4/22/2025 - Descript's Agentic Video Editor: Revolutionizing Content Creation

Transcript

1. 0:00[Music]
2. 0:15I don't know. [Music]
3. 0:36[Music] Yeah. [Music]
4. 0:58Load the car and write the [Music] note. Grab your bikes and grab your coat. Tell the ones that need to [Music]
5. 1:14know you we are headed north. [Music] One foot in and one foot back. Well, it don't pay to live like
6. 1:33that. So, I cut the ties and jump the tracks for never to return. [Music]
7. 1:47Brooklyn, Brooklyn, take me. Are you wear the shape I'm in? My hands they shake my head. It
8. 2:02spins. I'll Brooklyn take me. When at first I learned to speak, I used all my words to
9. 2:17fight with him and her and hurt you and me. That was just a waste of [Music] time. It was such a waste of time.
10. 2:35[Music] That woman, she's got eyes that shine like a pair of stolen polished eyes. She has to dance. I
11. 2:53satisfied. I'll see I'll see you in the morning. [Music] Oh, Brooklyn, Brooklyn, take me
12. 3:08in. Are you where the shape I'm [Music] in? My hands, they shake my head. It spins. Oh, Brooklyn, Brooklyn, take me
13. 3:24in. Three words that became hard to say. I and love and
14. 3:38you were you were then I am today. Look at the things I
15. 3:53do. Brooklyn, Brooklyn, take me in. Are you aware the shape I'm in? My hands they shake, my head is
16. 4:09spin. Oh, Brooklyn, Brooklyn, take me in. Three words that became hard to say. I and
17. 4:26love and [Music] you Tuesday evening people. We started
18. 4:41late tonight. I was out networking. It It was a fancy dinner. There there was drinks. We had to sit through a speech.
19. 4:49The price of dinner is that [ __ ] speech. And I'm telling you, $57 for dry chicken would have been [Laughter]
20. 5:04[Music] cheaper. Ah. What's happening, good people? Anybody here? Oops. I just did the wrong thing. Clicked the wrong button. We're going to
21. 5:18do a little short one here [Music] tonight. Well, every time I see you now,
22. 5:35get that look in mine. Every time I see your mouth, I hear that smile. In the early misty morning light,
23. 5:48I heard the engine turning and they all fought [Music] outside. You were leaving
24. 6:03me again today. You will convince me again today. You're leaving this hotel looking for someone else's golden ring.
25. 6:21[Music] should say so long [Music] Suzan now don't you [Music]
26. 6:36cry so long Susanna don't you cry for [Music] [Applause] [Music]
27. 6:58Hey, good lord. That's a good song. That is a beautiful song. It's a beautiful damn song. What a time for a share. Um
28. 7:18drinks always help with speeches. Yeah, that's true. This was actually a really good idea. It was it it was a uh it was
29. 7:24a technology plat AI platform company and the founder of the company was wise enough to realize that in order
30. 7:38to put their to to sell their platform they have to do all this services work and and the founder was like I don't
31. 7:46want to do services work and so this was a meeting to to invite people in to become the initial service delivery partners. Basically,
32. 7:54they'll funnel consulting leads out to their their AI consulting community basically, which I thought was pretty [Music]
33. 8:13smart. I also did something really obnoxious. We had to go around and introduce ourselves and I, you know, I
34. 8:19did my AI salon stuff and Story Vine and and AI learning lab. I go live five nights a week, all that [ __ ] And
35. 8:26then like I passed on the baton and I forgot to talk about the musical. So after everyone introduced themselves, I
36. 8:33was like, "Can I say one more thing?" And I told him I was working on a musical. And there were actually two
37. 8:41people in the crowd that actually have connections to New York Broadway producers. So take that. These [ __ ] musicals aren't
38. 8:50going to produce themselves. You got to be always be hustling. Pitch that [Music] [ __ ] I don't have any. So for tonight,
39. 9:06I don't have a I don't have a super clear thing I want to do. So it could absolutely be a tirade Tuesday if a troll comes in and says you
40. 9:18know those things she you know AI is useless because it was trained on it's just regurgitating the information it's
41. 9:25been giving it can't have a novel concept one of those I might go off but also if you just say uh how do you make cute puppy pictures we go
42. 9:39there we could do vibe I don't give a [ __ ] we can do whatever you want to Maybe I'll just Maybe I'll just sing my
43. 9:45sing my shitty three songs over and over [Music] again. There's not enough booze to get through a time time share speech. Yeah,
44. 9:59this wasn't time share. Luckily, this was actually what they were doing was interesting and there were really good
45. 10:04questions and it was really smart people who were up to some [ __ ] in the world. So, it was nice. Wait. [Music]
46. 10:19[Applause] [Music]
47. 10:42Start with roasting our little tool, vibes.diy. Don't Don't make me go in live. Don't make me go on live. I didn't check
48. 10:55it out yesterday. I won't do that to you. I I I I am very reticent to go into any tool that I haven't at least looked at and
49. 11:05understand what it is. Partially because I don't want to look like a tool, but also just because I just I can't shut my
50. 11:13[ __ ] mouth. As who someone call me last night the Gordon Ramsay of of AI? I'll just [ __ ] say it. Um, used
51. 11:25AI as a mock focus group date today. Got some valuable feedback. Yeah, I think the idea of doing mock focus groups is a
52. 11:32really smart use case. I think um doing an advisory board if you haven't made um custom GPTs of yourself and people
53. 11:43you know. It's a really good exercise. It's really fun. I could I could show you how to do that tonight. It's really quite straightforward.
54. 11:52Um, but then you can put those custom GPTs into a single GPT and have like an advisory board. You can also kind of do
55. 11:59a cheap version of it, right? You could say, um, start a new chat and say, I want, um, I want an advisory board that
56. 12:08consists of, you know, over opinionated Steve Jobs, Aristotle, um, I don't know, Dr. Phil, and producer Brandon. Well, I
57. 12:19won't know who producer Brandon is, but you know, we'll know the other ones. Um, and then you give it an idea and you
58. 12:25have each of those personalities comment on your idea, right? You set up the prompt so that you're getting feedback from all the different
59. 12:32people. Genie in a bottle, make a song about dried chicken on a Tuesday night. That's not a bad idea. Maybe we should
60. 12:38do that. We haven't made a song in a while. Uh, so you know what? Let's start. Oh, I have an idea. I do have an idea. All right. This
61. 12:48is a good This is a good one. Okay. All right. I got an idea. Genie in a bottle wins. Any ideas on where to look on the
62. 12:55salon to learn more about making agents? Jeffrey Hopkins, really great idea. Um, you should I don't think anyone right now is
63. 13:06is deeply digging into agents. The infrastructure isn't quite there and the salon is tends to not be a geeky kind of open-source crowd.
