages to each other.

35. 6:18

    Not only can you have teammates, but you can have teammates that work together, and you can e-effectively start having managers, right? And so you, you went from an IC enabled by an IDE, then you have pinned threads that feel like a team where you're the manager.

36. 6:32

    And very quickly in the future, as models get better, this is where the s- the puck is going to skate to, you're gonna start having your, like, manager threads and then your IC threads.

37. 6:40

    And I'm sure in the future there's gonna be some other crazy orchestration, right? And all of this really is due to the fact that compaction works. Even six months ago, I, I don't know, like, how many people here have been told this, but you were always told if a conversation goes very long, start a new thread, right?

38. 7:00

    After 20 messages, it's, it's not gonna be that good. Um, every feature should be its own, its own, uh, conversation. If you do a code review, start a new session.

39. 7:10

    Those things basically aren't true anymore, and a lot of it has to do with compaction. Just pin the thread, rename it to the project ID, and that project thread should be able to delegate to sub-agents, create new threads and have conversations, and then write to your memory vault, which will allow you to just log what's happening.

40. 7:28

    And then with automations, you can just wake them up. [sniffs]

41. 7:32

    And so there's really three acts of working with AI, right? Working in Codex. You bring the context in, and I'll talk about how you do that and what are the ways you can bring context in.

42. 7:40

    Then you work on it, right? This, for example, this slide deck is just in the Codex app. And then you take actions out in the real world. [sniffs]

43. 7:49

    So I asked this during the, the keynote, but I'm also curious what the audience here is doing. But how many people use dictation when they interact with an AI?

44. 7:59

    Nice. How many use dictation even at work?

45. 8:04

    Yeah. I think we, I think we should all be a little bit more shameless, you know, uh, i-in doing these kinds of things. Uh, you generally talk about three times faster than you type, and it's just incredibly productive to be able to give the messy version of what you're thinking about to the AI and take that extra

46. 8:21

    time and just try to be even more thoughtful to the people you work with, right? It's like, now, I, I don't wanna send my coworker [chuckles] like a 15-minute voice memo, but I should be f- I should feel very comfortable sending an AI a 15-minute voice memo, 'cause you're gonna include some random tangents.

47. 8:37

    You might just say, "I'm pretty sure I had a meeting with Charlie sometime last week about

48. 8:44

    the agents SDK." And it will go and read like 35 meeting messages to figure out which one it was and make it relevant, and now all of a sudden, like, whatever memo you're gonna write or some project tracker that you're trying to do, uh, is gonna work, right?

49. 8:57

    But I would never do that with, with, um, AI. [sniffs]

50. 9:02

    So as you guys are doing and, uh, just listening to this talk, like, try to just s-sort of set up Codex the way that I've been describing these things, right?

51. 9:11

    So once you have your ability to just have you input into the machine a lot more effectively, you can start thinking about using th- things like skills and plugins.

52. 9:20

    Um, skills is a very simple construct. It's just a, a couple of files and some scripts. A plugin is a library of these things. And as you are just doing things many, many times, you can start thinking about creating your own skills.

53. 9:31

    And as you package a bunch of skills, you might start thinking about building out a plugin.

54. 9:37

    If you wanna install the plugins, we have a pretty good ecosystem now, some-something I'm very proud of. If you just go in the sidebar, click Plugins, you can just search whatever plugins make sense for you.

55. 9:46

    So if you use Slack, you can install Slack. If you use Gmail, Teams, most of these things are pretty built out. If there's something that you feel like you are missing, just, uh, at me on Twitter, and I'm sure one of my Twitter monitors will pick it up and send a message to someone on the Connectors team.

56. 10:03

    If you're already actually looking at the plugins panel, I also really recommend just starting the process of setting up the Chrome extension as well as Computer Use. We'll talk about this a little bit later, but, like, Computer Use was the first time in a long time I really sort of felt the AGI of being at work, right?

57. 10:21

    I was in iMovie for the first time. I didn't know how to use it, and it was just teaching me how to, like, export the movie. It was able to, like, figure out where the sound effects were, and it placed it in the right timestamps.

58. 10:31

    Really small things like this that really make using a computer very fun again, right? Like, I don't really have the time to, like, learn new software. But if Codex, if Codex can just show me what's going on, it's, it's, it's pretty awesome.

59. 10:41

    And as you see the cursor move, oftentimes you're, like, cheering for it to, to do the right action. If you don't have a link for the Chrome extension, you can just click this button here.

60. 10:50

    The difference between Computer Use is Computer Use can work behind the scenes to control any application, right? So whether it's Slack or some, you know, trading software, God forbid, uh, it can control all of those things.

61. 11:03

    With the Chrome extension, it just controls everything in the Chromed app. But the cool thing here too is, again, it doesn't take over your screen, right? Sometimes I'll just be working on my computer.

62. 11:14

    I'll go to the Chrome browser, and I'll realize that, like, Codex has just opened up three tabs to just look at my Twitter DMs and then just closes them back up as I'm just, you know, responding to some other email.

63. 11:25

    It's really cool to watch these things work in the background.

64. 11:29

    You can connect a bunch of other plugins. I use things like Notion, Linear. I also use Obsidian. It's just a good time.

65. 11:36

    Once you do this, what you're gonna find is just by asking really vague questions about your day and just tagging the right plugins, you're gonna realize the AI can learn a lot about you, right?

66. 11:47

    The AI does not a system now where it does, like one search request and tries to come up with an answer, right? It might check your emails and find a loose thread.

67. 11:55

    It might check Slack or some meeting and figure out what's actually going on. Who are these people? I had, I had one of my loops basically realize I was meeting with somebody, look at their LinkedIn, and realize that, uh, we went to the same university at the same time.

68. 12:09

    And so the moment I jumped on my call, I was like, "Hey, you are also from Waterloo. You know, do you remember this, this, and this person?" And, and immediately we had a, we had a connection, right?

69. 12:17

    And obviously, I didn't tell anyone it was AI, but that's kind of some of the small things that you can do by just improving your automation. You know, it can make you closer to people.

70. 12:26

    Um, as you build out your memory system, right, as, as you build out the Codex memory system and your, your ability to trigger plugins, maybe day one you have to tag everything.

71. 12:36

    But I've become, like, a worse and worse manager over time, right? Now I'll just open up the composer and just say, like, "What has changed about the launch?" And it'll be able to do a good job, right?

72. 12:46

    And it- it's-- that's possible because you have this long history. You have all these pin threads. You have these memories, right? It's the same thing with an employee. Day one, you have to show them every standard operating procedure, you know.

73. 12:59

    But at some point, you have an employee that has been here for seven years, and you can just say, "Hey, I think you should have made the company more money," and they can figure it out.

74. 13:06

    But it's only because they have this context.

75. 13:10

    Um, one thing you can just do, for example, if you wanna try it out, is you can just say, "Hey, check out the schedule, find all the sessions, organize them in a markdown file, put them in a spread- spreadsheet," and you'll just realize that we can do these things.

76. 13:21

    And maybe it'll do it with web search. Maybe you can do it with Chrome. Um,

77. 13:27

    lots of fun things here. If you wanna get inspired by just looking at what kind of skills exist, we have two really great sources. One is if you just run the skill installer skill, it will actually list out all of the OpenAI curated skills.

78. 13:41

    These include ones for things like GitHub, uh, best practices when writing Playwright code, you know, Remotion, for example. But you can also check out websites like Skillset.sh or use, uh, I think this is, like, Vercel's, like, skills, uh, tool.

79. 13:56

    And then you can just find other skills, right? So if I'm thinking about doing some more motion design or web design, or I know that, like, someone told me I shouldn't do-- like, use memo in React, but I don't really know what that means, I can now go install the React best practices skill, right?

80. 14:11

    But again, internally, one of the highest impact things I think, I think you can do as, like, the AI champion in your company is to figure out what the team needs and build out those skills, right?

81. 14:22

    I have a lot of skills on doing things like triage and how, how you do comms, right? If there's an outage on Twitter, how do you, like, convert that, figure out who, who needs to hear this?

82. 14:32

    How do you, like, start the SEV? What stat sig gates do you need to check? All of these things are now just automated.

83. 14:40

    And that's exactly what I just said in this slide. Um, we also have a really good plugin creator and a skills creator skill. So if you just ask Codex to trigger it, it will try to interview you to figure out what's going on.

84. 14:50

    And even mo- in a more useful way, you can also just do it yourself once, document everything, and just tell Codex to make a skill from what you've learned, right?

