Table of Contents 0:00 My first task was too simplistic 0:37 Aaron Levie Tweet, using cloud software 1:31 A more complex task and prompt 2:25 Transcript Fetching, NotebookLM Usage, Google Drive Usage, Tweet Thread 3:16 Demonstration of Operator Session 5:02 Operator scrapes Seeking Alpha 5:36 Operator creates an organizational folder structure 6:42 Skipping permission checks 8:01 Operator using Google Drive 8:50 Operator searches for HTML to PDF converter 9:31 Security risks of Operator using search engine to find tools 10:40 Operator finds NotebookLM and uses it to extract insights, creates a podcast 11:40 Even though there's no NotebookLM API, we can automate it! 12:15 Operator creates a Google Doc 13:05 Operator creates a share link I didn't ask it to! 13:54 Operator creates a twitter thread with Alphabet earnings insights 16:17 Effects on advertising business? Can ads hijack Operator attention? 17:00 I've seen Operator get distracted and do something random 17:48 Can I just walk away and trust the agent did everything I asked? 18:10 Reflections, notes, thoughts on Operator, predictions 18:42 No code needed! 20:58 No API needed! Just let Operator use the app! 22:07 Teaching AI to use software via video? 23:18 Dedicated accounts for AI agents? 23:50 Iterative agent management, checking in on your agent's progress 24:49 Future integration with Scheduled Tasks feature? 25:53 App User Experience for agent users? 26:19 Security risks, whitelisting websites, organization policies 27:30 Github / version control for Operator tasks / recipes? 28:47 The diminishing value of writing code by hand 29:23 Operator Mobile App? 30:28 You may not use Operator, but Operator will use you! 31:07 More business use cases, less travel agents please! 32:05 Think in terms of chaining SaaS apps 32:26 Wrapping Up - I feel like I have a better understanding of agents now
Thanks for the great video! I am wondering if it is possible to upload more extensive instructions to Operator - for example a 50 page manual? If so, I'd like to try it out for working with specific software.
A few thoughts. First, from what I recall there is a character limit on the task box. There is a file attachment you can use too though, which I haven't explored much yet. From my experience so far, I think it would be better to break the instructions down and format / number them to make the instructions as clear as possible. I think an interesting approach might be to feed your manual to an LLM / RAG system to break it down and ask an LLM it to transform the manual into a set of numbered instructions - sort of asking an LLM to clean up the manual or rewrite it for clarity first. And when they make an API available, then if you were limited by characters, you could possibly chain multiple runs of the Operator together to do something really complicated. I think it is not reliable enough yet to do something requiring a 50 page manual without a lot of guidance, but who knows where it will be on official release - things are moving fast.
@@parttimelarry Thank you for the detailed response! Breaking the manual into smaller, numbered sections sounds like a practical approach. I'll try to experiment 🙂
![Interactive Brokers Web API [Part 1] - Docker Tutorial](/img/n.gif)
Table of Contents
0:00 My first task was too simplistic
0:37 Aaron Levie Tweet, using cloud software
1:31 A more complex task and prompt
2:25 Transcript Fetching, NotebookLM Usage, Google Drive Usage, Tweet Thread
3:16 Demonstration of Operator Session
5:02 Operator scrapes Seeking Alpha
5:36 Operator creates an organizational folder structure
6:42 Skipping permission checks
8:01 Operator using Google Drive
8:50 Operator searches for HTML to PDF converter
9:31 Security risks of Operator using search engine to find tools
10:40 Operator finds NotebookLM and uses it to extract insights, creates a podcast
11:40 Even though there's no NotebookLM API, we can automate it!
12:15 Operator creates a Google Doc
13:05 Operator creates a share link I didn't ask it to!
13:54 Operator creates a twitter thread with Alphabet earnings insights
16:17 Effects on advertising business? Can ads hijack Operator attention?
17:00 I've seen Operator get distracted and do something random
17:48 Can I just walk away and trust the agent did everything I asked?
18:10 Reflections, notes, thoughts on Operator, predictions
18:42 No code needed!
20:58 No API needed! Just let Operator use the app!
22:07 Teaching AI to use software via video?
23:18 Dedicated accounts for AI agents?
23:50 Iterative agent management, checking in on your agent's progress
24:49 Future integration with Scheduled Tasks feature?
25:53 App User Experience for agent users?
26:19 Security risks, whitelisting websites, organization policies
27:30 Github / version control for Operator tasks / recipes?
28:47 The diminishing value of writing code by hand
29:23 Operator Mobile App?
30:28 You may not use Operator, but Operator will use you!
31:07 More business use cases, less travel agents please!
32:05 Think in terms of chaining SaaS apps
32:26 Wrapping Up - I feel like I have a better understanding of agents now
Super dope tutorial of Operator. Thanks for putting this together. Such a crazy week in AI news and products
Great video and use case. Every other demo on video shows it booking a holiday and ordering pizza. Appreciated the interesting sideline insights also.
Also appreciate the mention of potential points of exploit, really crazy
amazing content. keep them coming bro
Absolute brilliance....and scary!
Thanks for the value !
Thanks for the great video! I am wondering if it is possible to upload more extensive instructions to Operator - for example a 50 page manual? If so, I'd like to try it out for working with specific software.
A few thoughts. First, from what I recall there is a character limit on the task box. There is a file attachment you can use too though, which I haven't explored much yet. From my experience so far, I think it would be better to break the instructions down and format / number them to make the instructions as clear as possible. I think an interesting approach might be to feed your manual to an LLM / RAG system to break it down and ask an LLM it to transform the manual into a set of numbered instructions - sort of asking an LLM to clean up the manual or rewrite it for clarity first. And when they make an API available, then if you were limited by characters, you could possibly chain multiple runs of the Operator together to do something really complicated. I think it is not reliable enough yet to do something requiring a 50 page manual without a lot of guidance, but who knows where it will be on official release - things are moving fast.
@@parttimelarry Thank you for the detailed response! Breaking the manual into smaller, numbered sections sounds like a practical approach. I'll try to experiment 🙂
Wondering if/when Operator will have API to push prompts to it :)
It's definitely a "when"", their post says they will release an API "Soon". Whether that's weeks or months is anyone's guess.
Do you know if operator works with zapier yet?
My prompt : “ acquire Mexican citizenship , log into bitmex, 100x it, close trade in 2026”
WTF. what the hell is this. 😱😱😱😱😱
machine friendly UI will make UI design even more minimalistic, boring and formulaic