64. 13:18Um, but I would go check out um um Oh, what the [ __ ] it called? Uh, Pates Pates. Uh, Damn it. I can go look. Go into go to clubs and
65. 13:36hubs and it is AI mechanics with Pate. Go into AI with mechanics with Pate also. Um, do a search for um, what's Peter's last name?
66. 13:51Jesus, what's with my brain tonight? Um, Peter Kaminsky. Go look for Peter Kaminsky. Um, and see what he's been
67. 13:58posting. See what he's been up to. He's a good one. Um, Nate St. Pierre has been doing some interesting stuff. Nate and Peter have a
68. 14:08business together. Um, oh, and according Yeah, I saw that. I knew that. Um, last week we had a learn out loud session about um about
69. 14:22agents. So there's a learn out loud session where where it's in past events. Okay. So go check out past events and
70. 14:29look for the LOL on agents. So there you go. Can we make an agent? Uh you can. I can't. I haven't made one. I don't
71. 14:39listen. I people are making custom GPTs and calling them agents and they're making chat bots and calling them
72. 14:45agents. That's not an agent. An agent to me is I tell it to do something and it [ __ ] goes off and just [ __ ] goes
73. 14:53right and it does 150 things and it's not programmed in with logic. Do this thing then do that thing then do that thing. If then else.
74. 15:03No, an agent is good. Just go. Just [ __ ] go. Take the beach. Take the beach. The LOL was on NAN. Okay. Oh, so it's not really agents.
75. 15:19It's an [Music] automation. And by the way, a you you like that? [Music]
76. 15:41Well, I heard there was a secret called David play that please a lot. You don't really care for music, do [Music]
77. 15:55you? It goes like this. The fourth, the fifth mount of fall and major lift. The Bible king composes. [Music] Hallelujah.
78. 16:12Hallelujah. Hallelujah. Hallelujah. Hallelujah. [Music]
79. 16:31Well, your faith was strong, but you needed proof. You saw her bathing on a roof. Her beauty and her moonlight overthrew
80. 16:44you. She tied you to the kitchen chair. She sp your throne and cut your hair from your lips. [Music] Hallelujah.
81. 17:01Hallelujah. Hallelujah. If you're wondering what the [ __ ] happened, like what? I know. I was mids sentence answering someone's question
82. 17:12and the dog [ __ ] did his thing and then I just sung a song. Welcome to Chat Add. If you haven't been here before, my name's Kyle Shannon.
83. 17:20Uh, I know what the term executive function means. Dictionary definition of it. I have just never experienced it.
84. 17:28So, welcome. Welcome. I will occasionally finish a sentence in this here uh in this here thing. Sing it.
85. 17:35Champ. Yeah, Champy. Sometimes Champ the muse just takes over him. What was I talking about? Oh, agents. Here's the thing. I have never
86. 17:45um built an agent largely because um right now it's a lot of [ __ ]
87. 18:01work. I I am a fan of the let the really smart people build the [ __ ] that's going to do the [ __ ] for me
88. 18:09automatically. Which, by the way, if did you did Oh, speaking of agents, did you did anyone I don't apparently I either
89. 18:20didn't drink enough tonight or I drank too much. Um, the tongue is semifunctional. Dcript today put out a
90. 18:29video about a new system that they've created where they basically put an agent layer on top of the Dscript tool.
91. 18:36So, the Dscript tool, if you don't know it, is this really, really powerful. It's kind of video editing,
92. 18:43but it's more like AI video analysis. It'll transcribe your video. It'll take take out the ums and the o's. You can
93. 18:50add, you know, subtitles and just, you know, whatever, uh, whatever they're called, uh, captions. You can add
94. 18:57captions. You can do editing in it. Well, they're launching a thing where it's now that tool. So, you can still
95. 19:06use that tool, but there's now a chat window and so you just chat with it and say, "Oh, go take out the ums and o's in
96. 19:13my video." And it'll figure out which tool to use and go do that [ __ ] for you. So, I'll play that video tonight. That's
97. 19:17pretty cool. ADHD is just spicy curiosity with side quests. That's a great way to put it. [Music]
98. 19:33What's dcript, Kyle? Dcript. So, if you go to dscript, I think it's aai, but it might be.xyz orso or dot It's It's really [ __ ] annoying. D
99. 19:52script. Dcript. Why is that not autopop populating? Oh, cuz I probably haven't been to Dcript under this user profile.AI. I think
100. 20:05it'sai. No. God damn it. Gosh, you son of a [ __ ] Okay. Word to the wise. For those of you vibe coding and uh and hiring
101. 20:22teams of offshore engineers and making software, making websites and SAS platforms, spend the [ __ ] money for
102. 20:32the.com. Okay, rant over. So many tools now hard to keep up. It's impossible to keep up. Source camp. It is It is
103. 20:39ridiculous. Okay. So, we're going to type in dcript AI and then we're going to go to chat GPT and it's going to tell us.