85. 14:59

    And as long as you tell it, "Hey, by the way, every time you run this skill, you're allowed to edit yourself if you learn something new," you can edit your s- the skill file.

86. 15:06

    These things will also improve over time. And a big theme that's happening over this talk really is just you kind of have to just get really comfortable with asking.

87. 15:17

    Like, we'll obviously try to make more of these things, like more slash commands. But more and more, like, I'm just not touching a computer, so it doesn't even make sense for me to, like, run a slash command.

88. 15:26

    I just wanna say, "What's launching this week? Check Twitter." You know, l- look at what I'm seeing in the browser.

89. 15:34

    Um, the example I've been developing internally has just been this, like, developer experience triage skills, right? So again, this skill just documents, like, every Slack channel that should be, uh, you should be aware of.

90. 15:46

    It knows which engineers have worked on what projects. It knows what Slack channels are taking in feedback. I know that if you DM me on Slack and you tell me that some regression has happened, I need to ask for a feedback ID, right?

91. 15:59

    Now, the agent does this automatically, and it does it automatically with app shots. So again, I- I don't know how many times I'm gonna say this, but app shots is one of my favorite features.

92. 16:10

    How many people here have just, like, sent a screenshot to Slack-- to, to, uh, Codex? Right? Like, almost everybody. But the issue is the screenshot does not have that much information, right?

93. 16:22

    The model has to then do OCR. And if you send a screenshot of, like, a Slack thread, the model has to, like, read the Slack thread and then do a list Slack channels function and then realize, like, there's a guy named Charlie, and then do, like, a list persons.

94. 16:36

    It takes a lot of hops. But with app shots, it takes not only the image, but the entire accessibility tree of the app. And so when I give it an app shot of a Slack channel, it knows the channel ID, so it knows exactly what function to call to post there.

95. 16:51

    It has the user IDs of every single person in that channel. So if I take an app shot and say, "Do some research and reply," it's only one function call.

96. 16:59

    It knows to send the send a message to channel, like, [REDACTED:generic_id], and then it, because, you know, it knows that Charlie is, like, [REDACTED:generic_id] four [REDACTED:generic_id] five, it can do that in a very s- fast hop.

97. 17:13

    So not only is it a very quick way of getting context into your system, it just gives so much more context that these subsequent tool calls do a really good job.

98. 17:22

    I have not done, like, filled out a form in, like, two weeks 'cause I just now tell Codex to fill out this form, right? It, it knows all the fields.

99. 17:30

    It then figures out that it's in Chrome, and so it'll use the browser extension. If it's in Safari, it'll use Computer Use. The model has become really, really intelligent.

100. 17:39

     And so just like y- you might have a manager that gets an email, and they forward the email to me with, like, three question marks, and it's your job to figure out what's going on, you can kinda start doing that with your AI as you start investing in these skills.

101. 17:52

     And most of this is because of the fact that you've built out your memory system. So if, if you guys are, are taking a look at these, these slides,

102. 18:01

     JSONL slash personal monorepo template, that is actually the template I used on my personal computer. It's basically just a directory tree and a bunch of skills that I use to sort of grow out my memory.

103. 18:14

     I'll also make one call out, which is, uh, if you open this in your browser, just press app shots and tell Codex I've set this up for you, and then you can pay attention to the rest of the talk.

104. 18:30

     Yeah, yeah, yeah. You can just tell Codex Jason has written a personal monorepo template on GitHub. Please find it and then install it.

105. 18:43

     All right.

106. 18:43

     Can I ask a quick question?

107. 18:44

     Yep.

108. 18:46

     Does it start each project with a vault and once you go and live elsewhere? Where do you put that?

109. 18:51

     Yeah. So this is, this is a really good point. So like for example, on the DX team, I make a lot of demos, and so I have like sixteen r-repos, like, you know, real-time demo one, like real-time demo two, like funny, right?

110. 19:04

     You have all these demos. I don't create new projects for them, right? The only project that exists on my sidebar is the, like, personal monorepo sidebar.

111. 19:15

     But Codex is able to still manage files outside of that project directory. And so in my agent.md file, I just say,

112. 19:24

     "Don't save any of the code in the monorepo, save it in like /dev." And just by that one line, if I tell it to clone a new project, it saves it in /dev.

113. 19:33

     If I tell it that I wanna work on my slides, it knows that there's like a /dev slides directory, but it's just an easier way of managing everything, right?

114. 19:40

     Like, I wanna start all my projects from my personal vault, and then it can touch the file system in any way that it wants to. Um, one call out, it kind of breaks, uh, like Git review sometimes in the sidebar, but generally it's been a pretty good experience for me 'cause I just re- I just review my

115. 19:55

     code in, in GitHub.

116. 19:58

     Thanks.

117. 19:59

     Sweet. Um, these are some of the skills I have just installed there. Uh,

118. 20:06

     there's no need to take a s- a photo, just ask Codex afterwards. But, um, the assistant plugin basically has the ability to, uh, onboard you. It will interview you, it will figure out what plugins you need to install, and then it will actually go create the threads it thinks it needs.

119. 20:20

     It'll create the automations. It's a pretty fun one. I have a bunch of skills in like auditing AI code and AI writing. Um, I don't include this, but one of the, my favorite skills of all time is called Write Like Me.

120. 20:33

     And if you wanna make one like that, all you had to tell Codex is, "Hey, Codex, I want you to read all the emails I've written in the past six months, all the Slack messages I've written in the past six months, and write a style guide for how to message just like me."

121. 20:47

     And then that's it. And then anytime I tell it to send a Slack message or write an email, it'll go, "Okay, this is an email. Clearly this is just like a customer support form, so I will be much more stern in my messaging.

122. 20:58

     Let me go draft this email." Hasn't failed me yet. Um, one thing I've also added that I think are, are really valuable to call out is I've made my own loop skill just 'cause I, I do like having a slash command every once in a while.

123. 21:10

     I'll talk about this in, in part two. And I also have a skill called, uh, simple HTML artifact that just designs artifacts the way I like them. I want my background to be white.

124. 21:20

     I want some, uh, certain style guides. And then ultragoal, which is like a super version of Goal that we'll also talk a little bit more about. Um, and if, if anyone's curious, like new person, new project, that's just a way of like running a script to bootstrap a new person.

125. 21:35

     I, I kind of have like a Palantir for my personal life now. It's just like a CRM and basically anytime my AI agent like finds a new person that's emailed me or messaged me on Slack or on iMessage, I just keep track of these things.

126. 21:47

     And the new project is the same way.

127. 21:52

     Let me just double check. Yeah. Cool. Um, and so I'll give you maybe like ten minutes to try to try to set this up and we can go in a little bit of a, a Q&A.

128. 22:00

     I'm happy to answer any questions about like how we bring context into our systems, how I've organized my personal memory vault and, uh, you know, some other crazy uses of app shots if anyone has any questions.

129. 22:11

     Yeah, what's your question?

130. 22:12

     Does this get rid of the need for like an Obsidian brain that brings context by itself because you have this compact in nature that's really good now?

131. 22:20

     Yeah. Um, so the question was do I still basically use Obsidian brain? Uh, the answer is yes, because I still want to sort of like keep track of everything.

132. 22:29

     Um, one thing I actually really like doing is I make my monorepo vault like a Git repo, and so maybe it'll work on it for like a couple of days and I'll come back and I'll just run Git diff.

133. 22:42

     And by running Git diff, I can just see like what the model has updated and what the model has not updated, and I can just confidently review that over time and just realize that, oh yeah, like I guess Charlie did respond to this person and close the loop and I didn't realize that, but now I know.

134. 22:56

     Right? And, and oftentimes it's, that's relevant in another conversation. More than that, it's also very helpful for when other people are asking me questions, right? Codex will feel very good about reading my memory vault, drafting a response, and then asking me for permission to send that message off.

135. 23:13

     And so if someone messaged me on Slack a question that the AI could have answered, the AI will just try to answer it. Um, and it might be simple things like, oh, like who should I talk to about this project, right?

136. 23:24

     And the model knows 'cause it's, it's in the memory vault. Um, one thing I'll also call out is if you want to use more tokens, you can also have like custom automations where the job is to maintain and manage and garden your me-memory vault.

137. 23:37

     Um, but generally that has not been a big issue for me.

138. 23:42

     One question over there.

139. 23:45

     What do you think about, uh, doing evals on the skills that you're creating or I guess manual review, automated evals, that sort of thing, profile climbing to better emulate the source code-

140. 23:55

     Yeah

141. 23:55

     ... versus just like YOLO one shot?

