DEVOURED
Planning with Agents: Divided Worlds, Boundary Objects, and Thicker Interfaces
Maggie Appleton argues that agent planning interfaces should be visual, multiplayer, and interactive rather than static Markdown text.
What: Maggie Appleton, a researcher at GitHub Next, critiques current text-based AI agent planning tools and proposes 'thicker' interfaces—similar to the Chopin prototype—that allow humans and agents to collaborate using visual, editable state machines and interactive reports.
Why it matters: It highlights that current agentic workflows force humans to adapt to a narrow, language-centric interface, ignoring our superior visual and spatial cognition.
Deep dive
- Limitations of text: Humans read at ~240 words/minute, while agents generate at ~4,500 words/minute, causing cognitive overload.
- Situated action: Plans are resources for ad hoc activity, not rigid instructions.
- Boundary objects: Tools like shared interactive reports facilitate better collaboration than static prompts.
- Chopin prototype: A GitHub Next project utilizing MDX for multiplayer, real-time agent/human planning.
- Design strategy: Prioritize visual representations (charts, state machines) to move computation from working memory into the interface.
Decoder
- Boundary Object: Information/tools that allow different groups with different perspectives to cooperate effectively.
- Situated Action: Actions that are improvised in real-time based on environmental context rather than following a pre-planned script.
- Ubicomp: Ubiquitous computing, the concept of integrating computing into the physical environment.
Original article
Full article content is not available for inline reading.
Explore all 58 articles in today's digest
Fresh dev news, AI-digested daily. Never miss a trend at devoured.fyi