104. 20:48There we go. No, that's deepgram. Wait, what's deep Graham? The [ __ ] is deepgram? Oh, it's a stupid [ __ ] content
105. 21:03site. Are are seriously Dcript video
106. 21:18editing it is dscript.com. What? What? Oh, I typed it wrong. I typed.ai and then No. Didn't I type.com after
107. 21:33that? I don't I don't [ __ ] know. All right, it's.com. Fine. See, I told you it was easier to find with com. You just got to spell it
108. 21:47right. Okay. So, what dcript is, let me see if I can go find projects. So, like I've got here's Paradox
109. 21:58Alley. Let me change my let me change my sharing so you can hear hear what the hell's going on on screen. Yeah, I'll
110. 22:06tell you what, boy. I'll tell you what. You're gonna you're going to whack this. So, I got I got me a song called Quantum
111. 22:18Cipher. And so, I made this picture of a cat. And I think this was a Oh, this is Paradox Alley. This was a this was a
112. 22:30followup to Quantum Cipher. And so I wanted to make a video that had captions in it, right? And so so notice the little the little
113. 22:47button here, the little things moving. And then here's the like it did all the transcription of the words and then it's
114. 22:54going to use those to make subtitles. I'm the quantum gum shoe running through sunless tunnels from corners of order
115. 23:02late in the cop and hanging quarter courses of old concepts clutter corners stumbling over subtle signals scribbled
116. 23:07in riddles by direct devils levels of gravity unravel rapidly my mind spinning like spiral bevels and so you can go in
117. 23:13and you can edit you can edit the script and you can cut [ __ ] out if it's not good you can edit you can put things in
118. 23:19you can do overlays you can do so it's basically think if if um if Final Cut Pro was in its first year and it had some really cool AI [ __ ]
119. 23:34in it. Um, but you would be more comfortable if you were a if you're a command line interface engineer type, you would be
120. 23:46more comfortable in this tool than if you were a video editor because it's just I it's it's very very powerful and it has a shitty
121. 23:56interface, a shitty complicated interface. Um, can you figure it out? Yes, you can figure it out. Is it a pain
122. 24:02in the ass to figure out? Yes, it's a pain in the ass to figure out. So, let's go look at I got to change my tabs back
123. 24:09now. Let's go look at the video. So, what they're adding is super
124. 24:24[ __ ] cool. And okay, I want you to look at this. Oops. Put that in the wrong place. I will not say it. A dirty joke popped
125. 24:39in my head and I am going to be a professional. B script uh script. Wait, script
126. 24:55agent. Okay. So, I'm going to play you a little movie and and you're gonna be like, "Oh, Kyle, that was a good movie. Can we Is
127. 25:03there another one? Can we watch another one? Should we get the popcorn? Should we get It's not that kind of movie.
128. 25:07Okay. Okay. Uh, what's that? Not sharing. Oh, well, you wanted me to share. Y'all are so [ __ ] picky. Okay, there's Andrew Mason, CEO
129. 25:22of Dscript. So, we're going to play this movie. Now, here's what I want you to I want you to watch for and
130. 25:29and Okay, this is I'm going to give you a little power up right now. So, a lot of people ask me like,
131. 25:40"How do I keep up with stuff? How do I, you know, do all this? Why do I do all this?" at this event tonight. I told people how
132. 25:51many how much I went live and they were like, "What are you doing?" Okay. What is going on here? What is that? Okay.
133. 26:06Um, so what this is is this is a simplification layer, an agent layer sitting on top of a complicated piece of software.
134. 26:17When you watch this, watch it with two minds. One is, oh, I need to make videos. The other one is conceptually
135. 26:25what's happening here is really, really important. We're going to start seeing this. So, there's this thing that
136. 26:34anthropic launched called M MCP, model context protocol. And what it allows is it allows for software developers to
137. 26:43create what's called an MCP server, which is basically a translation layer for their application that a large language model
138. 26:53can talk to. So you could say, "Hey, make me a 3D village in Blender." And it will know that it's got an MCP hook into
139. 27:02Blender. it'll know what Blender is and be and that MCP server will say here's all the things you can do inside
140. 27:10Blender, you know, from a from a large language model and it'll it'll just go use that. This is that for Dcript. We're
141. 27:18going to start to see this over and over and over again that the things that humans used to have to learn, we're not
142. 27:25going to have to learn them anymore for much longer. So, so watch this with that in mind. This isn't just about what
143. 27:31Dcript is doing. This is the we're at the very beginning of the insanely complicated world that we have to navigate. The world of the
144. 27:44internet and software and processes and [ __ ] We're going to start automating it away. And here's my prediction.
145. 27:57Yes, it's going to cost some jobs and a lot of the stuff that it automates away, we're going to realize we [ __ ] hated doing that. Good
146. 28:09riddens. Well, now what am I supposed to do? I don't know. But you don't have to do that shitty stuff. Maybe there's some
147. 28:15non-shitty stuff you've always wanted to do that you can go do now. So, this is the beginning of that. So, so watch it
148. 28:22with that in mind. All right, let's go. Let's [ __ ] go, people. All right. Uhhuh. These nuts script. Yeah. Yeah.
149. 28:33Yeah. Uhhuh. Years now. You know, before it was cool and there's been a lot of change in that time, but honestly,
150. 28:40everything feels kind of like a dress rehearsal for what is going to happen this year with agents. Hang on. Uh,
151. 28:47Source Camp just said, "I'm starting to see how many new jobs there are." Here's the thing. So Source Camp for one, she's
152. 28:55she's Kelly Camp. She's got an AI agency out of Dallas. So if you need someone who really knows her [ __ ] with this and
153. 29:03and you want to figure this stuff out, she's really good at it. She's on the front lines with this stuff. And
154. 29:11so she's working with companies that as they're discovering what AI can do and it, you know, eats up the tasks that used to be someone's
155. 29:24job. What she's saying is she's starting to see what the new jobs are starting to look like. She's she's starting to
156. 29:30understand what they're going to be, how they're going to be, right? She's starting to see the future of work. This
157. 29:37is very much like so so Kelly would have been in the industrial age, you know, we're all working on a farm, right? And
158. 29:45the steam engine comes and it's and most people just look at it and go, "Ah, that'll never take over my
159. 29:55horses. You can't plow a field with a big piece of steel." Right? Kelly Camp would have been the one going, "Uh, huh." So, wait,
160. 30:06that steam engine, that's going to plow all these fields. You know what I'm going to do? I'm going to move to New
161. 30:10York City because I bet they're going to put these steam engines where there's a lot of people around and it could do
162. 30:15stuff other than farm. I'm going to go see what they do there. And then she'd be in Manhattan, you know, where they
163. 30:21have poop on the streets and it it was probably miserable. She would have been miserable, you know, with her little
164. 30:27dainty kirchief with perfume in it to just, you know, dull the stench. But she'd have been there like working with people that are
165. 30:37inventing the service industry, inventing, you know, whatever the infrastructure you need to run a city is, which is very different than
166. 30:47running a farm. That all had to be invented. And people on the farm simply could not imagine what life would be
167. 30:55like in a city. Simply couldn't imagine it. And all it looked like was that this machine was going to destroy everything.
