KEYNOTE: The agentification of software has a UX problem
summary
Argues that chat-based agents overpromise and underdeliver, citing loss of high‑bandwidth context, the prompting dance, uncertain attribution, and a real example that promised 250 bucks per person but refused to act. Names concrete UI experiments—language guey, census.app, AI UX patterns—and proposes built interactions: mermaid-style diagrams, DAW‑like agent tracks, histograms for risk/ambition, and fine‑grain permissions instead of bot‑tokens. Shows what was built, what fails (steerability, siloed agents, god‑mode tokens) and what actually works enough to iterate.
AI-written summary · about
chapters (26)
- 0:00Introduction
- 3:13Agentification UX problem
- 5:25Defining utopian agent interaction
- 6:29Chat interfaces fall short
- 10:05Text-based limits
- 12:36Richer UI for agents
- 15:47Steerability in interactions
- 17:55Software patterns for agent UX
- 20:25Current agent interactions
- 23:15Chat like shower curtain
- 23:35Business context of agentification
- 24:17Nobody wants to be a tool
- 26:47Agent gateway browser wars
- 27:29Flight delay claim example
- 31:24Trust issues with permissions
- 35:42Most dangerous part of agent
- 36:02Potential futures for agents
- 37:27UX challenges for engineers
- 38:54Q&A
- 41:00Voice and video interfaces
- 41:20Whiteboard diagrams and outputs
- 43:05Why not build AI apps
- 45:12Open vs closed models
- 46:14Humans more dangerous than LLMs
- 47:18Personalization fatigue
- 49:27Closing remarks
related talks
57mDesigning Human-AI Experience
36mThe UX of AI: Making AI-Powered Apps Your Users Don't Hate
46mDesign Patterns For AI Interfaces
21mDesigning in the age of the 10x engineer
19mBeyond agents: AI as a creative partner
32mKeynote: Architecting Systems for Agent Experience (AX)
33mAgent Experience
20mMicrointeractions for the System: Design in the Age of Techlash