LukeW on agent interfaces: chat threads break at scale, you need many views
LukeW · x · 2026-10-06
Luke Wroblewski argues that as agents manage other agents in deepening hierarchies, chat threads break down — developers need new ways to see what agents are planning, what they've done, and how to steer them.
He contrasts two approaches:
- Management views: inboxes, Kanban boards, task lists, dashboards, and calendars — they answer "what are my agents doing right now?" and currently get most of the attention.
- Specific dynamic views: task-tied visualizations, which he argues are more useful in practice.
In Intent, his large-scale agent coordination tool for developers, he implements several patterns:
- Agents draw architecture maps, flowcharts, sequence diagrams, and state machines inside notes and chats, so you see the plan before spending tokens.
- Stepped walk-throughs move through a diagram one stage at a time, with short explanations and highlighted parts; boxes linking to files or notes open on click.
- Agents place before/after screenshots of UI changes side by side in notes, letting you judge results without reading code; notes can update per change or persist as history.
Core takeaway: multi-agent coordination needs task-specific dynamic visualization beyond inbox-and-dashboard patterns.
More from coding & agent
- Wardell: the AI opponents in the 2-hour agent-built game mode are surprisingly good — draginol · 2026-10-07
- Stardock CEO: AI producer agents cloned a game mode in 2 hours with Opus, Grok and GPT — draginol · 2026-10-07
- DeskForge generates 1.2M annotated desktop observations, lifting GUI grounding accuracy by 11.5 points — docling-project · 2026-10-07
- 10 lessons from 15 Grokbot guides: one chief-of-staff bot beats twenty chats — leveragedupside · 2026-10-07
- Cortex Harness lets frontier models run real robots, with human takeovers feeding training data — kimmonismus · 2026-10-07
- Blog: Multi-agent systems aren't a new scaling axis, just structured inference scaling — willcb · 2026-10-07