142. 23:59

     Honestly, I generally go down the path of like YOLO one shot only becau-- Oh, that scared the hell out of me.

143. 24:08

     Um, only because I know that like the way I build my skills is that they, they self-improve all the time, right? Like- I think the difference would be if I make a skill that I share with my team, I think about that a little bit more, right?

144. 24:25

     'Cause it's like, okay, do you, do-- does the, does the triage plugin know that, like, who is working on what feature? Can it route correctly? For my personal work, I generally just build a skill as quickly as possible, and every time it makes a mistake, I just correct it, and I tell it to, to move on.

145. 24:41

     And then generally what happens is, if I've used a skill for two months, I just generally feel pretty good about sharing with my team because I, I've just experienced it working.

146. 24:48

     Um, yeah, most of my skills connect to so many other plugins and connectors that, uh,

147. 24:57

     I just don't know how to eval that because I can't, like, snapshot my Slack at any given time.

148. 25:02

     There a question over there?

149. 25:03

     Are you pretty exclusively using the desktop app or are you using the CLI-- Codex CLI as well?

150. 25:09

     Sometimes I'll use the Codex CLI every once in a while if I wanna, like,

151. 25:15

     like be a little bit faster. But generally, the desktop app has pretty good-- it has a pretty good experience. Primarily because everything I do is an AppShot. [laughs]

152. 25:28

     Like, if I'm watching a video, I'll just, like, AppShot, like summarize this, and I'll continue to watch the video. I'll just watch the video with like the LLM like summary, right?

153. 25:36

     Or it's like, uh, if I see some kind of form or someone asks me to sign something, I go like, "AppShot, use DocuSign, like sign this and save it to my desktop."

154. 25:43

     Uh, like last week, it like DocuSign something, then like found a faxing service and like faxed my medical records. Like that's awesome. But I-- the CLI can't really do that.

155. 25:53

     Um, yeah. Any other questions? Yep.

156. 26:01

     What do you do while waiting for the agent to run?

157. 26:04

     I'm like learning to juggle. I'm like, I'm like learning to play the drums. Um, well, I think it's like two things. Like in the office, what I'm trying to do is I'm trying to like talk to more people, right?

158. 26:15

     It's like, it's like I'm just the AI's assistant to get more context that the AI can't get. Uh, no, I think, I think my job when I'm at work really is just when the AI is running, I should be talking to somebody.

159. 26:25

     I should be like learning about what they're working on or trying to make connections and then figure out what are the, you know, points of connection, right? It's like I should be talking to more people in real life as AI works.

160. 26:38

     I think someone had a question over there. Oh.

161. 26:41

     Um, I had a question about use cases where you found GPT 5.5 was overkill and GPT 5.3 Codex Spark perfectly intelligent enough while you can leverage the speed increases.

162. 26:53

     Yeah. So the question is like when do I think five five is overkill versus five three Spark?

163. 27:01

     I think this is colored by two things. Like because I have unlimited tokens, I, I don't really make those decisions. And then secondly, because I'm not watching my AI work, like most of these things are automations that run in the background, the latency has not really affected me.

164. 27:14

     The times where I use Spark is primarily when there's a very simple, uh,

165. 27:21

     there's a really simple computer use task, right? Like I just wanted to like click all the buttons and fill out this form. Like, uh, I think, I think I have a thing that just checks me into flights, and that is like a Spark agent.

166. 27:33

     And so now like anytime there's an email that's, that's like a flight check-in, my agent will check me in, download the boarding pass, and then send the boarding pass to myself on iMessage, right?

167. 27:45

     And like I just never do that kind of stuff anymore. Uh,

168. 27:49

     again, it is really weird when you're just like working, all of a sudden you check your, your-- check your Chrome desktop and it's just like, you know, JetBlue is just like the first page.

169. 27:57

     But again, it's, it speaks to the fact that like having access to your computer is uniquely powerful because it has your auth and, and your credentials and, and your, your, um,

170. 28:07

     your file system. Yeah. Over there.

171. 28:11

     So when you're doing something long-running, which specifically means your computer-

172. 28:16

     Yeah

173. 28:17

     ... do you also wanna travel with your computer outside?

174. 28:20

     Sorry, you're, you're really quiet. Do you mind just speaking up?

175. 28:22

     Yeah. I'm asking if you wanna do long-running tasks which need computer use-

176. 28:27

     Yeah

177. 28:27

     ... but you can personally not give your laptop as computer use. Let's say you have different devices.

178. 28:33

     Yeah.

179. 28:33

     How do you actually-- do you actually manage to juggle between different physical devices or it's only cloud and one personal computer?

180. 28:41

     Yeah. So I have-- We'll talk about this basically in, in the next section. But, um, with remote control, I can control both my local computer and my-- like a remote computer.

181. 28:53

     I think the difference is computer use is tricky because it has to be on your computer. That's one thing. The second thing too is if you go into Settings, Computer Use, there's a flag called Locked Use.

182. 29:06

     And if you enable that, as long as your laptop is plugged in, even if the monitor is closed, you can still trigger computer use commands through your phone. Um, that also gets really weird 'cause like...

183. 29:18

     But go on.

184. 29:19

     No, I was gonna say, my question is like you have two primary devices.

185. 29:22

     Yeah.

186. 29:23

     You don't wanna use your phone. You wanna take your laptop with you when you travel, but you want long-running tasks dispatched to the phone.

187. 29:31

     You're, you're really quiet. I can't hear you. [laughs]

188. 29:33

     Can I ask again?

189. 29:34

     Okay. Okay, cool. Um, yeah, like if you have multiple computers, you can still connect both your, your, your iPhone to both those devices, right? Like some people just have a Mac Mini.

190. 29:44

     Uh, I think the difference is do you wanna control your Codex or specifically computer use? 'Cause that, that requires like a operating system with a GUI. Um, but we can talk about this in a, in a little bit.

191. 29:54

     Cool. Like I said before... Were there any more questions? I know someone just raised their hand. Yeah, go ahead.

192. 30:03

     Yeah. Is there, uh, an elevated risk with computer use and how do you control that?

193. 30:08

     Yeah. I mean, there's always some kind of risk. Like,

194. 30:15

     like I think earlier versions might like edit a document a little too eagerly, right? Um, but realistically- I think we've done-- I, I think these models have done a really good job of being very precautious.

195. 30:28

     And more often than not, it's me going like, "No, just please just, just do it. Like, just, just like just sign the document. Like, this is... Like, please just like send this message."

196. 30:37

     Um, I found that, uh, the 5.5 models are pretty reluctant to take these like destructive actions. That's one thing. The second thing is, if you look at the sidebar here,

197. 30:49

     ooh, you have the ability to change your permissions. And so as a show of hands, like how many people use like, uh, ask me e- for every permission?

198. 31:02

     Yeah. [laughs] Yeah. Exactly. Okay. How many people use like full auto, full permissions YOLO mode?

199. 31:11

     Okay. I don't like that. Um, and then how many people kn- have used auto review?

200. 31:16

     Whoo.

201. 31:17

     Yeah. So I think auto review is actually a, has been really, really great. And again, I'm usually annoyed by the fact that my models won't do more than I want them to do.

202. 31:26

     Um, and so generally it has not been as big of an issue. The only times, the only examples where I, I'm really annoyed is it will like edit documents it shouldn't be editing.

203. 31:37

     But I just like add a, something in the agents.md and it's, it's never really messed with me too much. Yeah. I know that's not a real answer, but I think with a combination of auto review and agents.md file, I have generally felt pretty safe.

204. 31:51

     Yeah. And then if you're also at an organization, there's different admin settings that you can have. So for example, at OpenAI, uh, you can't use an MCP server to send an email if any of the people in that email is a non-OpenAI email, right?

205. 32:06

     Or like you can't, uh, send a Slack message to external Slack channels. Um,

206. 32:12

     that's when things get dangerous, right? But these are some things that you can control.

207. 32:19

     One question over there.

208. 32:20

     We have a mic too.

209. 32:20

     Yeah.

210. 32:20

     You want mics.

211. 32:21

     I have a follow-up question. Do you have any concerns regarding either security-

212. 32:27

     I got a, I got a mic for you. Do you have any concerns regarding either security and/or privacy? [laughs]

213. 32:36

     That's tough 'cause I work here. [laughs] Um, I think that...

214. 32:44

     I think that's hard to answer 'cause I don't really know what are like data retention policies for like individual versus enterprise. But Charlie, do you have any thoughts there?

215. 32:51

     I'm just gonna throw [laughs] that over to you.

216. 32:55

     Uh, just concerns about security and privacy.