168. 31:03No, no, no. It was just going to take over the tasks that these people used to do. There are new tasks that they can
169. 31:11do, right? So, really important, but that's awesome. Poop poop emojis. Now, if you don't want to hear me bloate
170. 31:19about AI and you just want to see the demo of our agentic video editor, I totally get it. Just skip ahead to this.
171. 31:25They It's okay. The irregulars love people bloating about AI because it's what I do two hours a night time marker. For the rest of you,
172. 31:36assuming you're down for a little bit of bloation. So, here we go. What we've been up to at Dcript, if you really boil
173. 31:41it down, we're just making a tool that makes it easier and faster for humans to make video. In fact, that's all any of
174. 31:47us are really doing. Those of us that make software tools, we make things easier and faster. So, you can save
175. 31:52time, save money, do more. But now, for the first time, we kind of have an alternate path. Maybe instead of killing
176. 31:58ourselves to make tools that are faster and easier for humans, while somehow simultaneously making them ever more
177. 32:04powerful. Instead, we software developers can just build these AI entities that use the tools for you.
178. 32:11They become the tool experts and now all you have to do is tell them what to do. So, that's what agents are. And I'll
179. 32:17just tell it's a really good explanation, right? He said, "Instead of us focusing on all these features and and engineering them
180. 32:26in a way that humans can use them, why don't we put our time into making these things that'll just [ __ ] do the work
181. 32:32for you?" Perfect explanation of a you upfront. I am fully agentpilled. I'm using vibe coding tools like cursor and
182. 32:39replet every day now and it's super clear to me that they are going to fundamentally change the nature of not
183. 32:45only how we build software but how we work together how we organize into teams what skills are valued will AI one day
184. 32:51replace us completely I don't know people keep saying that that's like chimney sweeps whining about electric
185. 32:56heating so we'll see I guess because there's no getting off this train but what I think is very clear is that
186. 33:01agents are going to change the nature of our jobs I think of it like they've come along saying to us. Hey, why don't you
187. 33:08humans go over here into the director's chair and your job? Wait, can I just We got to give props. I forget this guy's name. What's his
188. 33:17name? Um, Andrew Mason, CEO of Dscript. We got to give him props. So, what do I talk about all the time? Embrace the
189. 33:27jank, right? He made a picture that's supposed to say director and it says Irector and he didn't [ __ ] fix it. He
190. 33:35put it in his little launch video. [ __ ] genius. Love it. Embrace the [ __ ] jank. Faster and with total
191. 33:44mastery of these tools. And by the way, this is a right now thing. It is here now. We don't need to wait for another
192. 33:51breakthrough in artificial intelligence. We're really just waiting for product engineering teams to catch up with the
193. 33:56intelligence that's already landed here on Earth. Okay, enough of this nonsense, but it's actually a good segue into the
194. 34:03purpose of this video today. Dcript is taking our first step into this future with an allnew agentic version of our AI
195. 34:10assistant underlord. So before our AI assistant looked like this, a bunch of buttons and discrete actions and now it
196. 34:17looks like this. Surprise, a chat sidebar. From here you can talk to your underlord and it can do all of those
197. 34:23actions in the old sidebar as well as everything else in dcript. And it can read your script, listen to the audio
198. 34:29and see the visuals. So I mentioned cursor earlier. If you've heard of that, you can listen, read it, and see the visuals.
199. 34:39It's like cursor for video. Just imagine that you have an assistant editor who is an expert in dscript hanging out in a
200. 34:45chat sidebar at all times. You can ask them to do things for you and they can do it at 10 times, 100 times the speed
201. 34:52that a human could. Okay, enough talk. I'm going to show you different examples of how you can use Underlord to make
202. 34:57video. First up, this is like a recording for a feature launch. As a first step, I'd normally go through and
203. 35:02edit out all the mistakes and repeated lines. And now I just ask Agent Underlord, reads the script, sees what
204. 35:08kind of video I'm trying to make, and knows what kind of edits to make. So, it goes through and removes the repeated
205. 35:14takes, the diversions, and the long silent gaps. Next up, I want to add visuals, what we call scenes in Dcript.
206. 35:21So, I say, "Hey, can you do a visual pass? And by the way, maybe break it up in a few places with some stock video."
207. 35:27So, it comes back with a kind of plan. It tells me that it'll try to mask the jump cuts I made in the recording
208. 35:33earlier. Also, it know Wait, notice this. Look at all this. See all these this storyboard looking [ __ ] down the
209. 35:39side? It broke the script up and added a bunch of video in there. Bonkers. Notices that I seem to be talking about screen recordings, so
210. 35:50it's going to add those. This whole visual thing is maybe my favorite feature of the new Underlord because
211. 35:55going through and adding these visuals to make things look good, it's just tedious and timeconuming and now I can
212. 36:01just delegate it. And now that I see what it did, I just it's tedious and timeconuming. Most of the work that we
213. 36:07do, most of it is just tedious and timeconsuming. And it's like even [ __ ] that I really enjoy like ideiation. Okay, got to get got to find
214. 36:20a whiteboard. Oh [ __ ] someone didn't erase the whiteboard. Can I erase the [ __ ] on the whiteboard? Yeah. Yeah. Let
215. 36:25me write on Oh [ __ ] the whiteboard markers are dry. [ __ ] Does anybody have a whiteboard? Could just could we get a yellow legal?
