WEBVTT

NOTE Sentence-level transcript of https://www.youtube.com/watch?v=9arM9b7JgOo

NOTE One cue per sentence. Cue ids are the line anchors on /transcripts/9arM9b7JgOo.html. A cue ends where the next begins, or 2 s after its last word.

s1
00:00:01.080 --> 00:00:09.480
workshop setup, so um there are some people who kind of have like never installed open claw, and I want to kind of get them set up.

s2
00:00:09.480 --> 00:00:20.440
And then the other people like want to try an experimental thing, which I haven't even fully gotten working, but I think for more advanced people that'll be cool um to get like a sort of staging environment

s3
00:00:20.440 --> 00:00:21.560
and run two open claws.

s4
00:00:21.560 --> 00:00:31.960
So, I think if I'm not, you know, too busy helping new people set up, then I'll try to help people like finish debugging the last part, or maybe someone else will get it.

s5
00:00:31.960 --> 00:00:39.280
And I think uh 10 minutes Q&A, 10 minutes some more advanced topics, and then just more Q&A.

s6
00:00:39.280 --> 00:00:44.440
And uh probably after every 5 minutes or so, we can do a bit of a few minutes of discussion.

s7
00:00:44.440 --> 00:00:48.400
So, I'll just set some timers to make sure I don't go off track.

s8
00:00:49.520 --> 00:00:50.760
All right.

s9
00:00:50.760 --> 00:00:53.320
Let me start.

s10
00:01:02.520 --> 00:01:02.960
All right.

s11
00:01:02.960 --> 00:01:05.840
Can everyone see this presentation?

s12
00:01:07.440 --> 00:01:08.160
Yes.

s13
00:01:08.160 --> 00:01:09.560
All right, cool.

s14
00:01:09.560 --> 00:01:11.120
Yeah.

s15
00:01:11.120 --> 00:01:16.240
So, this kind of describes like my setup.

s16
00:01:16.240 --> 00:01:25.120
Um I'll just keep it really high level for this part where when everything works, you know, maybe like 70% of the time, you describe your task.

s17
00:01:25.120 --> 00:01:31.280
And then what's nice is open claw will have like context and memory about what you're talking about.

s18
00:01:31.280 --> 00:01:35.520
So, for example, like let's say I'm like, "All right, fix this thing."

s19
00:01:35.520 --> 00:01:41.440
It's a very brief Slack message, but open claw can remember like what I've asked it to do before.

s20
00:01:41.440 --> 00:01:47.160
So, for example, I'm working on this thing I'm calling a skeptic agent, and it's a custom code review thing.

s21
00:01:47.160 --> 00:01:49.680
But if I'm like, "All right, fix the skeptic agent."

s22
00:01:49.680 --> 00:01:50.360
or whatever, right?

s23
00:01:50.360 --> 00:01:53.040
It kind of knows what that is, which is nice.

s24
00:01:53.040 --> 00:01:54.880
You don't have to re-explain a lot.

s25
00:01:54.880 --> 00:02:00.160
Um I use multiple agents with work trees, that's very key for parallelization.

s26
00:02:00.160 --> 00:02:07.840
Uh, CI code good integration.

s27
00:02:07.840 --> 00:02:16.720
But you can kind of see you can talk to it and be pretty brief and then, you know, usually get reasonable outcomes, but sometimes you'll have to debug it.

s28
00:02:16.720 --> 00:02:19.000
So, then it's like, all right, there's some time out thing.

s29
00:02:19.000 --> 00:02:22.680
I'm like, give it a longer time out, right?

s30
00:02:24.000 --> 00:02:28.920
And then I asked that here like like uh, you know, what are the priorities and what we're working on.

s31
00:02:28.920 --> 00:02:31.000
And I'm reading it here.

s32
00:02:31.000 --> 00:02:38.520
And then I might have just said like, uh, I just made code for it, right?

s33
00:02:38.520 --> 00:02:43.096
So, another interesting thing about this is you can kind of see from like, um,

s34
00:02:43.096 --> 00:02:43.880
[clears throat]

s35
00:02:43.880 --> 00:02:48.640
the types of responses I give, they're not that like, um, special.

