Calm Down, TechBros. ChatGPT Agents Aren’t Killing Zapier, Make.com, or n8n
AgentKit Changes the Game — Just Not the One You Think.
Hey Friends!
It’s been a whirlwind few weeks — part Gramaternity Leave (welcome, Kieran!), part toddler-wrangling (so much K-Pop Demon Hunters with Elowyn), and part sprint to get Author Nation programming and four special Indie Author Magazine issues out the door. Will I see you in Vegas in 3 short weeks?
The Hub is officially up and running, with all users migrated, a few workflows active, and the first batch of prompts and events ready to go. We’re sitting at about one-eighth of the content migrated — and now that there’s a real home for everything, the rest is on its way. Let me just catch a power nap and answer 400 emails real quick.
Remember when I said APIs and webhooks aren’t just for devs in hoodies? That day has officially arrived. Sort of.
OpenAI just launched AgentKit, a suite of tools that lets you build AI agents that can connect directly to your business apps. This means you can now tell ChatGPT to go pull data from Airtable, create a Stripe invoice, check your calendar, send an email, or spin up a draft post based on last week’s sales. Claude’s already doing this too, by the way, using connectors that tie into similar backend structures.
The tech works. I’ve been testing it for months across ChatGPT, Claude, and TypingMind. It’s fast, powerful, and genuinely useful—in the right context.
But let’s not skip straight to the “RIP Zapier” hot takes. The internet is already announcing the death of Zapier, Make.com, and n8n. That couldn’t be more off base. We’re not replacing automation—we’re adding a new layer on top of it.
Let’s unpack what’s actually changed, where the risks are buried, and how to build workflows that won’t leave you stuck inside someone else’s sandbox six months from now.
What We’ve Had For a While Now
If you’ve been following along, you saw me write about MCP (Model Context Protocol) back in May. It’s the behind-the-scenes protocol that allows large language models like ChatGPT and Claude to securely talk to apps like Airtable, Notion, Stripe, and Gmail through a connector server.
With an MCP setup in place, you can say things like, “Show me everyone named Dave on my newsletter list,” and your AI assistant will call the relevant API in the background and deliver the answer right inside your chat window. You don’t need to know what an API is or how authentication tokens work. MCP translates your plain-language request into structured commands and gets the job done—assuming the server is configured properly.
That’s been possible for months now. MCP servers make it easier for non-technical folks to use AI tools in real-world workflows without needing to mess with raw API calls or build middleware from scratch.
What AgentKit Actually Does (And Why It’s Cool)
AgentKit is OpenAI’s next step: a developer-friendly, UI-driven toolkit for building smarter agents that can do more than just respond to a one-off query. It includes:
Agent Builder – a visual canvas where you define your agent’s logic (if/then conditions, functions, branching behavior).
Connector Registry – a centralized dashboard where you control what apps your agent can access and how it uses them.
ChatKit – tools to embed these agents directly into your site, app, or internal tools—so you’re not just working inside ChatGPT’s interface.
Evaluation Tools – to measure performance, identify failures, and improve how your agents behave.
Guardrails – built-in safety controls to mask sensitive data and reduce the risk of rogue outputs.
In short, if MCP is the translator between your tools and your AI model, AgentKit is the workspace for hiring, training, and supervising the agent doing the actual work.
It’s no longer just “pull this data” or “send this email.” With AgentKit, you can build an agent that checks inventory, drafts a message, and routes that message based on customer type—all in one flow. It’s genuinely useful for orchestrating multi-step tasks, especially when you want to keep a human in the loop.
That’s exciting.
But—and this is important—it still isn’t automation.
Agents Are Not Automation (Yet)
Here’s the critical distinction: ChatGPT’s agents (and Claude’s, for that matter) still require you to initiate the process. You have to start the conversation, provide the instructions, and define what should happen next.
That’s assisted workflow. It is not automation.
Tools like Zapier, Make.com, and n8n don’t wait for you to show up. They listen for triggers. They react when something happens—like a Stripe payment or a form submission—then execute a series of actions behind the scenes while you sleep.
They are proactive. They are dependable. And they don’t need your attention to run.
💡 Not All Agents Are the Same
The term “AI agent” is everywhere right now, but it means different things depending on the platform.
- The agents in n8n and Make.com are autonomous systems that carry out multi-step workflows once they’re given an objective. They act without further input.
- The agents in ChatGPT or Claude are powerful assistants—but they’re still reactive. They wait for a human to prompt them.
The difference isn’t just technical—it’s functional. One automates. The other assists.
If your business relies on workflows that need to happen whether or not you remember to open a chat window, you still need traditional automation tools. Full stop.
Let’s Talk About Security (Because Most People Won’t)
Right now, not every app has an official MCP integration. So, developers—some trustworthy, some questionable—are building third-party connectors and publishing them on GitHub or marketplace clones.
Here’s where things get dangerous.
When you use one of these unofficial connectors, you’re handing over access to your business systems. That includes customer data, email accounts, payment platforms, and internal documents. Unless you know exactly who built the connector and how they’re storing your keys, you’re taking a massive risk.
This isn’t fear-mongering. It’s just how security works.
Use official integrations when they exist. If you’re going third-party, make sure it’s someone you trust.
To help with that, I launched MCP Server Hub. It’s a curated collection of secure, vetted MCP connectors built specifically for authors and digital business owners who don’t want to bet their backend on a stranger’s weekend project. Right now I have connections for Stripe, Freepik, Airtable, TypingMind, Lulu, Make.com, n8n, and FFMPEG. If you want access, drop me an email. It’s by invitation/approval only.
Treat your credentials like gold-plated keys to the kingdom. Because that’s exactly what they are.
Lock-In Is the Bigger Threat
Even if you use a secure connector, there’s another long-term risk: vendor lock-in.
When you build your entire workflow stack on top of OpenAI’s AgentKit, you’re placing a lot of trust in one company. If they change their pricing, alter how connectors work, or introduce restrictions that impact your setup, you’re stuck.
This is why I don’t lock myself into a single LLM ecosystem.
I use TypingMind, which works across multiple models (Sonnet 4.5, GPT5, GPT-4o, Claude, Mistral, Gemini, Perplexity) and allows me to bring my own API keys. If something breaks—or if one provider suddenly becomes prohibitively expensive—I can switch models without rebuilding my workflows from scratch.
Being tech-stack agnostic isn’t just smart. It’s essential.
The Stack That Actually Works
Here’s how I recommend integrating these tools into your current workflow:
Use Zapier, Make.com, or n8n for structured, repeatable automations. These tools are your operational foundation.
Use ChatGPT or Claude for interactive, on-demand tasks that benefit from AI intelligence but still need your input.
Use secure MCP connectors from MCP Server Hub to bridge your data tools with AI—without compromising your privacy or security.
This isn’t about choosing one system over another. It’s about building the right blend of tools that work together, safely and reliably, without putting your business at the mercy of launch-day features.
Want to See The New Hub?
If you’re trying to figure out how AI fits into your workflow—without breaking what already works—I’m hosting a live session on Tuesday, October 14 at 11am Central.
We’ll do a full walkthrough of the new Author Automations Hub—how it works, how to install workflows, how the prompts are structured, what the tutorials cover, and how to use the MCP Server Hub to connect your apps securely.
No fluff. No hype. Just real tools, real risks, and what’s actually worth doing right now.
See you Tuesday,
Chelle