216. 36:34Everything is just this [ __ ] drudgery, right? Because the world is slow and complicated and our systems are
217. 36:41complicated. So this is just like the computer, you know what it is? It's kind of like this. you know, chat GPT right
218. 36:48now, you talk to it and it's just like it's like a golden retriever. Yeah, boss. What do you want me to do, boss? That's that's what's now
219. 36:57happening. Except the tools are not just the golden retriever for ideas and words. It's for all of the functions of
220. 37:04the software we're going to use. It's going to be insanely good. Again, back to this idea of find your inner Rick
221. 37:10Rubin, right? You're the taste maker. You're the one that says what you want this tool to do and then just have it go do
222. 37:21it. And then we don't have to get so [ __ ] precious about the thing we crafted. Listen, if you love crafting a
223. 37:29thing and you're all about the craft, great. Do that. Do it for the joy of that. But if you're about the output, let the machines do the
224. 37:38drudgery. Just want to make a few tweaks. This stock video choice is a little weird. So, I'm gonna ask for a
225. 37:44few different options. Great. This one's better. So, I now like the visual edit, but the style doesn't really match my
226. 37:50brand. So, I'm going to paste in a screenshot from like a branded Dcript presentation and ask if it can update to
227. 37:56match that style. It sees the font and colors. Wait, look what it did. It changed all of these images to match
228. 38:03match the branding. It like it like color graded them and redesigned them. Then it just goes through and updates
229. 38:10everything automatically. Okay, almost done with this video. I think I kind of just went through this flow by telling
230. 38:16Underlord what to do, but you can also ask what it thinks. Now, what I'm showing off here is that Underlord not
231. 38:21only reads your script, but it can also hear the audio and it can see the visuals, right? So, it notices that
232. 38:27Ramdy isn't making eye contact. It offers to fix that. And it notices that the audio quality isn't super great. The
233. 38:35host wasn't making eye contact. It chooses to fix that. [ __ ] amazing. So, it offers to fix that. Okay, my
234. 38:42video is ready. Now, I just need a promo clip. So, I ask for a one minute edit for Instagram and it creates a copy,
235. 38:48takes a stab at editing down to a coherent minute, makes it vertical and adds captions. All right, great. Wait,
236. 38:55did he's got this long form horizontal video. He's like, "Yeah, just make it a minute long and and make it
237. 39:03vertical." Right? Like Dcript could do that before you, but you had to know how to reformat, pick the tools, do the
238. 39:10thing. The agent knows how to do that. Now, now think about every piece of software you have. Imagine being able to do what he
239. 39:21just did with this video with every piece of work that you've got. That's where we're headed. Great. So, my video
240. 39:27is ready to go. I was able to basically vibe code the whole thing, delegating to Underlord, making a few tweaks myself
241. 39:34when I wanted to fine-tune details. In that first video, we started with the recording, but you don't even need that.
242. 39:39As long as you have an idea, you can make a video with Underlord. So, now I want to make some descript content for
243. 39:44social media. I've got this idea, but not really a whole worked out script. So, let me just paste the idea in here
244. 39:50and see how AI does at filling in the details. I like the visuals, but I think it would be even cooler with an AI
245. 39:57avatar. The one thing that I don't know here, I have my suspicions. It looks like this is happening instantly. I would imagine
246. 40:07that this is a sped up marketing video. So that recoloring of all those images probably took 10 minutes,
247. 40:17but it still only took 10 minutes and you didn't have to do the work. You could just go have a cup of coffee. Uh,
248. 40:23play some video you made. Like which one? Yeah. Yeah, exactly. Overlord Underlord. I know that it's a play on that.
249. 40:38Number three, listen to yourself as you're recording and riff until it sounds like you. And now I have my whole
250. 40:44video with this little AI mascot on camera. Now I'm just going to follow up and ask it to add an outro scene. And I
251. 40:51think this is ready to go. Now that I have this base version, I want to create copies for some of the other countries
252. 40:57and languages in which Dcript is popular. So I just ask it to do that and it'll go ahead and put in a translated
253. 41:03script. I'm also going to ask it to update the visuals to be less American and more local to the country and it can
254. 41:09make those changes as well. So we're obviously incredibly bullish on agentic editing and you might be wondering what
255. 41:15does that mean for the rest of Dcript? All this work making it easier for humans to edit video. Are we just
256. 41:20throwing in the towel on that? When you start using the new Underlord, what you are going to find is that the agent and
257. 41:25the human editor, they pair really nicely, like apples and cheese. You saw it in the demos. Sometimes you just want
258. 41:31to take the steering wheel. Sometimes it's faster to do something yourself than it is to describe what you want to
259. 41:36someone or something else. So, we still feel really smart for all this work that we've put into reinventing the video
260. 41:41editor to make it super easy. And we think Dcript is kind of the perfect cockpit for you and your agentic
261. 41:49co-pilot to go on your mission together. So I'm really excited for creatives to get their hands on Agentic Dcript and
262. 41:55frankly other Agentic tools. This is a little crude, but really there's no better way to describe what it feels
263. 42:01like than power. This whole idea of moving into the director's chair, I think we're all going to find it
264. 42:07incredibly creatively fulfilling, at least for a few years until AI is better than us at that, too, and we need to
265. 42:13figure out how to find purpose from hanging out and playing chess or whatever. So, you're probably wondering,
266. 42:19when can you get your hands on this? If the demo that I showed earlier hadn't been kind of faked, the answer would be
267. 42:24right now. We are going to start giving this to some of you. Okay, that was faked. who don't mind squinting a little
268. 42:30and like the idea of being able to tell us what we should do to make it better. So, head over to this website if that
269. 42:34sounds like you. So, go to dscript.com/agent and fill out the get on the wait list for this thing. I hope it's free. I don't think
270. 42:44it'll be free. I think it's it's uh 10 bucks, something like that. And sign up for the closed beta and we'll start
271. 42:49letting people in this week and hopefully speed it up very quickly. Okay. Crazy times, right? That was,
272. 42:58first of all, as far as CEO, you know, marketing launch videos go, that one was super, right? He's down to earth. He's
273. 43:07just talking. He's not [ __ ] reading off a script. He's not doing awkward table launches with inarticulate people,
274. 43:14right? He knows how to talk about his company. He spent a minute and a half telling us what an agent is, like
275. 43:21demystifying it, de scaring, unscaring us. And then, you know, they showed some really sexy demo [ __ ] that doesn't
276. 43:31really exist in that form yet, but it will, right? It'll it'll be some version of that. Um, so go sign up for that.
277. 43:38That was that was quite good. Quite smart. Quite smart. It was smart, I tell you. It was smart. It was very smart, Kyle. Hey, Kyle.
278. 43:49Kyle. All right. What's your browser bookmark? Prankshot color. Browser bookmark. Oh, prank shot caller. I think I think that was some
279. 44:04night here here on the the AI salon. Let's see. [Laughter] I wanted I wanted to use a Raspberry Pi to create a prank shock collar that
280. 44:20vibrated like a cell phone when someone sends a super comment on YouTube. Yeah, this was okay. So, so part of
281. 44:28the part of the joy of this channel and part of the bane of producer Brandon's existence is when I get into, you know,
282. 44:38a hyperfocus tunnel, I could just completely ignore him. Or like maybe I'm talking as if I'm sharing my screen and
283. 44:45I'm not actually sharing my screen. And so we have joked for a long time that I need a shock collar that that the mods
284. 44:52can just [ __ ] zap me when I need to pay attention to something. Wait, scroll down. This is a good example. Yeah, exactly. Well, let's see.
