From MCP to Agent Mesh: Building Proactive Agents You Can Trust
MCP is rapidly turning apps into AI-native building blocks, from creative workflows in Figma to coding agents in the IDE. What’s missing is the layer that makes agentic systems real in production: who orchestrates, how permissions are granted and audited, and how we stay safe when inputs are multimodal and actions are real.
Join Figma and Mighty (creators of Citadel) for two lightning talks and a panel AMA on what’s next after the MCP wave: the emerging agent control plane (identity, consent, policy, provenance, and observability).
Figma will bring a practitioner’s perspective on what it takes to develop and ship coding agents in production, tracing the full arc of Figma Make’s evolution from a simple chatbot to more complex architectural pivots and sharing learnings along the way.
Mighty will share lessons from deploying agentic security in practice, including:
- Citadel: a one-line multimodal protection layer for AI systems
- Text Guard (Go, OSS): a CPU-only text guard with a strong Hugging Face model.
- OpenClaw plugins: powered by Text Guard, with optional multimodal upgrades.
We’ll close with an AMA on how to make agents proactive and helpful while staying legible and safe for everyday teams.
Agenda (matches your table)
- 6:00–6:30 Arrival (bar open)
- 6:30–6:40 Host welcome (Mighty + Figma)
- 6:40–7:00 Talk #1 (Rohit Chouhan @ Figma):
The Evolution of LLM Development Powering Figma Make
- 7:00–7:15 Talk #2 (Johnny Hung @ Mighty): LLM-As-A-Judge Is Not Enough to Secure Your Workflows
- 7:15–7:45 Panel AMA: “The Orchestrator Problem” with Moderation by Johnny Hung @ Mighty
Ian Butler @ XBOW
Moritz Stephan @ Cognition
Zach Bai @ Warp
Rohit Chouhan @ Figma
- 7:45–8:30 Networking (food + drinks)