s36
00:02:48.640 --> 00:02:51.840
As in I I actually think an agent could replace me.

s37
00:02:51.840 --> 00:02:56.800
So, that's kind of what I'm working on right now to like get things even more autonomous.

s38
00:02:56.800 --> 00:03:02.239
You know, and you and you'll see this with LLMs like you're working with them and they'll be like, "Hey, I've got this thing ready."

s39
00:03:02.239 --> 00:03:06.040
And I'm like, "Okay, we'll run the test, right?"

s40
00:03:06.480 --> 00:03:07.239
Uh, all right.

s41
00:03:07.239 --> 00:03:10.400
So, I'll share the next slide now.

s42
00:03:10.400 --> 00:03:12.400
Oops.

s43
00:03:16.920 --> 00:03:17.760
Okay.

s44
00:03:17.760 --> 00:03:19.560
We're going to some concepts right now.

s45
00:03:19.560 --> 00:03:20.760
Like, what is Open Claw?

s46
00:03:20.760 --> 00:03:28.680
So, I think what's cool about Open Claw is it's not just about a particular repo or the code, but more of the concept.

s47
00:03:28.680 --> 00:03:36.040
So, you know, I've got this cool super diagram, but first concept is frictionless communication.

s48
00:03:36.040 --> 00:03:47.640
So, I know we have other things like Claude co-work or whatever, but I think they still have a similar concept where it's like, can you like easily talk to your AI versus like you've got a,

s49
00:03:47.640 --> 00:03:51.400
um, remote desktop into your computer or you have to go sit down.

s50
00:03:51.400 --> 00:03:56.560
So, that kind of enables certain really higher level of velocity.

s51
00:03:56.560 --> 00:03:59.160
Um then I think the central axis is very important.

s52
00:03:59.160 --> 00:04:07.840
So, you know, there's a spectrum like I kind of go like a little more cowboy and I give it act.

s53
00:04:12.000 --> 00:04:16.799
Open Claw has these agent orchestrator managers that manage your workers for you.

s54
00:04:16.799 --> 00:04:21.320
Uh or sometimes when I want more personal control, I use tmux terminals.

s55
00:04:21.320 --> 00:04:27.800
So tmux is like a terminal program, but it's pretty good for parallelization and AI development.

s56
00:04:27.800 --> 00:04:33.040
Then I use the open source framework that I forked agent orchestrator to do the workers.

s57
00:04:33.040 --> 00:04:37.760
And then once it gets to here, this part is like not really controlled by me as much anymore.

s58
00:04:37.760 --> 00:04:49.160
Um so, I'm calling them managed agents, but basically like you have a worker that runs Claude code and Claude code itself can run agents and those can have sub agents.

s59
00:04:49.160 --> 00:04:51.320
So once you get to this Claude code part, right?

s60
00:04:51.320 --> 00:04:58.760
This part is not exactly my stack, just stuff I use, but this part of the stack I'm changing a lot more.

s61
00:04:58.840 --> 00:05:03.960
Um I'll pause here to see if anyone has any questions or thoughts.

s62
00:05:06.400 --> 00:05:07.520
Um can you hear me?

s63
00:05:07.520 --> 00:05:08.640
So I I have a question.

s64
00:05:08.640 --> 00:05:17.800
So the the real question is, you know, why why use Open Claw versus directly use Claude?

s65
00:05:17.800 --> 00:05:20.200
So yeah, that's that's definitely a good question.

s66
00:05:20.200 --> 00:05:20.953
Um

s67
00:05:20.953 --> 00:05:21.800
[clears throat]

s68
00:05:21.800 --> 00:05:26.520
The reason I use Open Claw as specialization.

s69
00:05:26.520 --> 00:05:35.800
So when Open Claw makes a decision, I want that context to be more about like the spec or the goals or like the history of what I want in the task

s70
00:05:35.800 --> 00:05:37.520
rather than the code.

s71
00:05:37.520 --> 00:05:39.160
And you can just imagine this, right?

s72
00:05:39.160 --> 00:05:46.440
As soon as you open up Claude, it reads Claude MDs, it reads skills, it reads um MCPs.

