Agents need more than a chat
summary
Claims that agents need structured, persistent interfaces beyond chat and shows what that looks like: a legal vertical product (Legora) with >1,000 customers and >50 markets, a new UI and tabular review that flags issues, decision log to record agent choices, skills that encode judgment, guardrails (file and site limits), test-driven/browser-backed tasks to make subtasks verifiable, and a recent shipped UI. Demonstrations focus on reducing review overhead and increasing control and trust rather than vague examples.
AI-written summary · about
chapters (9)
- 0:00Introduction
- 1:52Vertical AI and complex agents
- 3:15Verifier's rule and task verification
- 5:20Control and trust in collaboration
- 7:48Guardrails and limitations
- 8:49Planning versus skills
- 10:51Elicitation and decision logging
- 11:56High-bandwidth artifacts
- 13:21Why chat is insufficient
related talks
33mAgent Experience
50mKEYNOTE: The agentification of software has a UX problem
26mAI agents 101
25mAgents are not Apps: Stop building interfaces, Start building coworkers
06mAgents are going to be our co-worker
29mBuilding Trust in AI Agents
15mThe 85% Beyond the Model
45mAn Agent Walks Into a Bar... and Meets Another Agent: Multi-Agent Systems in Action