- jaypeters.dev
- Posts
- Cursor, 1 Million Subs, and Corporation as a Service
The Weekly Variable
Full AI this week, podcasts going heavy on AI topics, generating code, and hacking away at Twitch APIs with AI.
Topics for this week:
Cursor
I’m sidetracking from my personal projects for the moment to focus on the Twitch Hackathon as the due date nears.
The hackathon group had a great session last weekend and we narrowed down the scope of an MVP so we’re gearing up to submit something by November 4th.
With that, I’m writing a bunch of code to get the ball rolling but I’m not writing it alone.
I’ve got a Copilot… ok not Copilot, but a Cursor.
Cursor has popped up in my feeds a few times now so I’ve been meaning to try it out and this seemed like the perfect opportunity.
Admittedly it’s been a while since I’ve written proper backend code and having Claude Sonnet 3.5 and gpt-4o-mini built directly into the IDE and codebase, ready to help me refresh on best practice has been a game changer.
I’ve written and generated a ton of code in a very short amount of time - only a few days.
On my own, I would never have been able to reach this kind of output.
AI isn’t writing the code on it’s own yet, it’s certainly made some pretty obvious mistakes that I’ve caught, and some that I didn’t catch so I had to go refactor, but by and large it helps me fly through whatever I’m trying to write.
It’s a little scary and exciting at the same time.
I was not expecting Cursor to make such a huge difference but I’m glad I tried it.
Cannot recommend it enough to my fellow developers.
The Dev Sync
If you want a little preview of my Cursor experience, it will be available next week!
Eric and I recorded a new episode of The Dev Sync where I do a quick demo of what I’ve been working on in Cursor and we talk more about coding with AI.
Naturally we jump to much wilder ideas like hiring an entire “company” that’s comprised of AI to do the work instead, but more on Corporation as a Software later.
New episode of The Dev Sync goes live Monday with episode 10. We made it double digit episode numbers!
The Road to 1 Million
Devin Nash outlined his plans to grow his YouTube channel to 1 million subscribers this week and the timing is great.
I’m not looking to gain 1 million subscribers on my channel, but I’ll be closely following his progress and how his approach works.
He plans to primarily stream 2 to 3 days per week with the intent of generating a few shorts and 1 long form video at some point in the stream, which makes a lot of sense.
He then has a team that will identify the best clips from the streams and they will help edit and upload those videos to all of his social media accounts.
This process is what inspired my idea for the AI Pipeline I’ve been working on.
Well more accurately, the AI Pipeline would be the foundation that has access to all kinds of AI models and tools like Whisper and ffmpeg, then you could build an AI media company on top of that.
And yes I could hire an outsourced team to watch all of my streams and teach them to find the best videos in those streams instead, or I could create a system once that does it for me automatically, and use that system forever and even offer that system to others.
I could “hire” AI to fill out that team and those roles.
It’s more work upfront undoubtedly, but I think it would be worth it in the long run.
I think Devin and other influencers absolutely have the right approach though, with hiring a media team.
They don’t have the time but they do have the budget.
I have a little extra time and an idea for how to build it so I’m willing to give it a shot.
At the same time, I plan on joining Devin soon enough in growing a channel so I’ll be keeping an eye on his pain points before I get to a similar spot in the journey.
Looking forward to see his channel blow up on YouTube!
Intelligence vs Work
This was a heavy one but I really enjoyed it.
Emad Mostaque founded Stability AI and has been a thought leader in the AI space for some time now.
He and Tom had an excellent sparring match debating what the future looks like as AI continues to play a larger and larger role in society.
Emad has interesting perspectives on where to take AI and how to build models for the future, including building specific for different nations, but the one point that stood out to me was about the idea of intelligence.
Tom was asking about an AI becoming as smart or smarter than Einstein and Emad pointed out how the need for that kind of intelligence is not very frequent.
What typically has more impact is a decent level of intelligence but applied consistently.
AI models don’t have to be super smart, but if they can continually work with minimal mistakes, they can have outsized results.
It’s a really powerful perspective that I hadn’t considered but I have to agree with.
I, like Tom, value intelligence very highly, but super high intelligence isn’t often necessary to get the job done.
It’s great for putting together new and wild concepts but for day to day activity, it doesn’t take a rocket scientist to improve things, it just a little bit of logic to decide the highest priority and then the work to get it done.
That’s one point among many heavy concepts talked about in this 3 hour conversation but (if you ignore the clickbait title and thumbnail) it’s worth skipping around this podcast to hear Emad and Tom’s considerations for the future of AI.
Corporation as a Service
The Impact Theory podcast with Emad made me think about entire workforces made of AI, which doesn’t feel very far off.
Honestly it’s here now, happening all over the world, as you read this.
Thousands of social media accounts are constantly posting all kinds of content: text, images, audio and video all generated with AI.
AI’s are already hard at work.
Which reminds me of the AutoGen idea of splitting an AI into a team or workforce.
And after hearing Emad talk AI on such a larger scale, it made me think of the AI team concept at the next level.
What would an entire AI corporation look like?
There could be a natural hierarchy of high powered, so called god model, AIs like OpenAI’s o1 model at the top, and underneath them would be the top contenders like Anthropic’s Claude, Google’s Gemini, the larger parameter versions of Meta’s Llama, and various other open source options.
All of those larger models could be sub-divided into respective roles like an executive team or c-suite.
Then below them, they could cascade down, employ the smaller models like gpt-4o-mini and other variants that are trained on way less parameters but also are much cheaper to operate.
The larger models could hand instructions down to the smaller models, review the quality of the results, and update the prompts and assignments accordingly.
All of this could be taking place in the cloud, available through a single API or service.
A Corporation as a Service.
It’s a weird and fascinating idea.
It may be the direction that the AI Pipeline is ultimately heading but we’ll see.
I’ll be sure to let you know if I stumble on a digital enterprise.
And that’s it for this week! Diving deep on AI while keeping tabs on Twitch and YouTube. What else do you need? Maybe a CaaS, but other that…
Those are the links that stuck with me throughout the week and a glimpse into what I personally worked on.
If you want to start a newsletter like this on beehiiv and support me in the process, here’s my referral link: https://www.beehiiv.com/?via=jay-peters. Otherwise, let me know what you think at @jaypetersdotdev or email [email protected], I’d love to hear your feedback. Thanks for reading!