285. 45:01Okay, so reason for 23 seconds. Um, what you'll need. So, we need a Raspberry Pi Zero, micro SD card, vibration
286. 45:11motor. Frankly, like what the [ __ ] are we doing? Why would we not make it an actual shock collar? Yes. Well, so I was
287. 45:20So, if you scroll back up to the top, and this was actually for YouTube super comments, which by the way, you can send
288. 45:25super comments and super likes now on YouTube. But if you notice when I first asked it to do a a prank shot collar, I
289. 45:34got booted. It said, "I'm sorry, I cannot comply with that." So, I went back and said, "Let's modify the design.
290. 45:40What if we're doing a I'm like I'm trying to think like did when did I come up with this? I forgot this was
291. 45:46something you sent me. You did this. Yeah. Uh the Post-it notes weren't doing it that night. Um and so uh I had to
292. 45:55modify it to do a a reminder wristband that offers gentle vibration. It spit out basically the instructions
293. 46:05for a shock collar just in the form of wrist. Yeah. You just need to put a capacitor in it and a couple and a
294. 46:11couple of electrodes and you're good to go. Car battery and you know you're That's awesome. You're good to go.
295. 46:19Um but yeah, super cool. But yeah, here's here's all of the elements. Um here's the software and then it it wrote the software,
296. 46:28right? Here's the description. There's the motor driver circuit. Absolutely incredible. You know, would be fun. I I I like the fact
297. 46:39at the if you scroll back up to that diagram. Yeah. My favorite part of this entire thread is that it uses the
298. 46:48completely subjective term. Scroll up just a smidge more. Appropriate battery voltage. That's great. That's [ __ ]
299. 47:02awesome. And then we could Yeah. This is so cool. So cool. Let's see. Uh, show the reminder
300. 47:16collar around a podcaster's neck. Oh, I wonder if it'll let you do that because I remember we're talking about a
301. 47:30a gentle reminder wristband. But, uh, thinking it was it was nicer to you than I was because it offered you a cancellation
302. 47:45button. The diagram is borderline pointless. Page [ __ ] all over your diagram. Yan, Yan, Yan dialing in from Finland.
303. 48:00We are We are officially global. We have We have uh folks from uh Australia, Finland. Beautiful. Where else is everybody from?
304. 48:10Pop your flag in there or type your country with letters. [Laughter] [Music]
305. 48:24[Music] Canada. Canada, our our 51st state, Canada, checking
306. 48:40in. I love I love how Canadians are like, "Oh, yeah. Yeah. You think we're going to be your [ __ ] 51st state, you [ __ ] hosers?
307. 48:53Hey, look. Look at the little vibrations. Oh, that's awesome. All right, there we go. That's what prank
308. 49:04shot collaller was. I don't know why I have it in my bookmark bar. It's pretty funny. Um, let me pop into my library
309. 49:12since I'm here. You're all seeing this, right? Yeah. Yeah. see if there was anything. Oh, yeah. This was a kind of cool thing I did. Um,
310. 49:23let me go to this one. How do I get [Music] to I'm I'm still kind of um you know, one of the things
311. 49:39that I've been talking about a fair amount this year is um training myself to not use my brain as much. And I know that sounds stupid
312. 49:57and weird, but rather than not using my brain, what I'm trying to teach myself to do is use this brain more. Like
313. 50:12like it's really hard having been on the planet for nearly 60 years to to not go into default mode. And default mode is oh if I see
314. 50:28something and have an idea I go into the mode of oh do I want to explore that idea? Oh I got to come up with more
315. 50:37ideas to refine the idea. I got to process the idea, right? And and it's just this aching, slow, neurosisfilled, crazy straw of an
316. 50:53intellect. And so, can I get better at just going, "Oh, here's this idea." Just go straight to whatever the tool is, chat, GPT,
317. 51:05Perplexity, whatever is your tool of choice. Can you get better at instantly going there to get the feedback? Because what that will allow
318. 51:14you to do is not descend into the implementation details so quickly. You can just have the idea, let the machine vomit on it for a
319. 51:25while. And you can go, okay, cool. Or you can go, oh, that's really good. And then dive in. Right. So, I saw this picture
320. 51:38today and it was a really cool picture. It's of a city. It was It was It was one of those shitty ass exposts where they're
321. 51:47like 10 crazy cities you never knew existed. A wild ride, a wild thread. Um, but I just like this first picture because it's just all these
322. 51:59buildings right on the river. Um, so I assume it's in somewhere in China it looks like, but who knows?
323. 52:08Um, and like there there's a part of me that was like that would be cool to sketch or it would be cool to
324. 52:16like imagine that as a as a fantasy thing and then turn it into a thing. And then there was something about sketching
325. 52:25that that was like, oh well yeah, maybe I should just sketch it. I'm not good at sketching. And with ADD, I don't have
326. 52:32the [ __ ] attention span to do something this intricate. So, I just said, "Turn this into a doodle someone
327. 52:40did with a ballpoint pen on ruled paper." And it did that. Look at that Tik Tok pen. Stoic in the corner.
328. 52:52Kyle, I had the exact conversation with Chat GPT about this process and going down rabbit holes. Yeah. Yeah. Exactly. It's
329. 53:01like like just like let it go there. And and then I said so so it did that and then I said make it more detailed. Wait,
330. 53:11make it more de make it like a more detailed more detailed like a talented doodler did it on nicer paper and it did this more like really
331. 53:26nice, you know, sketch like a really talented illustrator would if they were just doodling with a ballpoint pen.
