Legora CTO at AI Engineer: Chat Is the Wrong Primary Interface for Complex Agents
Jacob Lauritzen, CTO of Legora (vertical AI for 1,000+ law firms across 50+ markets), delivered a talk at the AI Engineer conference arguing that chat is structurally unsuited as the primary interface for complex long-running agents. His core claim: chat collapses a multi-branch agent work tree into a one-dimensional transcript, giving users low bandwidth to review, correct, and steer the agent. Legora ships durable artifacts — documents with per-clause commenting and tabular review sheets — as the primary collaboration surface, using chat only as an input mechanism.
Why It Matters
As agent workflows grow to 10× and 100× larger work trees, the chat transcript becomes genuinely unmanageable for human oversight. Lauritzen's framework — trust and control as orthogonal axes that engineers must push into higher quadrants through verifiability, decomposition, skills, and elicitation — offers a concrete design vocabulary for the agent UX problem that is currently stalling enterprise adoption.