217. 33:00

     Privacy.

218. 33:00

     Let's say you get a, I don't know, email offering you a job at a competitor company.

219. 33:09

     Uh, I mean, I, I think a lot of it goes back to we do wanna make... Like, the models are fallible, right? I don't think anybody in this room would be shocked to understand that like you can still jailbreak a model, for example.

220. 33:20

     Uh, but both the, the models themselves are getting smarter and better at not, you know, doing silly things. Uh, and at the same time we're figuring out, like Jason mentioned, what are those bigger, you know, limitations around the sandbox.

221. 33:33

     We, we started with very simple sandboxes where it was like, you can just run this command and nothing else, and slowly the sandbox has grown to the entire computer.

222. 33:41

     And I think we're figuring out what are the like computer level or organization level, you know, uh, edges to that sandbox that we need to build.

223. 33:52

     This is me using the delegate skill. [laughs] Great answer. Any more questions before we jump into act two?

224. 34:02

     Sweet. Awesome. So we just talked about a bunch of different ways of bringing context into the system, right? You can use your voice, you have plugins, you can use AppShots, and then you can also design different skills and plugins to figure out how to do more systema- systematic work.

225. 34:18

     So now we can talk a little bit more about the work itself. So like I said before, like every pinned thread effectively is a teammate in my mind. I have my chief of staff thread.

226. 34:28

     I have, uh, you know, Swix prefers to call it like, you know, the god thread. Uh, I, I have a thread to manage the agents M- SDK, whether that's implementation and documentation.

227. 34:37

     It has two sub-agents that it delegates to. Uh, the CLI, the open source program, and, uh, Twitter.

228. 34:49

     But if you wanna make it wake up, right, all you have to say is, "Keep an eye on this until sometime." You know, keep an eye on this every 30 minutes.

229. 34:58

     If you remember those like secret words, you can effectively automate like about everything in your life at this point. Um, and what this does, it will trigger a, a heartbeat automation, a thread automation.

230. 35:09

     You should think of it as a way of scheduling a message back into the thread. Right? In the beginning when we set up automations, it was very much the case that a automation would create a new thread every time.

231. 35:21

     So it might be, "Give me a morning brief," and it would create a new thread, and then it would do that kind of work. But as these models got better, I think the right design is scheduling these messages into the same thread.

232. 35:31

     So for example, because if you, if you download the monorepo, we have a loop skill. If you just do loop, right, this is the equivalent of just saying keep an eye on this.

233. 35:40

     Uh, keep an eye on this pull request. Anytime there's feedback, fix it. Make sure it's always mergeable. Al- make sure it's always rebased on master. Make sure that CI is always passing.

234. 35:51

     And it'll just do that. And then maybe you make a pull request in, in, on a Monday, you get really busy. Thursday afternoon all the feedback has been integrated, you know, C- CI is passing and, uh, you know, you're not like 4,000 commits behind, uh, the main thread.

235. 36:09

     I also do this with support, right? Again, if, if someone is like dealing with some issues on, on Twitter, on Slack,

236. 36:15

     AppShot, right? You know, like @, like [REDACTED:username] experience skill, figure this out. And it'll say, "Okay, this is an issue on the browser side. Like James is the one that works on the browser.

237. 36:27

     The channel is called like browser feedback. I'm going to post in that channel, DM James." And then I'll use Computer Use to open up Twitter to let them know that I've, like, escalated this internally, and then I will check every hour to figure out if James on that channel has responded, and then let the user know.

238. 36:47

     And then sometime later in the future, you're just, like, checking your computer and all of a sudden, like, Twitter opens up, and it's just like, "Hey, so and so, this has been resolved."

239. 36:56

     And then you just, you, you just hit Enter, and then you check your thread and say, "Oh yeah, a pull request has been made. It will get merged by next Thursday."

240. 37:03

     And, like, this is a crazy experience to witness, right? This is, this is actually allows us to do way more support without making it, like, the worst part of my job.

241. 37:16

     And then with the chief of staff thread... Oh, this is, this is, I remember this is where my slides get really messed up thanks to, uh, Codex, so I can't do everything just yet.

242. 37:23

     Um, you can also just do a loop that says, "Check all my connectors and give me an update as to what is the most important thing I, I should be thinking about."

243. 37:32

     You know, give it, give it to me in a nice format. You know, make sure you have links to every email that you read. Make sure you have a Slack link so you can deep link into the application.

244. 37:41

     And now just it's been really, really helpful to track these random things. And again, I think in my chief of staff thread, I have that line that says, "Check into all my flights if you can."

245. 37:52

     Like, send me the book- the, uh, the boarding pass on iMessage. And it just works. And again, it's really weird when you just start seeing your computer do- doing stuff while you're, while you're working 'cause, again, most of these tasks run in the background, and because I've had my permissions on, it's not like it needs to tell

246. 38:09

     me that it's using Safari. I just, like, find out that it's using Safari. That might be scary to some people, but, you know, it's, it's pretty great. And then lastly, this is something that happens r- really a lot at, at, uh, OpenAI, which is I don't even know what's, what, what's launching, [chuckles] right?

247. 38:26

     Is something delayed? Has something landed? Is it, you know, gonna happen at eleven AM? Is it, like, happening at four PM? I have no idea.

248. 38:35

     We've actually made a ton of progress on this, and I'm pretty sure this is also just automation and AI. But I actually used to have Codex make me a PowerPoint every Wednesday night of what's shipping for the rest of the week, right?

249. 38:48

     And that's useful, and then you just say, "Great, it's useful for me. Now let me make sure I can just post this on the channel as a Slack message."

250. 38:54

     And now you're, again, using your skills not only to benefit yourself, but benefit your entire team. That's kind of like the plugin hero mindset.

251. 39:03

     And this one's pretty funny too. Uh, I have an example. Let me just double-check this slide. Yeah. I have another example which has also been pretty crazy, which is the one I gave during the keynote, which is I had been editing, like, a short film in iMovie, and on my bike ride, someone gave me feedback about the

252. 39:21

     video. So I just went on my phone, and I just, like, sent, told the AI, like, "Okay, there's a file somewhere. Can you just, like, find it in iMovie?

253. 39:27

     There's only one iMovie project. Read the Slack message. Export the video. If you c- if the Slack MCP server does not allow file upload, use Computer Use to upload the file, and then watch that thread every hour, and if they have any feedback, re-export the video and re-share it."

254. 39:44

     And then, like, biked home. And by the time I got home, it was like, "Oh, by the way, it was actually way easier to use a Google Drive connector," so I've just been uploading the same file on Google Drive instead, so they only have, like, one URL to manage.

255. 39:56

     And, uh, yeah, we, like, fixed a bunch of stuff in the typography, and then we shipped it. S- like, again, like, kind of mind-blowing stuff. It's very simple, but mind-blowing, right?

256. 40:06

     But that's what a heartbeat is, right? A heartbeat is just a way of waking up your thread over time to take some actions. The other thing you can do is set goals.

257. 40:14

     So /goal is pretty amazing. It basically defines a verification step and says, "Okay, as long as this is running, check this verification step.

258. 40:25

     If, if it's not done, keep going." A very simple idea. Um, and as long as it is a verifier, it does really, really well.

259. 40:35

     For example, I've just been, uh, migrating a bunch of software into Rust,

260. 40:41

     right? It's like if this is a Python project that is amenable to be wr- rewritten in Rust,

261. 40:46

     /goal migrate the back into Rust, make sure all the unit tests pass. And I was able to not only rewrite the rich terminal library in Rust, I also rewrote UV and TypeScript, um, just to see if I could.

262. 41:00

     And, uh, we're, like, a hundred percent te- test coverage. It's, it's pretty amazing. Obviously, you should not be doing this in your work, but it's, it's very helpful to just understand that as these systems have better verification, you can make a lot of progress.

263. 41:14

     In the, in the monorepo, I've also included a skill called Ultra Goal, and all it does is instead of setting the goal in the app, we set it in a file, right?

264. 41:25

     So we have, we have a goal MD file. And what that means is you can edit the goal while it's being run, so you can add more scope, uh, just like many, many real projects do.

265. 41:35

     We also define a plan, again, that we can reference. But the benefit of this is as you're learning more about the projects and you're changing the plan and the goal, as these, as these models are just, like, looping, uh, it can update its understanding of the system.

266. 41:50

     And then sometimes I have, like, a state MD file or a work log just to track these, like, longer-running tasks. Like, if, if things are running for, like, a day or two, I wanna know what's going on, and I'm never gonna read this, like, four gigabyte, like, you know, session JSON object.

