Vercel SVP of Product Aparna Sinha breaks down the fundamentals enterprises should think through if they want to keep up with AI technology

💭 Model selection for different use cases
🛂 Controls and fine-tuning
💸 Price

🧵👇

Building an external-facing MCP server? Be prepared to burrow down the identity and OAuth rabbit hole.

🆔 Tool and scope based access
⚡ Supporting DCR (with proper security controls)
📄 Supporting CIMD
✔️ User consent flows

The rabbit hole goes deeper. We did some digging at the AI By the Bay 👇🧵

Do LLMs care about the health of baby seals?

They might, and that can be a problem if used by an adversary. Andre Landgraf (DevRel @ Neon) shared this helpful example (among many others) at the @descope.com Global MCP Hackathon Launch Party earlier this year.

Catch our recap of his session 🧵👇

🏃 Move fast with AI and MCP without breaking things

@descope.com Head of DevRel Kevin Gao shows how the Descope Agentic Identity Hub can apply granular policy controls to govern AI agent access to tools like Google Calendar.

🧵👇

Why is agent identity different than user identity?

⚙️ Agents are task-based, current API keys and OAuth scopes won't map 1:1
🤖 Agents are autonomous, so the concept of least privilege is critical
🤝 Agent intentions are "delegated", the bound user's identity is important

🧵👇

Want to change one of your auth methods?

You can either make it an dev project, or "drag & drop" the new auth method to have...

✅ New auth method in prod
✅ Same brand-aligned login as before
✅ Securely merged user IDs

...without touching your codebase.

With @descope.com, you can just do things!