TL;DR: OpenAI just launched apps in ChatGPT, and the tech press is calling it revolutionary. But here’s what they’re not telling you: the underlying technology can do something far more powerful than what ChatGPT apps actually deliver. Why build a limited version? Because true multi-agent collaboration is harder to control, monetize, and keep inside a walled garden. And creative professionals are missing out.
Here’s the thing about walled gardens.
They’re beautiful at first. Lush. Well-maintained. Everything you need in one place. No hunting for tools scattered across the internet. No switching between apps. Just one gate, one path, everything organized and accessible.
Then you realize: there’s only one gate.
And someone else controls it.
But there’s something worse than a walled garden: a walled garden that could have been a forest.
OpenAI dropped apps in ChatGPT last week and the response has been predictable. Tech journalists calling it “revolutionary.” Partners like Spotify, Canva, and Zillow betting their distribution on it. Marketing LinkedIn already buzzing about “consolidating workflows.”
The pitch is seductive: chat with your apps instead of clicking through them. Want a playlist? “Spotify, make something for Friday’s client drinks.” Need slides? “Canva, turn this outline into a deck.” Looking for a restaurant? “OpenTable, find somewhere impressive near the office.”openai
Natural language. Integrated experiences. Everything in one conversational interface.
It’s convenient as hell.
But here’s what nobody’s talking about: the technology underneath can do something far more interesting than what OpenAI actually built.
What MCP Actually Is (And Why It Matters)
ChatGPT apps are built on Model Context Protocol. And before your eyes glaze over, stick with me—because this is where it gets interesting.
Think of MCP like this: You know how your creative team works best? When the researcher, strategist, copywriter, and designer are all in the same room, riffing off each other in real-time? Someone mentions an insight, the designer immediately sketches a visual direction, the copywriter builds on that idea, the strategist sees an opportunity nobody else spotted?
That collaborative flow—where multiple specialists work simultaneously, sharing context, building on each other’s output—that’s what MCP was designed to enable for AI agents.
Before MCP existed, connecting AI to tools was like carrying different chargers for every country you visit. Custom-built. Expensive. Fragile. Every integration required separate engineering work.
MCP is the universal adapter. But more importantly, it’s designed for multiple AI agents working together at the same time.
Here’s what that could mean for creative work:
Imagine briefing a campaign and having AI agents collaboratively:
- Pull relevant audience research from your CRM while
- Analyzing competitor campaigns from your saved references while
- Generating strategic territories based on both inputs while
- Sketching visual concepts that ladder up to the strategy while
- Drafting copy that integrates all of the above
All happening simultaneously. All sharing context. All building on each other’s work in real-time.github
Not sequential. Not one-tool-at-a-time. Collaborative intelligence.
That’s what the technology can do. That’s what MCP was built for.
ChatGPT apps don’t work that way at all.
What ChatGPT Apps Actually Do Instead
Here’s how ChatGPT apps actually function:
You invoke one app at a time. “Spotify, make a playlist.” That conversation happens. Then it ends. Want to use Canva next? Start a new interaction.
One app. One task. Done.
It’s not collaborative. There’s no simultaneous multi-agent coordination. No shared working memory across tools. No real-time building on each other’s outputs.
It’s like having your entire creative team in the same building but forcing them to work in separate rooms with locked doors.github
Sure, they’re technically “integrated” in the sense that they’re all accessible from the same interface. But they’re not actually collaborating.
The technical term for what ChatGPT apps do: sequential single-tool invocation.
The technical term for what MCP enables: parallel multi-agent orchestration with shared context graphs.
One of these is dramatically more powerful than the other.
What You’re Actually Missing
Let me make this concrete with a real creative workflow.
Scenario: You’re developing a campaign for a new product launch.
What MCP Could Do (Multi-Agent Collaboration):
You start: “Help me develop a campaign strategy for [product blah] targeting [audience blah].”
Simultaneously:
- Research agent pulls competitive analysis, audience data, cultural trends from connected databasesgithub
- Strategy agent identifies white space opportunities based on that researchquadone
- Creative agent generates conceptual territories aligned with the strategy
- Design agent sketches visual treatments for the strongest territories
- Copy agent writes headlines that integrate the strategic insight and visual direction
They’re all working at the same time. Sharing what they learn. Building on each other’s outputs.github
When the research agent discovers a cultural insight, the strategy agent immediately incorporates it. When the strategy solidifies, the creative and design agents adapt their work accordingly. In real-time.
The output isn’t five separate things you then manually combine. It’s one integrated, strategically coherent campaign developed collaboratively.
What ChatGPT Apps Actually Do (Sequential Single-Tool):
You say: “Canva, make me a presentation deck.”
Canva makes slides. End of interaction.
Now you want research? New conversation. Different tool. Start over. The research doesn’t inform the design you already created. Because the tools never actually talked to each other.
See the difference?
One is a creative team working together. The other is a very polite receptionist transferring your call to one department at a time.