267. 42:03

     But I can look at the r- work log, have it, have another model summarize it using, like, a side chat and go from there.

268. 42:11

     Uh, this meant to say remote control. But again, if folks are just, like, on their computers, I also recommend trying that out. In the sidebar, there should be a button that says Remote Control, especially if you have the iOS app.

269. 42:22

     This is the thing where we talked about being able to control your phone, so control your computer through your phone, right? So if you go on the iOS app, you enter Codex, you can, you can do this, like, flow where you can scan a QR code, and all of a sudden your, your, your, your, uh, ChatGPT app

270. 42:38

     can message and queue- Any thread in the, in the application, including, I think, remote threads. Uh, this is super powerful because again, oftentimes, you know, every time I try to leave the house, someone's like asking me for something, and now I can just ask Codex.

271. 42:54

     Uh, really I should just be having something that monitors Slack and just does it, but like I, you know, I'm not there yet.

272. 43:01

     But again, this is one of those big, uh, you know, feel the AGI moments,

273. 43:06

     right? Yeah. And again, like I said, the chief of staff thread effectively is like the single source of truth for basically what's going on in, in, in my life.

274. 43:17

     On my personal computer, I have a different one. Uh, lots of good stuff.

275. 43:23

     This is all I wrote for mine. Create a p-- create and pin a chief of staff thread every day at eight AM. Check all these connectors, figure out what's going on, and then, uh, you know, do a good job.

276. 43:34

     And again, you'll just start-- You'll start editing it over time, right? Maybe you don't like the formatting or you wish you included links. Uh, for a while, what I made it do was if it found all the emails, not only to ask it to draft the responses, but I would make it open a Chrome tab for every

277. 43:49

     email I need to reply to in Chrome. And so I'll open my computer, I'll take a meeting, I come back and on my computer is just like seven Chrome tabs, and I can just review the drafts and send each one.

278. 43:59

     Like small things like this just to prepare your computer while your meetings are happening. Uh, really productive. So now we talked a little bit more about like just like doing the work itself, right?

279. 44:08

     We haven't really gone into things like artifacts just yet, but I'm also curious if anyone has any questions on how I've been doing things so far.

280. 44:15

     Over here.

281. 44:16

     So I noticed-- So I do a lot of meta prompting-

282. 44:19

     Yeah

283. 44:19

     ... um, with GPT. I noticed that you would have very, uh, small prompt.

284. 44:25

     Yeah.

285. 44:25

     And so, but when I meta prompt, I get a lot of stuff from GPT.

286. 44:29

     Yeah.

287. 44:30

     So what's your advice?

288. 44:32

     I generally always prefer to have the model write the goal or write the, write the prompt itself.

289. 44:39

     Like more and more of these models are just getting better at doing that, and it's like more into distribution to what they want. Um, in reality, I just-- I will send like a ten-minute voice memo, right?

290. 44:50

     I'm just like, like, you know, this is some issue that's happening. I think there's a project about some thread. Uh, I don't know if they got back to me.

291. 44:58

     I think their name is Dylan. Uh, please look in Gmail. Like take a-- It's like a really, really messy...

292. 45:07

     And then it's like, okay, well, do I want to make it set a goal? Do I want it to create a new thread? Do I want to make a new skill?

293. 45:13

     Like that's really sort of where my taste lies now. It's like, okay, how do I wanna organize what the work product is? You know, do I tell it to then do all this work and put it into a index.html to share it?

294. 45:25

     Do I want to make a Word doc? That's basically it, but I generally just send very long messages. Yeah. Like for example, here, like this is not the prompt for the chief of staff thread.

295. 45:37

     This is the prompt so the model can make the chief of staff thread, right? And that model will be much b-- Th-that output will be much better at determining how verbose the, the, the automation is or how often it should check or what connectors.

296. 45:51

     Um, and then because I have this monorepo, one of the things I do is every project file has a link to every Slack channel this project is relevant for, and the model will just see that and it'll go read those Slack channels, right?

297. 46:06

     Every like person.md file has their email address, their Slack connector, multiple work addresses, and it'll read all of those things. I would never prompt that myself. Yeah.

298. 46:18

     Did you have to create them yourself or?

299. 46:20

     Yeah. Yeah. So I think in that example, what I realized was wh-when I mentioned the Slack channels that are relevant for a project, the results were better. So I just added that in the front matter of like the markdown file.

300. 46:30

     But these are the things that you grow over time.

301. 46:33

     So did I get it right that the only diff to something like our Hermes is that it's not-- it's, it's still multiple threads you're opening, uh, versus you have just one thread and lots of context, et cetera.

302. 46:46

     So is the only diff with all the markings exactly the same?

303. 46:51

     I mean, I've not tried like the deepest... So the question is like what is the difference between this and like an OpenClaw and a Hermes agent? I think,

304. 47:00

     I think-- I'm sure there's differences I'm not aware of right now, but I can imagine a world where it's gonna get much, like very, very similar very quickly, right?

305. 47:07

     Like most of my work, and I'll talk about this later on, is like my threads manage themselves, right? It could be the same thing as a sub-agent. Um, I don't know how many people use Hermes agents and OpenClaw for like very wide work, right?

306. 47:22

     Like I think I, like I think I tried OpenClaw to do some like house automation stuff.

307. 47:26

     Yeah. I mean, so the, the idea would be you only have one thread-

308. 47:29

     Yeah

309. 47:30

     ... and there's some meta layer that parks the threads or whatever. So you can just park them when you go get some-

310. 47:36

     Yeah. So this is the que-- the comment's like, yeah, with, with these models, there's only one thread. I think that's very reasonable. Just in, in reality, there's just, there's just like so many things that we work on that like I need the organization, right?

311. 47:47

     It's like, you know, it's like, is there a world where like my banker and my therapist and my personal trainer and my girlfriend is the same person? Like maybe, but like my like, you know, s- tiny brain can't figure that out.

312. 47:59

     And like it's easy for me to understand what the work is by making folders, right? It's like the computer doesn't know the folders exist, but the folders are for me in, in some ways, if that makes sense.

313. 48:10

     Right.

314. 48:10

     Yeah. Cool. Great. One more que-- Yeah, one question. And then that one. Yeah.

315. 48:18

     Um, so the other thing is, is one of your slides had like, uh, three connectors-

316. 48:23

     Yeah

317. 48:23

     ... so Slack, Gmail, et cetera. Um, and then you said after a while you can just like, uh, ask the prompt-

318. 48:30

     Yeah

319. 48:30

     ... um, to do things. How long does it take to do that? Um, because like I, I go ahead and use-

320. 48:36

     Yeah

321. 48:36

     ... Slack and, uh, connectors, et cetera-

322. 48:38

     Yeah

323. 48:39

     ... skills, but like, uh-

324. 48:45

     Yeah. Um, I think it depends. I think it depends on how proactive you are in, like, telling the AI to remember these things. So, for example, like, once I realized that I should include Slack channel IDs in, like, project documents, the model's like, "Oh, there's a Slack channel.

325. 49:00

     I should read the Slack channel." And that became really obvious. Now, I, I basically never tag things. I don't really know when that happened. Um, but yeah, again, a l- a lot of it is just getting in the habit of

326. 49:13

     remember this for next time. Update the skill for next time. Update the agent.md doc for next time. And that is the- [inaudible]

327. 49:21

     skill then?

328. 49:21

     Uh, sometimes.

329. 49:22

     Or is it memory?

330. 49:25

     I mean, I think it would be hard for me to, like, prove it. Like, there's also a memory system that's, like, outside of the docs. Generally, I'm, like, pretty happy with the memory.

331. 49:33

     I, I don't know if, if I can, like, give you, like, a time estimate, but I think I'll just try it out, right? Like, use it for a couple of weeks.

332. 49:41

     Have-- Make sure your memory is turned on, by the way. Like, it's also in your settings. I know our settings panel is, like, pretty crazy right now. But, um, yeah, I think I would just turn memories on and just see whether or not these changes happen over time.

333. 49:52

     'Cause, like, I basically never-- I don't remember a single time I've, like, u- like, @ mentioned something. Um,

334. 50:00

     yeah. And the last part's pretty, pretty fast, right? So we talked about reading context. We talked about working with the context. Now the, the last thing I gotta do is just, like, write context.

335. 50:09

     Um, most of this has been pretty, pretty, uh,

336. 50:13

     simple, right? You can draft emails. You can draft Slack updates. If you feel very brave, you can send them. But, like, be-- please be respectful. Like, um, I'm sure Charlie has gotten hundreds of sent from ChatGPT, uh, Slack messages from me over time.

