Aiden JohnstonInstagram

day 90 of posting every day until @curatedating has 3,000 daily active users

1:59English(auto-generated)7 segments531 words, 3 min read

Search inside any video

SavedThat transcribes your saved videos and lets you search across all of them instantly. Save this video and find any moment.

TL;DR

This video explains how to effectively use AI agents by granting them authority to operate autonomously, improving efficiency in startup tasks.

AI agents for startupsgranting authority to AIautomating business tasksefficient AI managementAI decision makingbuilding with AIusing AI in entrepreneurshipAI safety features

Chapters

  1. 0:00Introduction to AI Agents
    01
  2. 0:30Understanding Authority in AI
    02
  3. 1:00Building a Startup with AI
    03
  4. 1:30Setting Up AI Safeguards
    04
  5. 1:50Conclusion and Next Steps
    05

Transcript

0:00

If you ever wondered why some people's AI can build entire startups overnight while yours is still giving you wrong homework answers, this is AI Agents 101 and by the end of this video you'll be on track to use more water than any human ever should. In short, the models are good now. That's not the bottleneck anymore. The bottleneck is authority. Back in August, I'd have four cloud code terminals running once in parallel and I was still the bottleneck because they kept stopping to ask for permission.

0:17

Think of AI like a really capable employee. Say you hire someone to manage your convenience store or something, but they're not allowed to spend any money and you don't want them to call you. Imagine a pipe burst. Now they have to call you and your time is wasted calling a plumber. The manager could have done the plumber call in the first place. So back up, same manager, but this time you give the manager authority up to $1,500 per

0:39

se. Way fewer calls, maybe even zero. You can't give it any responsibility without authority. Now this is a gag Jersey store for AI researchers. Claude built the whole thing in about four hours. The prompt consisted of these magic words. We will make all of the decisions at the start. Then I'll leave you to run autonomously with as many sub agents as you need. Any questions, permissions, anything. After four hours, it was pretty much completely done.

0:55

All it needed for me was an API key for the print company, a payment link, and a domain. It's live right now at poach.fyi. You can check it out, but you can't buy anything because Stripe flagged my account. I'm gonna LARP and say it's because I didn't do the online assessment they sent me for an internship, but it's because my startup is in a super regulated niche.

1:10

Just know the less authority you give it, the more work lands back on you. If the agent is set up right, it shouldn't do something stupid like delete your entire database, but it's nice to baby proof it so it can't. This is through a feature called Hooks that blocks commands and explains to the agent why they're blocked. Then I bought its own sandbox. This Mac mini belongs to the agent. It doesn't

1:26

have any of my passwords. It's meant so that the agent could build, test, break stuff and loop over and over and over until it gets it right. That way, when I come back, it's something that's already pretty good and battle tested instead of just slop. So next time, don't babysit it. Front load all the decisions and permissions and planning and download T3 code. It's free,

1:41

it's open source, and it has all the features any agentic engineer could ever need. And if it's missing a feature you want, make a pull request. If any of this stuff about AI and using it to automate stuff in your business interests you. This is day 90 of posting every day until mine has 3000 daily active users fall along for the journey. Peace.

Are you the creator or rights holder of this video? Request removal of this transcript.

Related Transcripts

Never lose a moment again

Save videos from YouTube, Instagram, and TikTok. Search everything that was said, and jump to the second.

Start Saving Videos - Free Trial