Why Build The Limited Version?
This is the question that should bother you.
The technology for true multi-agent collaboration exists. Open-source implementations are already demonstrating it. The Model Context Protocol was literally designed for this.
So why did OpenAI build the single-interaction version instead?
Because isolated, sequential app interactions are easier to control.
With true multi-agent collaboration, you need:
- Apps that can actually communicate with each other
- Shared memory and context that persists across agents
- Parallel processing with distributed coordination
- Open standards so agents from different developers can work together
That’s harder to monetize. Harder to gate-keep. Harder to lock into one ecosystem.cybernews
Single-app invocations? Much simpler business model:
- Each app is a separate relationship with OpenAI
- Each interaction is trackable, monetizable, controllable
- OpenAI sits in the middle of every transactionopenai
- No need for apps to work together—so no need for genuine interoperability
The limited implementation isn’t a technical constraint. It’s a business model choice.
And that choice is costing you access to genuinely powerful collaborative AI.
The Playbook You’ve Already Lived
Here’s where the technical limitation meets the business model trap.
If you work in advertising or marketing, you know this story by heart.
Adobe: Buy Photoshop once and own it forever? Those days are gone. Forced subscription model, then steady price increases—Photography plan up 50% in 2025, All Apps jumping from $60 to $70/month. Oh, and remember when your software was yours? Now you’re renting it permanently.
Pantone: For decades, Pantone colors just worked in Adobe apps. Then suddenly in 2022: pay for a separate Pantone Connect subscription or watch your existing files turn to black. Colors you’d already used in client work. Held hostage until you paid the new fee. That’s not a feature update—that’s ransomware with better PR.
Facebook: Build your audience here! Free reach! Direct connection to fans! Then—surprise—organic reach drops to nothing. Want your own followers to see your posts? Pay up.
Instagram: Beautiful platform. Easy posting. Chronological feed. Then the algorithm arrives. Suddenly your most engaged followers don’t see your work unless Instagram’s machine learning decides it’s “engaging” enough.
Twitter/X: Open API. Developers build businesses. Ecosystem thrives. Then $42,000/month for API access. Entire companies destroyed overnight.
Cory Doctorow calls this enshittification. Three stages, always the same:
Stage 1: Be good to users. Attract them with value, convenience, free access.
Stage 2: Abuse users to benefit business customers. Now that they’re locked in, squeeze them. Algorithm changes. Paywalls. Forced monetization.
Stage 3: Abuse everyone to benefit shareholders. Extract maximum value as quality craters.
ChatGPT apps are Stage 1 with Stage 2 architecture already built in.
Stage 1: The Seduction
Right now? It feels great.
800 million ChatGPT users getting free app access. Natural language making everything frictionless. Your favorite tools—Canva, Spotify, Kayak—right there in the chat.
The experience genuinely is better than juggling separate apps.
And while you’re enjoying that convenience, here’s what’s happening:
Your chat history? Locked in OpenAI’s ecosystem. Your workflow patterns? Captured by their platform. Your app discovery? Controlled by their algorithm.
Apps don’t work outside ChatGPT. They’re not standalone. They’re not portable. They only exist where OpenAI allows them to exist.
And they’re deliberately built to not work together.
Because if they actually collaborated—if you could build genuinely powerful multi-agent workflows—you might realize you don’t need OpenAI sitting in the middle of every interaction.
Stage 2 Signals: Hiding in Plain Sight
Here’s what’s interesting. OpenAI isn’t even hiding the business model. It’s right there in the announcement. You just have to read between the lines.
“Monetization details coming soon.”
Translation: we’ll tell you what this costs after you’ve rebuilt your workflow around it.
App directory with “prominence” and “featured” placement coming later this year.
Translation: algorithmic curation. Paid visibility. The app store model where developers pay for discovery.
“Instant checkout” through the Agentic Commerce Protocol.
Translation: we’re taking a cut of every transaction that flows through this platform.
Apps “currently” work without paid promotions.
Translation: enjoy this while it lasts.
Oh, and one more thing: apps aren’t available in the EU or UK. Why? Probably because European regulations require genuine interoperability and data portability—things that don’t play well with walled gardens.
The walls are already up. Most people just can’t see them yet.
What Actually Powerful Looks Like
This isn’t theoretical. Real implementations of MCP’s multi-agent capabilities exist right now.
Agent-MCP frameworks are demonstrating true collaborative intelligence: multiple specialized agents working simultaneously, sharing memory, building on each other’s outputs without platform gatekeepers.
Open orchestration systems let you combine agents from different providers, running different models, all coordinating through standard protocols.
Self-hosted MCP servers give you the infrastructure for genuinely collaborative AI without vendor lock-in.
The technology exists. It works. It’s more powerful than what ChatGPT apps deliver.
So why aren’t you getting it?
Because OpenAI chose to build the controllable version instead of the powerful one.