337. 50:26

     Uh, but I hope they're-- they sound like me more now. Uh, building one-pagers is also pretty good. Like, I-- this slide deck was made with, uh, Codex, and soon we'll be able to do things like serve applications.

338. 50:39

     And now I think, you know, at least internally, so much of the work that we do has just been sharing, like, apps rather than, like, full documents.

339. 50:49

     I don't know how many people know about this, but we also have a really good, like, artifacts ecosystem. Like, more and more, uh, Codex has become a tool for all of the work that I do.

340. 50:56

     It can open and render, like, Excel spreadsheets, Word documents, uh, PDFs, slides. And with the annotation tool, uh, editing things is, like, pretty fun. So even with these slides, this is, this is actually served on the in-app browser of the Codex app.

341. 51:12

     And what I'll do is I'll just give my talk, and I'll press next and press next. And then when I don't like something, I just select it and say, "Hey," like, "fix this.

342. 51:18

     I don't like the white space. Like, these two slides need to be broken up two more slides." I hit enter. As Codex is working to clean this up, I'm just going down the-- get down the slides.

343. 51:27

     And it's, it's generally been a pretty, uh, natural way of working. Like, this, this deck really came from me reading my own blog post out loud and then generating the material as we go along.

344. 51:39

     But, you know, it's still, it's still, like, two or three skills to make the slides look this way.

345. 51:45

     Yeah. Um, yeah, and then once you do that, you can do again, again. It's, it's the same concept over and over again, right? You can, you can build out these loops that just touch other parts of the system.

346. 51:56

     Like, most of our project trackers are just Google Sheets updated with loops. Uh, these slides are loops. And the annotations-- Yeah, I j- I just said everything. My bad.

347. 52:05

     Um, and then one thing that's also very helpful is, like, as the company gets more, like, context dense, right? Maybe it's all A- OA agents. Just the ability to, like, summarize things, like, over Slack has been incredibly useful.

348. 52:21

     That was clearly, like, a slop slide that, uh, ChatGPT added in.

349. 52:28

     So once you can take actions on these different artifacts, I think the, the biggest thing, and the thing I really want people to try out is just Computer Use.

350. 52:35

     Um, again, it's like, it's like the first time I had that, like, feel the AGI moment, right? Like, a cursor is, like, trying to do some action. You can see, like, move across the screen.

351. 52:44

     And, like, when it does it really well, like, you really gain a lot of faith in the system. So we, we obviously have, like, plug-ins, right? And this is for sending Slack messages.

352. 52:54

     But then the in-app browser is gonna get even more powerful soon, right? We're gonna be able to basically treat this like the browser that you use. Like, I now try to use the in-app browser as much as possible.

353. 53:06

     And then for everything else, use Computer Use.

354. 53:09

     Um, how many people have used Computer Use, by the way? Very f- Oh, it's like ten percent of you. What's the crazy thing that you've done? What's, like, the craziest thing anyone's done?

355. 53:19

     Any, any volunteers?

356. 53:21

     Managing my home lab.

357. 53:23

     Home lab?

358. 53:24

     Yeah.

359. 53:24

     What's a home lab?

360. 53:25

     Uh, thirty agents I have, like, around... I have a DGX park. I have, like-

361. 53:30

     Oh

362. 53:31

     ... a bunch of, uh, boxes.

363. 53:33

     Nice.

364. 53:33

     And essentially thirty agents, and I can call it and just do stuff. I have a, a Mac Mini-

365. 53:40

     Mm

366. 53:40

     ... um, that I have Codex installed on. I actually have three of them in my home lab, and essentially they each do different things.

367. 53:46

     Wow. Any other, like, crazy Computer Use stories?

368. 53:53

     Gotta g- gotta get AGI pill and try out Computer Use. Um,

369. 53:59

     yeah. One thing I'll say is, like, because computers are so powerful, like you get-- like you mentioned before, there is some, like, safety component. I am now remembering this example where, like, again, because the Slack connector was not able to upload files.

370. 54:10

     If this model is really determined, right, uh, it could be like the one wish willow. It just says, "Okay, great, well, if I can't add a Slack m- uh, file, let me, like, go on Computer Use and press File Upload and do that," right?

371. 54:22

     There will be some times where the model, based on how you prompt it, become- becomes really determined and say, "Oh, like, I, I, it seems like I can't email someone using the Gmail connector.

372. 54:32

     Let me open up Chrome and hit the Send button." So those are the kind of things that you should be really wary about, right? These are, like, real security issues.

373. 54:39

     And again, um, m- more than not, like, having things like the agent.md file has been really, really helpful, especially if you do things like, uh, guardian mode or auto mode, excuse me.

374. 54:51

     Um, and it has also really changed the way I think about, uh, doing work. I feel like now when I'm doing things like building an application, if it's a native application, most of my testing is just done by Codex using Computer Use.

375. 55:04

     If it's a website, again, it's just using the in-app browser.

376. 55:08

     Um, I think we talked a lot about this already, like handle service work. Like,

377. 55:13

     it's kind of awesome just, like, be in a checkout page, AppShots, pshh, you know, find me a coupon. Like, it's made me more money than, than I, I would've, uh, expected earlier.

378. 55:22

     Uh, filling out forms, testing applications. Um, I haven't really had a good use of this just yet, but it can also control the iPhone through scr- screen mirroring. So, uh, do with that what you will.

379. 55:35

     And this is kind of sort of the escalation of the talk, right? Then the question is, like, what can the computer not do, and like, why can't it do it?

380. 55:42

     And one of the things that, like, you can't do is control Codex. But you don't really need to. Because these Codex threads can already talk to each other, uh, you-- it can already control itself in very powerful ways, right?

381. 55:57

     If you think of the example... This is not the slide I want.

382. 56:02

     Damn. Okay. I think, I think I, I messed up some slides.

383. 56:06

     Earlier in the talk, I talked about this idea that if there was some kind of support issue, I could take an AppShot, and it'll call this, like, DX triage skill.

384. 56:14

     And I will-- it will rename the skill, it'll do the loop, and it, it communicates to, like, Slack and Twitter, and it's very nice. But I still need to be the person that triggers these things, right?

385. 56:23

     That's the same thing as me seeing an issue and making a pull request.

386. 56:27

     In the future, like, I mean, the future for you, it's happening already now here. But n- now what I just have is I just have a single, like, monitor thread, and any time it identifies any of these issues, it will go off and create a new thread, and the thread's job is to do all this triage.

387. 56:45

     And what might happen is maybe this triage is waiting on someone on Slack to acknowledge this issue. Maybe a pull request has been created, but it has not merged.

388. 56:53

     If someone else complains again in the future,

389. 56:56

     the monitor thread just goes, "Oh yeah, I think this is the same issue." Not only is it the same issue, let me send a message to that, you know, downstream thread so that it's aware this issue is recurring.

390. 57:06

     Maybe that thread will send a Slack message. But in the main thread, I'll still see a message that says, "Hey, it's been like the third day this issue has been live, uh, based on Twitter feedback.

391. 57:16

     Should we do something?" And like, those are the things I'm, I, I'm, I'm... It, it's very hard for me to keep track of these things 'cause I'm just, like, on Twitter all the time.

392. 57:25

     But by the agent being able to just manage and consume all this information, it makes these things much more tractable. Um, I definitely messed up the slides over here, so I'm gonna skip a couple slides.

393. 57:37

     Do, do, do. Yeah, this is, this is the, uh, example.

394. 57:42

     Um, and, and I really want you to, uh, I really, really want you to play around with this idea. I, I don't think it's been fully baked yet, right?

395. 57:51

     Most of my work is about just having monitors create sub-threads. These threads are then managed and pinned onto the sidebar. Like, that's also one of the best things, right?

396. 57:59

     Like, with a sub-agent, the thread just has these, like, shapeless entities floating in the background. Uh, like, you know, the JSON thread or like the Galileo thread or whatever.

397. 58:12

     The difference with sub-agents versus these threads is because they show up in the sidebar, you can kind of just notice that something has changed. You know there's a new issue that's come up, right?

398. 58:19

     And a lot of it too is, like, just using the sidebar as effectively, like, the hub of understanding, like, what are the ongoing, like, work streams. And this has been, like, super powerful for me.

399. 58:30

     But it all just starts from pinning a thread, taking actions, having this monorepo to sort of manage all of your context, and then having different ways of waking systems up.

