Figuring out my relationship with AI

Some of you know me as a pretty good growth person.

Like a lot of us, I've been figuring out my relationship with AI.

In my personal life, I'm basically always using it for all kinds of things. And I use it a lot transactionally for the job.

My premise with AI is that whatever my AI puts out, I'm responsible for their work. Just like how you'd be responsible for your team member.

Right now, I've been building stuff like:

• a content factory

• a brand design sprint

• a performance marketing sprint

• a friction finder

• aeo tracker

• and so on..

It's really quite advanced, compared to the stuff I used to do about a year ago.

For example, with the content factory, I can essentially bring in my Google Search Console, Google Ads, and SEMRush data in one place. Then funnel keywords through an engine. There are multiple sub-agents that specialize in each role. For example, there's a creative director agents who has zero SEO constraints and will find ways to make the pages as engaging as possible. I can also switch in and out of models to make sure that it performs really well!

With Google Ads, I can monitor anomalies before they can happen. I can stitch full funnel product data alongside marketing data. And then find insights. I haven't yet given it access to edit campaigns directly because I'm worried about API restrictions.

AI is helping me do things that were quite advanced. So, obviously I'm loving these superpowers.

---

Although, the one real downside is that you do things you wouldn't otherwise do. Vercel CEO has a really good take on this:

I find that I lose a lot of time just waiting on Claude to respond. Even with the fast mode or dangerously-skip-permissions. It's just a lot of time spent just clicking on an enter key.

---

Claude is what I work most with, but I recently purchased a state-of-the-art GPU. They told me anything higher than this isn't for consumer use cases.

So, I am experimenting with running different models on that machine. If I can find ways to get the cost down without sacrificing quality and run things locally, I would do that.

---

Oh and vertical AI tools, which effectively specialize in just one category have also been on my radar.

I've been finding a ton of value in specialized tools like MagicPath.ai and Ploy.ai. MagicPath is a design tool, something that Figma should've been. It's much better than Claude Design. Not great for logo designs though! But the chat function is much better than Claude, since it feels like you're speaking to senior designer. Claude feels like speaking to a junior designer.

Ploy is another tool built for website building. Replaces Webflow, Framer and a bunch of other website building tools. Honestly, I had stopped using Webflow/Framer/Ghost and a bunch of these intermediaries because they didn't have a good relationship with Claude Code.

So now my process is Claude > Github > Vercel. I just need a few components and some freedom to be creative.

With Ploy, I tried it out, and got to a good output with the build process.

But the pricing was a bit wonky so I had to cancel my subscription. Connecting to Github was $300/month. And I didn't know what I'd do once the inital lift was over. I'm also not sure if they easily let you export the code. I tried and failed.

---

We are living in very interesting times. AI can make us do things way better, but it can also magnify the slop we produce, if you don't know what good looks like.

Thanks for reading. This entire article was human-written :)

Khushi