The Real Cost of This Choice
Let me paint you a picture of what you’re trading away.
What you’re getting: Convenient single-app interactions. One at a time. Isolated from each other. Controlled by OpenAI’s platform.
What you could have: True creative collaboration between AI agents. Simultaneous work. Shared context. Emergent insights from multiple specialists working together. Platform-agnostic tools you actually control.
The gap between those two things? That’s not just about features. It’s about fundamentally different approaches to AI tooling.
One approach treats you as a user who needs controlled access to isolated tools.
The other treats you as a professional who needs a collaborative team.
One approach keeps you dependent on a platform.
The other gives you genuine leverage to do better work.
And here’s the kicker: 18 months from now, when the pricing changes and the paywalls arrive, you won’t just be locked into an expensive platform.
You’ll be locked into an expensive platform that’s delivering a fraction of what the underlying technology can actually do.
What Lock-In Actually Costs (The Business Model Endgame)
It’s 18 months from now. Your entire creative workflow runs through ChatGPT apps. Research, ideation, design, client presentations—all integrated, all convenient, all in one place.
Then OpenAI announces ChatGPT Professional Plus.
The apps you use daily? Now they’re in the premium tier. Basic users get limited access. Want full functionality? Upgrade.
Or maybe: featured apps get priority in the algorithm. The tools you prefer aren’t “prominent” enough. OpenAI’s partners get suggested first. Your workflow disrupted because someone else’s business deal matters more than your preferences.
Or perhaps: apps require revenue sharing. Developers who don’t pay the platform tax lose visibility. Your favorite tools disappear or get buried because they won’t play ball.
And here’s the worst part: even if you’re willing to pay, you’re still getting the neutered single-interaction version instead of true multi-agent collaboration.
Sound far-fetched?
Every single one of these scenarios has already happened on existing platforms. Instagram buried posts from accounts that linked externally. Twitter destroyed its developer ecosystem overnight. Facebook turned organic reach into a pay-to-play game.
Why would OpenAI be different?
What Actually Open Looks Like
This isn’t theoretical. We have examples of what genuine interoperability enables.
True MCP implementations support multiple agents from different providers working together. You could have a research agent from one company, a strategy agent from another, and a design agent from a third—all collaborating in real-time.
Open orchestration frameworks let you build workflows that move between platforms. Start in ChatGPT, move to Claude, incorporate open-source specialized agents. Your workflow. Your choice. Your control.quadone
Self-hosted infrastructure means you can run collaborative AI systems without platform intermediaries extracting value from every interaction.github
Browser-based tools work anywhere without vendor permission or gatekeeping.byteplus
ChatGPT apps claim to be built on an “open standard.”
But here’s the test: can these apps actually collaborate with each other? Can they work outside ChatGPT? Can developers distribute them independently? Can users access true multi-agent orchestration?
No. No. No. And no.
That’s not open. That’s proprietary distribution with an open-source SDK.
It’s like Apple calling their devices “open” because developers can write apps—while conveniently forgetting that every single app must go through Apple’s review process, live in Apple’s store, and give Apple 30%.
What Creative Professionals Should Demand
Look, I’m not saying never use ChatGPT apps. If they solve real problems, use them.
But demand better. Because the technology for better already exists.
Demand true multi-agent collaboration. Not sequential single-tool invocations. Actual parallel coordination with shared context. AI agents that work together the way your creative team does.
Demand platform-agnostic tools. Apps that work across interfaces, not just inside OpenAI’s walled garden. Standards that enable genuine interoperability.
Demand transparent data practices. Know what’s being captured, how it’s being used, who benefits from the value your workflows create.
Keep alternatives. Don’t let your entire workflow depend on one vendor’s limited implementation of powerful technology. When they change terms—and they will—you need options.
Support genuinely open implementations. The developers building real multi-agent MCP systems without platform lock-in. The frameworks enabling actual collaborative AI. The standards that give you control.
The 2023 OpenAI leadership crisis taught smart companies a lesson: never bet your entire business on one vendor’s API. When the company nearly imploded over a weekend, the companies that had built multi-vendor strategies felt vindicated. The ones locked into OpenAI exclusively? Panicked.
The same principle applies to ChatGPT apps—except now you’re not just risking vendor lock-in. You’re accepting a deliberately limited version of genuinely transformative technology.
The Thing About Walled Gardens
They’re beautiful. Until you try to leave.
They’re convenient. Until the prices change.
They’re innovative. Until you realize what they’re not showing you.
It’s not about making AI tools as powerful as they could be.
It’s about making them as controllable as the business model requires.
Don’t settle for the neutered version just because it’s convenient. Don’t give them the only key to your creative workflow. And for fuck’s sake, don’t accept isolated single-app interactions when true collaborative intelligence is already possible.
You deserve better tools. The technology already exists for better tools.
The question is whether you’ll demand them—or settle for the walled garden because the gate looks pretty.
//A 🌱⚡