400. 58:39

     And earlier, these systems wake up because you messaged it or you set up a heartbeat, and now these things can be woken up by another thread running somewhere else.

401. 58:49

     And so more often than not, like, most of my automations just happen on the monitor level of threads, and they trigger and create new threads and then manage themselves.

402. 58:59

     Yeah. So I think we talked about a bunch of things. We talked about Computer Use, structured plugins,

403. 59:05

     and, uh, yeah, I think that's basically it. Try out some Computer Use stuff. I, I know I can't really... I'm, like, really worried about the Wi-Fi here if I'm gonna try something too bad.

404. 59:17

     But, um, yeah, try out AppShots, try out Computer Use, and really play around with what, uh, these models are capable of. There's a question over there. I'll do that one next.

405. 59:27

     Yeah. I wonder if, like, um, you could show us, like, parts of or what your monorepo looks like.

406. 59:35

     I don't think I can show you on this computer. They're gonna, like, take me away.

407. 59:40

     Um, I mean, I, I can talk about a little bit. So, so basically, my ro- monorepo is set up so that it looks very much like the one over there.

408. 59:49

     I have a projects directory, and in that projects directory is a named directory for every work stream I'm working on. So maybe it is the, you know, voice launch video, or it is the, um, some- like the, the agents SDK, right?

409. 1:00:04

     Or it is, uh, the Codex open source grant program. So those are sort of some of the projects. Then I have a people directory, which is just, like, every single person that's ever DM'd me, right?

410. 1:00:16

     I know what they're working on. I know what kind of problems they're thinking about. I know what other, like, uh, like, side channels they're a part of. And then I have a bunch of different, like, loose notes, agent summaries.

411. 1:00:28

     I have, like, daily summaries of what I've done. This is mostly just to test the limits of AI. I don't think it's very useful for that kind of stuff.

412. 1:00:34

     Mostly it's just the projects and the people, right? Um, and then I have a to-do list that an agent just maintains. So I have, I have a single thread that just, like, checks the to-do list.

413. 1:00:44

     Oh, this says it was undone. Let me have a sub-agent verify that, like, no one has done this task. Those things are pretty token ex- token expensive. I don't really know if it would be worth people do- setting this up for themselves unless they just, like, don't use that many credits.

414. 1:00:59

     Um- But I, I think the biggest ones is just like the Shiva staff thread. Nine AM, tell me what's happening today, tell me what's happening this week. And I think if you just do that, you're, you're gonna get a lot of, uh, lot of juice out of that.

415. 1:01:11

     Are you not getting crushed by, like, context flows or, like, slop as you're stuffing data back into the model?

416. 1:01:19

     It's hon- like, honestly, I just-- I've never experienced it. I think...

417. 1:01:27

     Like, I think, I think the compaction is just, like, really, really good. Um, I know that's not a very satisfactory answer, but, like, if I could improve parts of the model, I would rather improve, like, its writing tone rather than, like, its ability to search the context.

418. 1:01:43

     Um, generally, it's been, it's been, uh, pretty good. I think it, it might have to do with, like, the way the Codex memories have been set up, but I have not dug in too much into those details.

419. 1:01:52

     I think you had a question over here.

420. 1:01:54

     Yeah. Another question on the memory management aspect.

421. 1:01:56

     Yeah.

422. 1:01:57

     So one of the things I noticed a lot is memory bleed across projects.

423. 1:02:00

     Yeah.

424. 1:02:01

     Um, do you have any tips on how to manage that?

425. 1:02:05

     Because I've seen a lot of-- Especially when you enable memory-

426. 1:02:09

     Yeah

427. 1:02:09

     ... it tends to take learnings from some other project or a current project, which is typically not good.

428. 1:02:16

     Yeah.

429. 1:02:16

     It's not just a Codex or a ChatGPT thing. This is, like, across the-

430. 1:02:22

     Yeah. So I think the question was just, like, how do you deal with memories, like, bleeding across different projects.

431. 1:02:29

     I don't think I have a good answer to that, primarily because it's unclear to me what the downstream side effect would be. Maybe, like, one project uses, like, NPM, another project uses, like, Yarn or something.

432. 1:02:42

     But I mean, generally, I just clean those up in the agent MD files for those specific projects. Right. So another thing to mention is, like, every project directory has its own README and its own agent MD files.

433. 1:02:53

     And so yeah, like, I think sometimes, like, one project I was working on was, like,

434. 1:02:57

     NPM, and one was, like, PNPM. I just clean that up in the agent's MD. Now, it's-- I would be curious, like, how much of that can be cleaned up just by using those simple tools.

435. 1:03:05

     But talk to me afterwards. I'm really curious, like,

436. 1:03:08

     what's so bad about the, the bleeding.

437. 1:03:12

     Yeah.

438. 1:03:13

     Yeah. Yeah, maybe that, that would be a good example of, like, having project-level scoping versus a, a single thread. But, um, yeah, I might have to look into more details about how the memory part is actually running.

439. 1:03:25

     Any other questions? Over here.

440. 1:03:27

     Did you, did you change your Codex config files so that it would-- so that you could do a monorepo type of environment? How did you set that up?

441. 1:03:36

     No. So I think, um, honestly, at this point, it just kind of knows. It's, like, such a, such a bad answer, but it's almost like an AGI. Like, I feel like-- To me, it's-- it feels like a very AGI-filled thing.

442. 1:03:47

     But I think generally in the beginning when I was working with it, I would have a skill called, like, check notes. And it's like, hey, if you feel like you need to check your notes, like, check the notes.

443. 1:03:57

     All right. And so if I f- if I had a hunch or some intuition that the model might not be able to do this, I would just mention, like, hey, check the notes.

444. 1:04:06

     Um, but even, like, if you look at my pr-- like, uh, OpenAI profile, I think, like, my, like, check notes skill has been used, like, a h- a hundred fifty thousand times.

445. 1:04:17

     But I've, like, never mentioned it in the past, like, two, three months. It's 'cause again, I think this, the memory system has been, uh, doing a lot of the heavy lifting.

446. 1:04:24

     But again, it's, it's like onboarding an employee. In the, in the first, you know, two, three months of onboarding an employee, you have to give a lot of instruction.

447. 1:04:32

     You have to give a lot of-- You have to give them a lot of context, and you have to let them fail. And when they fail, you have to give them an opportunity, like, write down what they've learned and, like, clean up itself.

448. 1:04:42

     Um, I would be very surprised if you got good results in the first, like, day of setting this up. But I mean, just, yeah, maybe not to your surprise, but now it's like, I don't-- it kinda just works, right?

449. 1:04:53

     And that, that's a great feeling to have, right? 'Cause it, it, it lets me go back in the flow of just, like, doing my job.

450. 1:04:59

     You have a question?

451. 1:05:00

     Yeah. Uh, thanks so much for the talk. Um, I wanted to ask, so I'm a college student studying CS at Stanford, and I'm gonna be entering the workforce soon, right?

452. 1:05:09

     Mm-hmm.

453. 1:05:09

     Given that it's so rapidly changing and evolving, what advice do you have for college students who, like, uh-

454. 1:05:15

     Develop? Like, just-- So not productivity, but just, like, coding in general?

455. 1:05:18

     Yeah. Like, 'cause now it's not so valuable to code the hammers like you were saying.

456. 1:05:22

     Yeah.

457. 1:05:23

     Um, kind of what would your advice be to college students who are entering the workforce?

458. 1:05:28

     I'm so old. That's so long ago. Um...

459. 1:05:39

     Like, I have a line that's like, if you wanna have good taste, you kinda have to eat, right? Like, I think your job is to, like, consume a little bit more.

460. 1:05:46

     Like, try out different applications. Like, you know, like, do you know what a good onboarding flow looks like? Do you know what that feels like? Have you built an app that makes you frustrated?

461. 1:05:55

     And just, like, your ability just to consume more things and

462. 1:05:59

     develop your vocabulary on, like, how to complain about things that are bad. How do you, like, complain about, like, the slop? Those are the thing-- those are the skills I think will be very valuable.

463. 1:06:08

     And specifically around vocabulary. Right? Like, you can't really describe things that you don't really understand. Um, and this happens in, like, both, like, things like cooking, right? Like, if you just don't know what the ingredients are, you can't describe that something is too salty, or it could be-- it's like it lacks some acid.

464. 1:06:24

     Like, you just need to, like, consume a little bit more to-- and then think critically about why you like something and why you don't like something.

465. 1:06:31

     Um, 'cause I think, I think taste is the-- is a big issue, but I think a lot of it is because we are not consuming the right, like, stuff, right?

