Building agent-first developer experiences with ElevenLabs

Paul Asjes has spent years building SDKs at Stripe and WorkOS.

Now, as DevX Lead at ElevenLabs, he's watching developer education become agent education, and building a three-part strategy to make sure agents actually know how to use his product.

In this episode, we cover:
Why your agents are generating code against APIs that don't exist anymore, and why that's an architectural problem, not a training one

The OpenAI research showing LLMs hallucinate because guessing beats silence

Why dumping your entire docs into an agent's context makes it perform worse, not better

What agent skills are, how they differ from MCP tools, and why you need both
How ElevenLabs built a music player inside Claude Desktop with MCP apps

The security and versioning problems nobody's solved yet for skills registries

Why Paul thinks "software engineer" becomes "agent manager" within a few years

The 99-cent app store vision for MCP apps.

⏰ TIMESTAMPS:
0:00 – Orchestrating agents instead of writing code
1:01 – Intro and the tab key debate
2:17 – DevRel shift: docs and StackOverflow to LLM-first
3:56 – ElevenLabs: developer education becomes agent education
5:17 – What “Skills” are and why they matter
6:01 – Why LLMs hallucinate (the guessing problem)
8:31 – MCP best practices: tasks over 1:1 API wrappers
11:26 – ElevenLabs MCP server and Agent Skills standard
18:49 – MCP Apps and the UI layer unlock
36:54 – The open problem: versioning and keeping skills updated

🔗 LINKS & RESOURCES:

ElevenLabs Agent Tools

ElevenLabs Skills: https://github.com/elevenlabs/skills
ElevenLabs MCP Server: https://github.com/elevenlabs/elevenlabs-mcp
ElevenLabs MCP Player (MCP App): https://github.com/elevenlabs/elevenlabs-mcp-player
ElevenLabs Blog, Agent Skills: https://elevenlabs.io/blog/elevenlabs-agent-skills
ElevenLabs Blog, MCP Server: https://elevenlabs.io/blog/introducing-elevenlabs-mcp

Skills Ecosystem
Agent Skills Specification: https://agentskills.io/home
Agent Skills GitHub: https://github.com/agentskills/agentskills
Skills.sh Registry: https://skills.sh/
Tessl (skills evaluation): https://tessl.io/
Vercel Skills CLI: https://github.com/vercel-labs/skills

MCP Apps & Specification
MCP Apps Spec: https://modelcontextprotocol.io/docs/extensions/apps
MCP Protocol: https://modelcontextprotocol.io/

Referenced in this episode
OpenAI hallucination research: https://openai.com/index/why-language-models-hallucinate/
llms.txt: https://llmstxt.org/
XKCD 927 (competing standards): https://xkcd.com/927/
Lakera on prompt injection risks in skill registries: https://www.lakera.ai/blog/the-agent-skill-ecosystem-when-ai-extensions-become-a-malware-delivery-channel

About the guest
Paul Asjes, DevX Lead at ElevenLabs
Website: https://asjes.dev/
Bluesky: https://bsky.app/profile/asjes.dev
X: https://x.com/paul_asjes

MCP MVP is a series from Arcade spotlighting the people building the agentic ecosystem. New episodes dropping regularly.

💬 Join the conversation: https://discord.gg/GUZEMpEZ9p

🔧 Build with Arcade: https://arcade.dev

📖 Arcade Docs: https://docs.arcade.dev

Видео Building agent-first developer experiences with ElevenLabs канала Arcade