332. 53:36Don't you hate hate those kind of artists that they're like, "Oh yeah, in class I just doodled this." They like rip it out
333. 53:43of their notebook and like this is what they did for the past hour and you're like, "I'm worthless." Well, now you can doodle. Like does it
334. 53:54matter that I didn't actually do that with my fingers? No. Because I w I couldn't have I wouldn't have. I don't have enough joy
335. 54:03about drawing because I'm not good at it to get here. I don't have the attention span to do this much little
336. 54:10crosshatching, but I kind of like that aesthetic and it it it feels fresh and interesting. Like why not? Why not? And like I feel like every
337. 54:20time I do one of these little teeny tiny, what would you call it? like running down the rabbit hole, like discovering a rabbit hole and just
338. 54:27dipping my toe in it, I learn a new capability or I learn a new nuance to the capability that's already
339. 54:35there. And then and then that led me to go um I was thinking about oh I saw there's a there's a picture on X right now
340. 54:47someone did of like what it'll look like when when the Mars colony is like fully populated with buildings and [ __ ] like
341. 54:54that. And so I just I thought, "Oh, that would be something fun for chat chat GPT to imagine, right?" And so I said, "Let's
342. 55:03say we started developing a colony on Mars 10 years ago and the supply from Earth hasn't slowed. Descript what it
343. 55:10would look like. Describe what it would look like from a drone shot." And so it described it. And again, I don't read
344. 55:16any of this stuff. I just have it do it because I know it's thinking about it. And then I said, "Make that
345. 55:22image." And like there's the image. Cool. Love it. And then I said, "Make a sketch of that like, you know, an astronaut did
346. 55:37it. How cool is that?" And then I said, "Put it in an astronaut's hands with the city in the
347. 55:52distance like like out of nothing." So, like this is my new my new my new thing. It's kind of like actually, you
348. 56:05know what it's kind of like? No. Kyle, what? Maybe you could tell us sometime. Shut up. Where are we? Did I I lost my tab. Well, maybe if
349. 56:23you didn't have so many tabs open, it wouldn't be so hard to find everything. Shut up. I I think I must have closed it. Oh,
350. 56:30no. Here it is. Yeah, there it is. Okay. Um, so my new thing is just like Oh, let me show you another thing I
351. 56:42did. Um, if I go there, can you see that? No, because I got to change something.
352. 56:59Okay, so I saw something cool on on the Twitter. [Music]
353. 57:24I'll get there. Calm down, people. Okay, this was it. And then this was this was the post. Okay, so there's my original post. Okay,
354. 57:33so I'll get back to those in a second. So this dude, Michael Rabon or Rabon The junior high higher in me is like
355. 57:48Michael Raboner. God, he had to live with that his whole life. He's probably still living with that. He's like, it's a
356. 57:57Rabon. Anyway, Michael Rabone made these cool travel posters, right? And then he I think he put the prompt in here, what
357. 58:05the prompt was. So, if you go show more, you know, there's the prompt. And so what's happening a lot on
358. 58:17Twitter is someone will see something cool, they copy the prompt, and they go make the exact same thing. And I thought,
359. 58:25well, maybe we could make something different. Like, I like the aesthetic. I like the idea, but like, do I really
360. 58:31want to do one of Denver and then New York and the other places I've lived? And right, who gives a [ __ ]
361. 58:40So, what I did was I took the prompt and I went to ChatGpt and I said, "Hey, ChatGPT, this prompt made a cool poster
362. 58:47that I like. I want to keep the same basic, I don't know, vibe of it, but I want you to come up with 20 different
363. 58:55concepts that are kind of similar kinds of things." And then it gave me 20 shitty ones. I said, "Give me 20 more."
364. 59:02And then one of them was um retro tech. And I was like, "Oh, that'd be cool. Let's do retro tech." And then I said, "Write me the
365. 59:13posters." Like, "Write me the prompts for all the posters of like the top 20 pieces of nostalgic tech from the 80s and 90s." And
366. 59:24um so it wrote them all and then I was like, "Oh, you know what would be cool is like to have stats about the device."
367. 59:32So, if you're talking about the Mac, you know, it had 128K of RAM and, you know, it was the first GUI, whatever it was,
368. 59:39how many units it sold, how much it cost. Well, 03 does research really good. So, I said, "Go research all of these
369. 59:50devices and come up with cool things and then let's start making the posters." And it took me, I don't know, 20 minutes of rounds of them
370. 1:00:02sucking, but I ended up with four really bitching cool posters that are kind of t-sh
371. 1:00:18t-shirts. I think this is my favorite one, right? Asteroids from Atari. It's got the the the whole cabinet there. Released 1979. Price
372. 1:00:29$2,95. Units sold $70,000. Cost per credit 25. Man, like if you grew up when I grew up, taking a pocket full of quarters to
373. 1:00:41the mall to go to the arcade to play this [ __ ] game. How [ __ ] cool is that? like you know so like this thing that gives me immense
374. 1:00:55joy. I didn't have to figure out what [ __ ] font that is. I didn't have to go find source art. I didn't have to go
375. 1:01:04do the research. Like all of that work, all of the work of realizing this idea. A, I would have never done it, and
376. 1:01:18B, I never would have had the idea in the first place had I not run down the Chat GPT rabbit hole. And I wouldn't have done run down
377. 1:01:26the Chat GPT rabbit hole had Mark Rabone not done these travel posters. So, my new jam is to go through LinkedIn or X or
378. 1:01:40Facebook or whatever it is and if you see someone that does some good work, [ __ ] just grab it and then and then iterate off
379. 1:01:49that YouTube comment archetypal. I love playing Asteroids at the pizza joint. Of course, that's the way you do it, baby.
380. 1:02:03So good. And then the original Mac. And then there's the Nintendo system. And then there's the Walkman. Price 150 bucks. Remember the
381. 1:02:16orange the orange foam earbuds or you know headphone cos. I play table Pac-Man. There you go. It does do some fuckups. Like notice
382. 1:02:31there's a there's a Macintosh mouse next to the next to the Walkman and and the cassette is square instead of cassette
383. 1:02:39shaped. But like you know, embrace the jank. Doesn't really [ __ ] matter. But this my new thing is like
384. 1:02:46like in 15 or 20 minutes instead of just doom scrolling you could you can be like vibe creating just oh no. Select two
385. 1:02:57objects that are the same. Are you [ __ ] kidding me? Now I've got to do intellectual physical unable to verify you [ __ ] kidding
386. 1:03:09me. [ __ ] Tik Tok [ __ ] Oh lord. Anyway, how cool is that?