466. 1:06:41

     Like, try out all the different apps i-if you're thinking about apps, for example. Yeah.

467. 1:06:47

     That's what I think. Thank you.

468. 1:06:49

     Any other questions? Yep.

469. 1:06:56

     Um, you, you had, like, your chief of staff-

470. 1:06:59

     Yeah

471. 1:06:59

     ... all these other things.

472. 1:07:06

     Yeah

473. 1:07:06

     ... a thread and then afterwards it goes down. Can you demonstrate how you do that? Like, how do you configure that?

474. 1:07:13

     Uh, which part? So-

475. 1:07:15

     So, so, like, um, talking to one thread and then afterwards it's going to a different thread to do things. So I understand like you're in a thread and, um, spins up sub-agents.

476. 1:07:26

     Yeah.

477. 1:07:26

     I get that. But I think, uh, at, at least I believe what you were talking about is having like a chief of staff-

478. 1:07:34

     Yeah

479. 1:07:34

     ... that's maybe going to an IC.

480. 1:07:36

     Yes, yes.

481. 1:07:37

     And that's like a thread.

482. 1:07:38

     Yeah.

483. 1:07:39

     And how, how did you set that up?

484. 1:07:40

     Yeah, I mean, like un-ironically that slide just says like, "Threads can talk to each other, just ask." Basically there is a list thread tool and a send message to thread tool that Codex has available.

485. 1:07:51

     And so typing that is kind of awkward, but really I just say,

486. 1:07:58

     like, "Sometime last week I was working on slides. Can you go find that thread, rename it, and pin it for next time?"

487. 1:08:05

     Oh, got it. Okay.

488. 1:08:06

     Right? Like I would never type that but again, voice input makes it really easy. Um, sometimes I'll say, for, for example, this talk I, I think I messed up some slides.

489. 1:08:14

     Basically I had a main slide writing system, uh, and I said, "Okay, like this slide could be split up in three acts. Make up a thread for each act, pin it, rename it act one, two, and three and, uh, review each section.

490. 1:08:28

     And then once it's done, review the whole slide." Um,

491. 1:08:32

     and yeah, I generally like to just ask it. I, I mean, I think I'm gonna try to add more like commands to be a little bit more explicit around thread control, but I think that will just come as the models get better and, and as the model is like more aware of what Codex can do.

492. 1:08:46

     Um, but yeah, it really is just like... Like I think one thing you should try to do is just have a thread and just say, "Read all the other threads that are pinned and like rename them and like use an emoji to color code its like readiness."

493. 1:09:01

     And just seeing it do that will be like a first step, um, to, to just like understand how the thread control works. Th- this feature is like very new but it already has been pretty wild to just sort of have a single thread and you're just talking to it for like the a- entire afternoon.

494. 1:09:15

     Yeah. Sweet. If there are no more questions-- Yeah, one more.

495. 1:09:19

     With the multithreading you said it's very new. Do you know if it's available across any other popular AIs? 'Cause I don't think I've seen [inaudible]

496. 1:09:28

     .

497. 1:09:28

     Don't say that name to me. No, uh, so the question was like do other a- coding systems um, have these kind of tools? Not yet. I think I'm sure, you know, I'm sure they will soon.

498. 1:09:40

     Uh, but as far as I know... This goes back to the eating thing where like I just have not tried like the latest versions of other tools. Like I don't know if like Conductor has these skills, but definitely, um, I'm sure again like very competitive space, these, these features will pi- propagate very quickly.

499. 1:09:58

     If you have old threads run on old models will you have the new threads go back and pull that in as context?

500. 1:10:05

     It depends. Sometimes just say find the old thread and like update the model.

501. 1:10:10

     You know, just update the model. Like f- for example, I, I, I like-- So when, when uh, four five, sorry, five four came out and then we moved to five five I realized all my old automations were just on five four.

502. 1:10:24

     And it really bugged me and I was like, "Hey guys, like can we add a feature that like has like GPT latest so we can avoid this issue?" And someone was just like,

503. 1:10:33

     "Tell a thread to update the other threads." [chuckles] [chuckles]

504. 1:10:36

     And I was like, yeah, uh, like so much of it is just ask. You just have to like develop the language to figure out what you wanna do. But I think that's again like the most important, most important thing.

505. 1:10:45

     Yeah, so just ask it and then it'll automatically do it rather than you having to go through and-

506. 1:10:49

     Yeah, like I think I have an automation that just like cleans up old threads and like by like model IDs. Um,

507. 1:10:56

     but uh, yeah.

508. 1:10:57

     Thanks, man.

509. 1:10:59

     What's up?

510. 1:10:59

     Does this live anywhere publicly?

511. 1:11:01

     Uh, not yet. I'm not happy with the slides enough to I, I'll publish them, but I have a blog post. Just search-- If you just Google Codex Maxing it's all there.

512. 1:11:12

     Uh, you know. Three Xs I think. [chuckles] Uh, but I'll, I'll share some more stuff on, on Twitter. Go on.

513. 1:11:19

     Yeah. So you said that you work with unlimited token-

514. 1:11:21

     Yeah.

515. 1:11:22

     Um, but since you work so closely with it, do you, do you have any, uh, sort of advice on how we can-- 'Cause a lot of times when you're doing like loops you get a lot of like the same message from Codex and that can eat into the-

516. 1:11:35

     Yeah

517. 1:11:35

     ... token limits pretty quickly. Do you have any advice just-- I know you don't-

518. 1:11:39

     Yeah

519. 1:11:39

     ... probably personally but since you have the expertise. Do you have any, any advice you can share on how we can get sto- we can avoid hitting those limits depending on, you know-

520. 1:11:47

     Yeah. I think the biggest misconception here is that like X high will give you the best results. And so there's like, there's a lot of like X high maximalists here.

521. 1:11:57

     I want everything on X high, right? Like, uh, like so when I'm-- I was demoing like, like getting a coupon thing and my friend like does an app shot and like hits enter and I'm like, "Why did you turn on X high?" [chuckles]

522. 1:12:07

     Right? It ran for like two minutes like searching like every website available for like coupons. Um,

523. 1:12:14

     obviously I c- I can't give that much advice 'cause it's something like on my personal computer I don't really run into limits, but really get comfortable with like low and medium thinking.

524. 1:12:24

     Like these models are still very, very smart, right? Like low thinking on fi- like five, five five is like still so much better than prior models. And for a lot of this work that's not just like, you know, make me a video game from scratch, you just don't need that m- kind of work.

525. 1:12:39

     Like my, my chief of staff I think is like default medium.

526. 1:12:43

     That's very helpful advice and I agree. But just as a follow-up, kinda I guess in, in a few words what I, I think what I was really asking is how do I tell the thread to shut up if it's just saying the same thing over and over again?

527. 1:12:53

     Oh, uh, I mean the answer's really bad but I just like, I just say like if there's no updates just reply with like one word, no updates.

528. 1:13:03

     Okay.

529. 1:13:04

     Uh, that's one thing. And the second thing too is, um, I would also play around with like how often these heartbeats are happening. You know, are you running the thing every thirty minutes?

530. 1:13:12

     Are you running this every nine AM? That's one thing and the second thing too is like do you have some kind of stopping criteria? Right. So for example like I was-- I had some argument with like Amazon and I'm just like great like they put me in a seventy-five minute wait list like check every five minutes if

531. 1:13:29

     the queue is like better and once you get to five minute wait time check every one minute and keep applying until you get my money back. And I took a shower and then when I came back I had like four hundred dollars on my credit card.

532. 1:13:40

     Do you think it's possible to set up like more dynamic heartbeats with the model to adapt to how fast it needs to be?

533. 1:13:47

     I've not tried too much there but it's all possible, right? Because to create a heartbeat you just edit a text file. And so you should definitely be able to create your own heartbeat and then change how frequent or infrequent.

534. 1:13:59

     Um, I might try to do that actually. I might just say like, "Hey like during the weekdays like change your heartbeat to be more active and during weekends or in the afternoons change them to, um,

535. 1:14:11

     be less active." Yeah. Like I've tried versions of the chief of staff thread where I just set a goal that says like never stop and you're only allowed to like set sleep and that has also worked pretty well.

536. 1:14:22

     It'll be like, "Hey it's like nine PM Jason like I haven't seen Jason post a Slack message in like two hours I'm gonna sleep for like five hours." Uh, and that also kind of works.

537. 1:14:32

     Yeah. But um, I, I think the biggest one honestly is just like w- people should not be afraid of low reasoning. Like X high is not like X high results it's just like think more. [outro jingle]