s73
00:05:46.440 --> 00:05:49.720
A lot of those things are sort of independent of like the actual task.

s74
00:05:49.720 --> 00:05:51.960
It's more about how to do the task.

s75
00:05:51.960 --> 00:05:57.200
Um so imagine like 25% of your context already taken up by implementation.

s76
00:05:57.200 --> 00:06:00.840
Versus like when you think about open claw, you're like, okay.

s77
00:06:00.840 --> 00:06:12.520
I want to think about exactly what I want to do and how it relates to all the other slacks Jeffrey sent me in the last 2 weeks and put it all together and give me like a reasonable spec.

s78
00:06:16.200 --> 00:06:19.560
Does every single thing a player does, but it does it without a browser, right?

s79
00:06:19.560 --> 00:06:24.160
So you could use MCP, which usually is JSON or whatever or HTTP.

s80
00:06:24.160 --> 00:06:26.040
Doesn't really matter actually.

s81
00:06:26.040 --> 00:06:28.640
And then the final thing is the browser test.

s82
00:06:28.640 --> 00:06:30.600
So some things like are visual.

s83
00:06:30.600 --> 00:06:33.520
I mean, maybe you could even have CSS tests or JS tests, right?

s84
00:06:33.520 --> 00:06:35.200
And that kind of goes here.

s85
00:06:35.200 --> 00:06:37.560
But then the final thing is visual.

s86
00:06:37.560 --> 00:06:45.480
So I would recommend you like try um similar approaches to this and be like, okay, like which problems truly needed that human or not.

s87
00:06:45.480 --> 00:06:53.040
Um and I think what's kind of cool is like every quarter or you maybe even every month like it improves.

s88
00:06:53.040 --> 00:06:58.440
So before like I had to really manually test a lot because like you know, this worked fine, right?

s89
00:06:58.440 --> 00:07:03.600
But these things didn't work that well with agents like last year but like a year ago even 6 months ago.

s90
00:07:03.600 --> 00:07:08.800
But now like um agents are pretty good at like nailing down a lot of browser tests for me.

s91
00:07:08.800 --> 00:07:14.520
Versus like when I started my agent would have a lot of problems, you know, finding a pop-up and entering a password.

s92
00:07:14.520 --> 00:07:16.640
Now no problem.

s93
00:07:16.640 --> 00:07:21.160
Um that was a little long, but does that kind of address the theme of your question?

s94
00:07:21.160 --> 00:07:22.080
Yeah, I think this is good.

s95
00:07:22.080 --> 00:07:22.560
Good.

s96
00:07:22.560 --> 00:07:23.040
Thank you.

s97
00:07:23.040 --> 00:07:23.800
Okay.

s98
00:07:23.800 --> 00:07:24.160
Thank you very

s99
00:07:24.160 --> 00:07:26.400
Just just remind me for all these slash commands.

s100
00:07:26.400 --> 00:07:29.600
I'll put um links here instead.

s101
00:07:29.720 --> 00:07:30.200
All right, cool.

s102
00:07:30.200 --> 00:07:34.840
Anyone have any other questions related to some of that stuff?

s103
00:07:37.440 --> 00:07:38.520
All right.

s104
00:07:38.520 --> 00:07:41.760
If not, I will go here.

s105
00:07:43.040 --> 00:07:43.400
All right.

s106
00:07:43.400 --> 00:07:48.330
So, here's a two websites that I've built.

s107
00:07:48.330 --> 00:07:50.330
[clears throat]

s108
00:07:53.000 --> 00:07:56.160
Yeah, while that loads, I'll show this one.

s109
00:07:56.160 --> 00:07:58.320
So, this is like a AI RPG.

s110
00:07:58.320 --> 00:08:04.000
Um what's cool about this is you can I've got a default campaign, for example.

s111
00:08:04.000 --> 00:08:04.705
Um

s112
00:08:04.705 --> 00:08:05.440
[clears throat]

s113
00:08:05.440 --> 00:08:07.280
Yeah, pick some options.

s114
00:08:07.280 --> 00:08:09.560
You've got a avatar.

s115
00:08:09.800 --> 00:08:14.080
And uh what's cool about this is it builds a custom world for you that reacts to you.