387. 1:03:25And you know what it's going to be in the future, right? You'll be able to make this graphic and then this game is actually
388. 1:03:33playable. You just zoom in on it and start playing the game. Which two squares have Madagascar vanilla beans? Anyway, ain't that
389. 1:03:44cool? Yeah, man. Yeah, man. All right. All right. It's 10 o'clock, people. Do you know where your children [Music]
390. 1:04:00are? I think I'm going to get out of here. 1:00 a.m. for Tams. Sorry about [Laughter]
391. 1:04:14that. Um, where are you? You must be in New Brunswick or something like Are you Nova Scotia? Cuz that's three hours. I'm
392. 1:04:22two hours from the from New York time. Montreal. Maybe I'm in the future. New Brunswick. Oh, cool. Oak Island, baby. Curse of Oak
393. 1:04:33Island. My favorite show. My favorite nerdy Nothing ever happened show. You know, it's so funny. Like some of my favorite shows in life
394. 1:04:44are ghost shows where they never find any ghosts and gold shows where they never find any gold. treasure treasure
395. 1:04:51shows. Um, the one I don't know if you I don't know if anyone caught this one. There was I think it was two
396. 1:05:01seasons. There was a group the the show was called The Curse of Snake Island, I think, or I think it was the Curse of
397. 1:05:10Snake Island. So, Snake Island, Snake Island is literally an island with the most venomous snakes on the world. And I think the the average
398. 1:05:23population of snakes on Snake Island is one venomous snake per meter. So, if you're like walking through the jungle,
399. 1:05:33there's like 10 snakes above you, 10 at your feet. Like, it's it's a ridiculous island, right? And so apparently there's
400. 1:05:41some treasure buried on this thing. And so these people are walking through this stuff and like you know
401. 1:05:48every three minutes like someone is like this close to a thing that if it bites them they're just dead. There's no
402. 1:05:54there's no antivenenom. It's just [ __ ] awful. Um and they find some clue, right? And the clue sends them to South
403. 1:06:02America to somewhere in South America. And then the season ends, right? So, season two, they're in South America and
404. 1:06:10they're they're on some [ __ ] river and they go up the river. They go to this fall and they find the fall and
405. 1:06:16they look up and there's a big giant rock like perched on a cliff that doesn't look normal and they climb up
406. 1:06:25the cliff and they get on the rock and some dude like jumps over a gap that if he fell it would be 100 feet. He gets on
407. 1:06:32the rock and he climbs around the back of the rock and there's some [ __ ] thing there, right? And it's some other
408. 1:06:40clue and they're like, "Oh, I think this points to some falls, right?" Which is like, you know, some other
409. 1:06:48place. And then they go to the some other falls and they're they're sort of traversing their way down these crazy falls with
410. 1:06:59all these rocks and undercuts and dangerous [ __ ] and they're just about to give up. It's like it's like the you know the the
411. 1:07:09last half of the last season, the last episode of the season and they're in scuba gear and and this this woman and
412. 1:07:17the one woman was from Denver. I don't know if she was the first one to find it, but anyway, she dives into this hole
413. 1:07:23and and like they cut to the hole and they show this flash of gold, right? This flash and you hear her go and like
414. 1:07:30bubbles go up from her mask and they [ __ ] found the treasure. They found the treasure. It was like this Mayan
415. 1:07:39crazy ass temple of the gods gold things and they're literally just pulling out gold artifact after gold
416. 1:07:47artifact. Never [ __ ] heard from them again. It was like you can't you can't have a gold hunting a treasure hunting
417. 1:07:56show and actually find the treasure. You [ __ ] up the show. Spoilers. Yeah. But no, it was like it was so exciting because it was
418. 1:08:09like like the [ __ ] they found was just it was, you know, and and like how they found it like they they [ __ ] it was
419. 1:08:16like the fantasy of you find the clue on the island with the poisonous snakes that makes you go somewhere else in the
420. 1:08:23world and then you find the rock on the cliff and points to the [ __ ] thing and you get to the thing and there's the
421. 1:08:30[ __ ] treasure. So, I don't know. That was nice. All right, everybody. How much did the crew take before they let the
422. 1:08:39contestants lose? Yeah, Kyle loves ciphers. I You know what? I love ciphers. Like, I'm I'm I'm in this NFT
423. 1:08:47thing called Neotokyo, and it it was a whole bunch of ciphers. Um I didn't solve any of them, but I I cidled up to
424. 1:08:54people that were really good at ciphers. Um I like the idea of ciphers. Like, I loved the uh the Dan Brown book, uh Da
425. 1:09:03Vinci Code, right? loved it. But like I couldn't be the dude. I couldn't be the Tom Hanks character because I'm just like, yeah, I'm just
426. 1:09:12bored. Like I don't have the tenacity to to figure out ciphers. But now that I have AI, like I love that ciphers are
427. 1:09:20solvable and I just want to know the answer. I want to be on the other side. So Oh, good lord. All right.
428. 1:09:29Beautiful. Yeah, Da Vinci Code's the best, isn't it? That was a good one. All right, fancy Bob. All right,
429. 1:09:37Bob. Ah, yeah, Bob. Yeah, Bob. Um, tomorrow's wedness day. I don't think I have anything planned, although I didn't
430. 1:09:45think of anything tonight. I forgot I had this networking thing. Oh, tomorrow podcast. Um, 400 PM Mountain time. Uh,
431. 1:09:54Ann Murphy and I do the AI readiness project podcast. So, if you go to We need to make a banner for this, uh, Brandon, uh,
432. 1:10:07airedness.com. Um, that'll give you all the information. It'll have links to the live streams and things like that. So, I
433. 1:10:14also realized tonight I've got too many projects going that are not on anywhere. Like, I don't have a single place with
434. 1:10:21all my links and I got to fix that. So, I'll do that in the next week or two. Aire.com. That's it. No. AI readiness
435. 1:10:31project.com. Aire readiness project. Let me just confirm that. AI readiness project.com. Boom. Boom. Boom. Boom.
436. 1:10:45Boom. Look at that. You got You got me, you got Ann Murphy. Both of us being all sassy and sort of corporate
437. 1:11:00headshotty. And then here's who we got tomorrow. And there's past episodes. Look at us. We're We're like an actual
438. 1:11:12podcast. Tell me that's not good. Invent a link tree for events. That's not a bad idea, actually. All right. I you
439. 1:11:25good. I like it. All right. Uh, so that's tomorrow 4 pm. Watch it. Be there. Be square. You'll probably be
440. 1:11:33square because you're all irregulars. Get it? All right. All right. I'm out of here. Peace out, everyone. Hope you had
441. 1:11:41fun tonight. I'll see you tomorrow. [Music]