OpenAI dropped AgentKit this week, their new framework for building AI agents directly into your workflows.

The marketing is slick. The demos are impressive. Brands are trying to figure out what they can use it for. And N8N junkies are already trying to come up with their next template.

Here's the problem nobody's talking about: You're about to lock yourself into a single model provider as the AI landscape is exploding.

I've watched this movie before. Six months ago, GPT-4o was the king of images. Today? Gemini’s Nano Banana designs circles around it. It’s not even close.

Worst yet? Anyone who had a custom GPT around copywriting had everything destroyed when GPT-5 came out and somehow got worse at copywriting. It now writes like a corporate committee on a binger.

Here’s a simple truth that OpenAI, Anthropic, and Google don’t want you to believe:

If you recall from what we were talking about [creating AI content that slaps], to get the best results with AI, you’re going to have to use multiple model providers.

OpenAI is betting you'll build so deep into their ecosystem that switching becomes painful. Custom GPTs, AgentKit integrations (Agent Builder), API dependencies, each one is another brick in the wall keeping you locked in as they also get more of your data.

Meanwhile:

But if you're all-in on Agent Builder? You’re stuck with underperforming models in the majority of use cases. This isn’t like an Apple vs. Android battle of preferences, we’re talking about real business results.

An agency I spoke with recently put it perfectly:

Use platforms that let you use model-agnostic workflows. This is part of why N8N won’t die overnight (even though workflows on N8N are very brittle). You want a tools that let you swap models like changing your socks (it’s why we have a model switcher in Raleon).

When Anthropic drops Claude 5 and it's 2x better at analyzing customer reviews? You switch. When some Chinese lab releases a model that costs 1/10th the price for the same quality? You switch.

Don't marry the model. Marry the outcome.

Does this mean you shouldn’t try Agent Builder? No. I 100% would try it. In fact, I have. But it feels like a developer tool right now. It will get better. But until they allow me to use other models, I won’t use it outside of very specific use cases.

For now, use better options (like what I’m about to cover below).

The head of social for Gains in Bulk at Commerce Roundtable asked me a question that made me physically uncomfortable:

"We need to analyze 15+ YouTube creators. Our current tool does one video at a time. It's a major time sink. Do you know how to make it do all 15 at once?"

One video at a time. In 2025. That’s like using your browser with only 1 tab open at a time (I currently have 67).

What they needed:

What they had: Poppy AI doing its best with a one-video-at-a-time process.

I built them, and now you, something better in 10 minutes with Gemini app builder (it’s criminally underrated).

Drop in 15 video links, hit analyze, go get coffee. When you're back, everything's done

I shared this with a friend. His response? "Why not build this in N8N?"

Because I like my Saturdays with the family. And my sanity.

The N8N way: 6+ hours of node configuration, API wrestling, and that special moment at 11pm where you're Googling 'why does my webhook hate me' while your spouse asks if you're coming to bed (if this line makes no sense to you, you should be thankful you never used N8N).

The Gemini app builder way: 10 minutes. And it actually works.

Here's my controversial take: Tools like N8N and Make are great... if you enjoy spending half your day connecting nodes and debugging workflows that break every time an API updates.

For most DTC operators? They're overkill. You don't need a visual workflow builder. You need working automation that doesn't require you to feel like a junior developer decoding API documentation.

I fired up Gemini 2.5 Pro and built this thing in the time it takes to make a decent pour-over coffee. No fancy node-building platform, just straight prompting, a dash of vibes, and a bit of structured thinking.

Here's the exact prompt I used to start. Nothing fancy:

Pro tip: That last bullet is magic. It lets you actually adjust how it analyzes the videos without having to rebuild the app each time.

Pro tip #2: Not sure how your prompt should look to build or analyzer, ask Gemini to write one for you (seriously, it works).

Here's what this app actually does for you:

Total build time: 10 minutes. Three of those were spent yelling at my dog to stop barking at the Uber eats delivering guy while Gemini coded. The other two were me poking at OpenAI Agent Builder debating how much more I should subject myself to the tool.

Look, I know you're busy. You don't have time to figure out code structure or debug someone else's half-baked GitHub repo. So here's the actual working code, ready to copy-paste.

Using it is stupid simple. Open Google Gemini and paste this

That's it. You now have a working YouTube analyzer that would've cost you $2,000 from a dev shop or 6 hours of your Saturday in N8N.

Now with those 6 hours you just saved, you can focus on operating. You can analyze the actual creative content, come up with new angles, and grow your brand even more.

Here's my prediction: In 12 months, most N8N workflows will be replaced by apps like this and Agent Builder. Not because N8N is bad, it's great for complex enterprise stuff where you can have a dedicated team. But because most DTC operators don't need enterprise complexity. They need simple tools that work without requiring a minor in computer science. Tools that reduce Opex.

Your next step: Make an adjustment to this app. Ask it to add API support if you hit your Gemini limits too fast, or ask it to give you a video score. You'll be shocked how fast you can ship when you stop making it complicated.

And that's it for this week's edition. OpenAI’s “Agent Builder” (AgentKit) looks impressive, but it's a trap. Just going in knowing that! Test the YouTube Analyzer. Modify it for your brand, and experience the glory of practical AI tools.