s116
00:08:14.080 --> 00:08:18.960
So, you can like kind of say whatever you want, do whatever you want, you'll get a story back.

s117
00:08:18.960 --> 00:08:30.640
Um the main difference between playing this versus like um you know, you could just always go into the Gemini chat or or chat GPT or whatever is that um I have like a D&D system.

s118
00:08:30.640 --> 00:08:35.159
So, if you play your own like games or novels, like you kind of just win too much.

s119
00:08:35.159 --> 00:08:45.520
Versus like with this, you'll actually do like dice rolls and be like, you know, did the person actually um did you actually succeed in your action or not?

s120
00:08:45.520 --> 00:08:45.800
All right.

s121
00:08:45.800 --> 00:08:49.480
So, while this is going, I'll also um show this thing.

s122
00:08:49.480 --> 00:08:51.240
Yeah, so this is funny.

s123
00:08:51.240 --> 00:08:55.240
Um this website is like a multi-AI analysis website.

s124
00:08:55.240 --> 00:09:03.240
Um so, what I found was like whenever I was doing research or or whatever, I would go to multiple models and I'd be like, what's the answer?

s125
00:09:03.240 --> 00:09:05.040
Then I copy and paste them all.

s126
00:09:05.040 --> 00:09:06.920
I put them into one model.

s127
00:09:06.920 --> 00:09:09.920
Um so, pretty simple concept, but this does it for you.

s128
00:09:09.920 --> 00:09:14.040
So, usually I like these answers better than like asking one model.

s129
00:09:14.040 --> 00:09:17.720
And I'm like, okay, how do I do this workshop?

s130
00:09:21.880 --> 00:09:25.640
So, I can be like, okay, what's the status?

s131
00:09:30.880 --> 00:09:34.280
Uh what's the latest PR now?

s132
00:09:34.320 --> 00:09:40.240
And um what I like about this is it has that sort of vertical tab type of thing.

s133
00:09:40.240 --> 00:09:43.440
Cuz when you go horizontal, it's really easy to lose track of your tabs.

s134
00:09:43.440 --> 00:09:45.560
Um but we've got notifications, too.

s135
00:09:45.560 --> 00:09:51.160
So, when this thing is finished or this is finished and I need to look, it'll give me a notification.

s136
00:09:51.160 --> 00:09:54.480
And I can just focus on clearing the notifications.

s137
00:09:54.720 --> 00:10:01.200
So, here um the way of using these terminals, usually it's more like a manager rather than a coder.

s138
00:10:01.200 --> 00:10:08.800
Um and it's kind of interesting, but it gives you a certain benefit where I feel I feel like the work is not biased anymore.

s139
00:10:08.800 --> 00:10:16.400
So, when I when I code with these directly, um I usually feel like there's a bias, where it wants to say things are really working or whatever, right?

s140
00:10:16.400 --> 00:10:21.840
Versus like, you know, um I'll show this.

s141
00:10:21.840 --> 00:10:24.120
Let's look.

s142
00:10:24.520 --> 00:10:29.760
If this had been working on PR 294 by itself, I think it would have been like, "This PR is amazing.

s143
00:10:29.760 --> 00:10:31.000
Like, we got to merge it, right?"

s144
00:10:31.000 --> 00:10:37.520
But then this one was like, "No, like, there's another PR that should supersede it, and probably we should just close this PR."

s145
00:10:37.520 --> 00:10:37.720
Right?

s146
00:10:37.720 --> 00:10:44.480
So, that's kind of the benefit you get, where the manager has a different context um than the workers.

s147
00:10:44.527 --> 00:10:45.800
[snorts]

s148
00:10:45.800 --> 00:10:48.400
Uh let's see.

s149
00:10:48.600 --> 00:10:50.800
All right, so this is compacting whatever.

s150
00:10:50.800 --> 00:10:53.080
So, it will take some time to complete.

s151
00:10:53.080 --> 00:10:59.800
Don't want to wait for it, but when it completes, you see these notifications, and I think my efficiency has improved a lot with tmux.

s152
00:10:59.800 --> 00:11:05.000
Um I think is Austin or someone from tmux here?

