Matt Jedraszczyk | Product Designer | New Tech | Startups
@mattjed.bsky.social
Agentic Systems Architect. Building Morphic Interfaces & AI Swarms. Bridging Enterprise (Microsoft) with Open Source (Langflow, Shadcn). AI Lead @ Grant Thornton. No black boxes allowed.
UX Designers in 2026 need to understand Orchestration flows just like they had to understand HTML/CSS in 2010.
If you can't design the "Supervisor Interface" for the swarm, you are designing legacy tech.
If you can't design the "Supervisor Interface" for the swarm, you are designing legacy tech.
December 15, 2025 at 10:21 AM
UX Designers in 2026 need to understand Orchestration flows just like they had to understand HTML/CSS in 2010.
If you can't design the "Supervisor Interface" for the swarm, you are designing legacy tech.
If you can't design the "Supervisor Interface" for the swarm, you are designing legacy tech.
The challenge isn't building it (Low-code tools like Langflow or n8n make it easy). The challenge is Governance.
Under the EU AI Act, B2B creates a "No Black Box" rule artificialintelligenceact.eu/article/13/. If Agent B rejects Agent A's work, the Human-in-the-Loop needs to know why.
Under the EU AI Act, B2B creates a "No Black Box" rule artificialintelligenceact.eu/article/13/. If Agent B rejects Agent A's work, the Human-in-the-Loop needs to know why.
Article 13: Transparency and Provision of Information to Deployers | EU Artificial Intelligence Act
← Back to index
artificialintelligenceact.eu
December 15, 2025 at 10:21 AM
The challenge isn't building it (Low-code tools like Langflow or n8n make it easy). The challenge is Governance.
Under the EU AI Act, B2B creates a "No Black Box" rule artificialintelligenceact.eu/article/13/. If Agent B rejects Agent A's work, the Human-in-the-Loop needs to know why.
Under the EU AI Act, B2B creates a "No Black Box" rule artificialintelligenceact.eu/article/13/. If Agent B rejects Agent A's work, the Human-in-the-Loop needs to know why.
The new standard involves specialized agents, not one general LLM:
1. Agent A (Researcher): RAG + CAG for data gathering (depending on case).
2. Agent B (The Critic): Validates A's output (Hallucination & govenance check).
3. Agent C (Executor): Only acts if B approves.
1. Agent A (Researcher): RAG + CAG for data gathering (depending on case).
2. Agent B (The Critic): Validates A's output (Hallucination & govenance check).
3. Agent C (Executor): Only acts if B approves.
December 15, 2025 at 10:21 AM
The new standard involves specialized agents, not one general LLM:
1. Agent A (Researcher): RAG + CAG for data gathering (depending on case).
2. Agent B (The Critic): Validates A's output (Hallucination & govenance check).
3. Agent C (Executor): Only acts if B approves.
1. Agent A (Researcher): RAG + CAG for data gathering (depending on case).
2. Agent B (The Critic): Validates A's output (Hallucination & govenance check).
3. Agent C (Executor): Only acts if B approves.