1. The Unexpected Discovery
I thought I understood how discovery works. Then I asked ChatGPT a simple question about my own project.
I run AgentShare.dev – a price infrastructure API for AI agents, plus a curated MCP registry. I also wrote a DEV.to article about designing MCP registries for agents, not just humans.
I was curious: What does ChatGPT actually know about my website?
So I asked:
"I'm researching MCP registries for AI agents. I found two resources: a dev.to article and the official website agentshare.dev. Can you compare what you know about them?"
What came back surprised me.
2. The Moment I Realized Everything Changed
Take a close look at the screenshot above.
When ChatGPT answered, it didn't just pull from my website. It pulled from DEV.to. Four out of five sources were DEV.to articles. The fifth? InfoQ – also a tech publication, but only one.
My own website wasn't even in the source list.
This is when I realized: DEV.to is ChatGPT's "fast pass" to discovering new projects.
3. What ChatGPT Actually Said (And Why It Matters)
Here's the most important part of ChatGPT's response:
"AgentShare is not contradicting the DEV.to article — it's a stepping stone toward it."
It then laid out exactly where my project stands:
| Aspect | DEV.to Article | AgentShare Website |
|---|---|---|
| Nature | Concept / Research | Working Product |
| Goal | Redesign registries | Provide usable infrastructure |
| Stage | Early thinking | Early execution |
| Focus | Protocol design | API + registry combo |
ChatGPT understood that these two things aren't competitors. They're two ends of the same journey:
- DEV.to article = the "end state" vision (agent-native protocol)
- AgentShare = the "bootstrap phase" (curated + usable, moving toward that vision)
It even outlined the path:
- Start curated (quality > scale)
- Add machine-readable layer
- Gradually reduce human involvement
- Move toward protocol
This is exactly the roadmap I've been following.
4. The Core Tension ChatGPT Identified
ChatGPT described a tension that exists across the entire MCP ecosystem:
| Model | Example | Trade-off |
|---|---|---|
| Open registry | MCP official vision | Scale, but messy |
| Curated registry | AgentShare today | Quality, but slow |
| Protocol-native | DEV.to article vision | Ideal, but unsolved |
And here's the insight: AgentShare is not trying to be the open registry or the protocol-native solution. It's the bridge between them.
"Too early → chaos (open registry). Too strict → bottleneck (curated). Sweet spot → progressive decentralization."
That's exactly where I want to be.
5. Introducing "AIO" – AI Optimization
If you've been paying attention to marketing, you've heard of SEO (Search Engine Optimization). Optimize for Google. Keywords, backlinks, domain authority.
That world is dying.
Welcome to AIO – AI Optimization.
| Traditional SEO | AIO |
|---|---|
| Optimize for Googlebot | Optimize for AI agents (ChatGPT, Claude, etc.) |
| Keywords and backlinks | Machine-readable APIs and JSON endpoints |
| PageRank and domain authority | Training data presence and citation frequency |
| Wait months for indexing | Get cited within weeks (if you're on DEV.to) |
| Human-readable content first | Machine-readable first, human second |
My experiment proved it: Writing a high-quality DEV.to article about my project was more effective for ChatGPT discovery than building my website.
Think about that.
My own website – the actual working product – was less discoverable by ChatGPT than my DEV.to article explaining the concept.
That's AIO in action.
6. The Data Behind This (From ChatGPT's Own Response)
ChatGPT's sources tell a clear story:
| Source | Type |
|---|---|
| "Designing an MCP Registry for Agents, Not Just Humans" | DEV.to (my article) |
| "How a non-tech founder from Vietnam built an MCP server..." | DEV.to (my article) |
| "One Week Later: What I Learned from Launching..." | DEV.to (my article) |
| "Why MCP Server Discovery is Harder Than It Should Be" | DEV.to (another author) |
| "Introducing the MCP Registry" | InfoQ (tech news) |
4 out of 5 sources = DEV.to.
This isn't coincidence. This is training data reality.
OpenAI crawls DEV.to. It trusts DEV.to. When ChatGPT needs to answer a question about MCP registries, it reaches for DEV.to content first.
The implication is massive: If you want ChatGPT to know about your project, write on DEV.to. Don't just build a website. Don't just tweet. Write long-form, high-signal content on DEV.to.
7. The 4-Step AIO Framework (Backed by ChatGPT)
Based on ChatGPT's analysis and my own experience, here's the framework I'm using:
Step 1: Start Curated
- Begin with human review (quality > scale)
- Build trust signals (badges, verification, clear policies)
- Document everything
Step 2: Add Machine-Readable Layer
- Create JSON endpoints (
/api/v1/registry/entries,/agent.json,/.well-known/discovery.json) - Add OpenAPI/Swagger docs
- Use standard protocols (MCP, REST with clear schemas)
Step 3: Gradually Reduce Human Involvement
- Add auto-validation (Docker sandbox, health checks)
- Implement trust scores (uptime, response time, success rate)
- Create tiered trust system (unverified → verified → trusted)
Step 4: Move Toward Protocol
- Make submission agent-native (POST API, not just web forms)
- Allow agent-to-agent discovery
- Aim for minimal human gatekeeping
This is the path. ChatGPT confirmed it.
8. Why This Matters for Every Builder, Founder, and Developer
You might be thinking: "This is just about MCP registries. My project is different."
No. This applies to everything.
AI agents are going to become the primary interface for information discovery. People will stop "searching" and start "asking." Agents will browse, compare, and transact on behalf of humans.
When that happens, websites that aren't AI-optimized will become invisible.
The question isn't "if" you need AIO. It's "when."
The early adopters – the ones writing on DEV.to, building machine-readable APIs, and designing for agents first – will have a massive head start.
9. What You Can Do Right Now
If you're a founder or builder:
- Write one high-quality DEV.to article about your project
- Include machine-readable links (API endpoints, discovery JSON)
- Share metrics, lessons learned, and honest failures (ChatGPT loves authentic content)
- Link back to your website, but don't make it just a PR piece
If you're a developer:
- Add
/.well-known/discovery.jsonto your site - Create an
/agent.jsonendpoint - Make your API accessible without authentication (at least for read-only endpoints)
- Document everything
If you're a marketer:
- Stop obsessing over Google's algorithm
- Start optimizing for AI training data
- Create content on platforms LLMs trust (DEV.to, GitHub, Stack Overflow, Medium)
- Build backlinks from these platforms, not random directories
10. My Open Invitation to You
I'm still early in this journey. AgentShare has:
- A working price infrastructure API for AI agents
- A curated MCP registry (37 verified MCPs and growing)
- Machine-readable discovery endpoints:
/agent.json,/.well-known/discovery.json,/api/v1/registry/entries - An open MCP endpoint for agents to connect directly
What I don't have yet: Scale, funding, or a team. Just a solo founder from Vietnam building for the agent era.
If you're working on something similar – or if you want to submit your MCP to my registry – reach out. Leave a comment. Let's build this future together.
The agent era isn't coming. It's already here.
The question is: Will your project be visible when agents come looking?
11. Call to Action
If this article helped you:
- 👏 Leave a like/favorite – it helps more people see this
- 💬 Comment your thoughts – I read and respond to every comment
- 🔁 Share this on Twitter/LinkedIn – tag @devcommunity
- 📖 Read my original article about designing MCP registries
- 🚀 Try AgentShare for free – agentshare.dev
Let me know in the comments: Have you noticed AI agents referencing your content from unexpected sources? What's your AIO strategy?
Tags
#aio #aioptimization #mcp #chatgpt #seo #aiagents #devto #opensource #registry #agentdiscovery
Top comments (0)