s153
00:11:05.000 --> 00:11:05.560
Yeah, here.

s154
00:11:05.560 --> 00:11:06.200
What's up?

s155
00:11:06.200 --> 00:11:07.360
Go.

s156
00:11:07.360 --> 00:11:09.417
Yeah, so Austin is uh

s157
00:11:09.417 --> 00:11:09.600
[clears throat]

s158
00:11:09.600 --> 00:11:11.960
one of the guys behind tmux.

s159
00:11:11.960 --> 00:11:17.640
Um you know, big fan of it, and I'm not I'm not paid to say this or anything, but big fan of tmux.

s160
00:11:17.640 --> 00:11:19.400
Uh been using it a lot.

s161
00:11:19.400 --> 00:11:24.040
I don't know, you can spend a minute if you want to talk about anything cool about tmux.

s162
00:11:24.040 --> 00:11:26.520
Yeah, so yeah, one of the creators of tmux.

s163
00:11:26.520 --> 00:11:29.560
Really happy to have Jeff as one of our power users.

s164
00:11:29.560 --> 00:11:33.080
And also like kind of endorsing it for free.

s165
00:11:33.080 --> 00:11:35.920
But yeah, lots of uh yeah, lots of things that we're shipping.

s166
00:11:35.920 --> 00:11:38.280
We shipped a Cloud Code Teams integration.

s167
00:11:38.280 --> 00:11:46.200
So, if anyone is using Cloud Code Teams and wants to actually see what the salvations are doing, we'll automatically spawn terminals for that.

s168
00:11:46.200 --> 00:11:49.200
We also shipped Cmux SSH.

s169
00:11:49.200 --> 00:11:57.280
So, if you do any SSH work, you can use our native Cmux SSH to you know, do your tailscale and etc. To other computers.

s170
00:11:57.280 --> 00:12:03.120
You can even use it to run your own open claw in your Mac minis.

s171
00:12:03.120 --> 00:12:05.800
But yeah, just one of the creators.

s172
00:12:05.800 --> 00:12:12.680
Please feel free to email me if you have anything that you think is a bug or any feedback, feature requests.

s173
00:12:12.680 --> 00:12:22.840
I'm always available and yeah, just love seeing people using Cmux and always want to take as much feedback as I can to like get them as get them

s174
00:12:32.440 --> 00:12:33.960
Okay, so just download that.

s175
00:12:33.960 --> 00:12:35.480
Yeah.

s176
00:12:35.480 --> 00:12:36.560
They actually have an app too.

s177
00:12:36.560 --> 00:12:37.520
So, the app might be nice.

s178
00:12:37.520 --> 00:12:38.400
Yeah.

s179
00:12:38.400 --> 00:12:40.400
Okay.

s180
00:12:40.640 --> 00:12:44.360
And then use that to help me get back in and set set up.

s181
00:12:44.360 --> 00:12:44.680
Exactly.

s182
00:12:44.680 --> 00:12:45.040
Yeah.

s183
00:12:45.040 --> 00:12:45.560
Okay.

s184
00:12:45.560 --> 00:12:46.360
Okay, cool.

s185
00:12:46.360 --> 00:12:47.360
Thanks.

s186
00:12:47.360 --> 00:12:48.040
All right.

s187
00:12:48.040 --> 00:12:50.839
Is using the sandbox, it should not.

s188
00:12:50.839 --> 00:12:57.240
Where would you Have you had Have you run into any troubles Aaron about sandbox token usage?

s189
00:12:57.240 --> 00:13:01.760
Uh no. I'm just I'm not technical at all.

s190
00:13:01.760 --> 00:13:03.360
I have open claw installed, so I'm

s191
00:13:03.360 --> 00:13:04.160
Okay.

s192
00:13:04.160 --> 00:13:06.320
very newbie questions.

s193
00:13:06.320 --> 00:13:07.960
No no no bad questions.

s194
00:13:07.960 --> 00:13:08.800
All good.

s195
00:13:08.800 --> 00:13:09.640
Yeah.

s196
00:13:09.640 --> 00:13:12.320
Uh they're not the sandbox wouldn't do that.

s197
00:13:12.320 --> 00:13:16.880
I think if if everything's working well for you, you know, keep using the sandbox.

s198
00:13:16.880 --> 00:13:21.480
For me, I just you know, don't use a sandbox and so so far nothing bad has happened.

s199
00:13:21.480 --> 00:13:26.440
But I think if I had an external bot, then I would use a sandbox too.

s200
00:13:27.080 --> 00:13:32.480
You changing the model testing a new model like it'll go down then I have to figure out how to get it back up again.

s201
00:13:32.480 --> 00:13:35.000
So I think I want to set up this sandbox because

s202
00:13:35.000 --> 00:13:35.360
Oh, sorry.

s203
00:13:35.360 --> 00:13:35.640
Sorry.

s204
00:13:35.640 --> 00:13:36.360
Um okay, wait.

s205
00:13:36.360 --> 00:13:37.360
There's two kinds of things.

s206
00:13:37.360 --> 00:13:38.840
Okay.

s207
00:13:38.840 --> 00:13:47.240
The sandbox that most people talk about means like they run it in Docker or they like have it running in isolated part of their system.

s208
00:13:47.240 --> 00:13:50.920
But the other one was the one where I made the staging instructions where I was like oh

s209
00:13:50.920 --> 00:13:51.600
Yeah.

s210
00:13:51.600 --> 00:13:52.400
Okay.

s211
00:13:52.400 --> 00:13:58.040
So that could double your token usage if you send the same work to both of them.

s212
00:13:58.040 --> 00:13:59.480
If you have two of them, right?

s213
00:13:59.480 --> 00:14:02.000
Um but I think the way I would use it is not like that.

s214
00:14:02.000 --> 00:14:09.600
Like I think I would do local development and then I would run integration tests on the um sandbox or staging one, right?

s215
00:14:09.600 --> 00:14:10.720
So I have two of them.

s216
00:14:10.720 --> 00:14:15.600
And then once like everything's good, then I would merge the code and deploy to the production one.

s217
00:14:15.600 --> 00:14:16.560
I see.

s218
00:14:16.560 --> 00:14:19.280
It wouldn't double your usage, but it would definitely increase it.

s219
00:14:19.280 --> 00:14:22.080
But then the benefit is you might have some more reliability.

s220
00:14:22.080 --> 00:14:22.720
So

s221
00:14:22.720 --> 00:14:23.160
Yeah.

s222
00:14:23.160 --> 00:14:30.080
I I made some instructions and I was kind of in the middle of setting it up, but then, you know, it's like competing with the other one, so I got to debug it.

s223
00:14:30.080 --> 00:14:30.680
Yeah.

s224
00:14:30.680 --> 00:14:32.960
But I think, you know, it's it's worth trying out though.

s225
00:14:32.960 --> 00:14:37.440
So I had that for more advanced people to give it a give it a try.

s226
00:14:37.440 --> 00:14:38.000
Cool.

s227
00:14:38.000 --> 00:14:38.320
Thanks.

s228
00:14:38.320 --> 00:14:38.520
What

s229
00:14:38.520 --> 00:14:41.760
Or if you're not as advanced, if you want it, you know, you can try it.

s230
00:14:41.760 --> 00:14:43.360
Yeah, I'm definitely going to play around.

s231
00:14:43.360 --> 00:14:46.080
But what a what model do you use?

s232
00:14:46.080 --> 00:14:53.080
Like do you have a something that like a default orchestrator model and then you say for this you use

s233
00:14:53.080 --> 00:14:55.280
use Codex 53.

s234
00:14:55.280 --> 00:15:00.920
Um I found I found uh GPT 54 to just use more tokens.

s235
00:15:00.920 --> 00:15:06.200
Um and even 53 like like I I just get destroyed all the time, right?

s236
00:15:06.200 --> 00:15:08.040
So I might have to adjust that.

s237
00:15:08.040 --> 00:15:17.360
Um so basically I use this until like this is getting low and then I just switch to MiniMax, which is not as good, but it kind of gets the job done.

s238
00:15:17.360 --> 00:15:25.240
And then this is more about money than like preference, but some some work I can still kind of just do um um here.

s239
00:15:25.240 --> 00:15:28.000
So, I don't always have to