Skip to content
Silent Potato.
SearchTagsEN/中文
  • Start
  • Writing
  • Columns
  • Projects
  • Research
  • Work with me
  • Photos
  • About

Research · preprint · 0.2

Making an Agent a Collaborable Object: State, Feedback, and Result Confirmation

Based on ChatLab design work, this paper records how presence, response, waiting, and delivery patterns from human collaboration can make Agent runtime states easier to understand; it is an HCI design observation about feedback, evidence, intervention, and result confirmation.

LiyukLiyukPublished August 19, 2026
  • #Human-Computer Interaction
  • #Agents
  • #Feedback
  • #Observability
  • #Recoverability
  • #Authorization
  • #Human-AI Collaboration
  • #Systems Design
  1. 1The Five Lenses: Connection-Oriented Problem LocationDeep research · 19 min read
  2. 2Data Measurement as Organizational Protocol: Definitions, Measurement, Tiering, and RetrospectivesDeep research · 11 min read
  3. 3Developer Productivity Is Not a Tool Catalog, but a Feedback SystemDeep research · 9 min read
  4. 4Defining the Boundaries Before Bringing AI Capability into an Engineering OrganizationDeep research · 9 min read
  5. 5When AI Lowers Workflow Barriers: How to Redivide Functional Lines and Business LinesDeep research · 14 min read
  6. 6Decompose First, Then Schedule: A Review of Multi-Model Task Decomposition, Capability Switching, and Subtask RoutingDeep research · 13 min read
  7. 7Making an Agent a Collaborable Object: State, Feedback, and Result ConfirmationDeep research · 13 min readThis chapter
  8. 8Let the Agent Execute: Emotional Adaptation, Trust Calibration, and Human Relief from Execution PressureDeep research · 13 min read
  9. 9The Laws of Human Motivation: The Situational Motivation ModelDeep research · 29 min read
  10. 10AI Does Not Automatically Create Productivity: From Local Acceleration to System ValueShort · 5 min read

Share to

XWeiboTelegramWhatsAppLinkedInFacebook

WeChat

Scan with WeChat

Open this article on your phone, or forward it to a friend.

Enjoy this site?

orSubscribe via RSS
← PreviousLet the Agent Execute: Emotional Adaptation, Trust Calibration, and Human Relief from Execution Pressure
Next →Decompose First, Then Schedule: A Review of Multi-Model Task Decomposition, Capability Switching, and Subtask Routing
View the column “Engineering & AI Judgment”← Previous: Decompose First, Then Schedule: A Review of Multi-Model Task Decomposition, Capability Switching, and Subtask RoutingNext: Let the Agent Execute: Emotional Adaptation, Trust Calibration, and Human Relief from Execution Pressure →

Keep reading

Maybe related to this one.

  • WritingAI Does Not Automatically Create Productivity: From Local Acceleration to System ValueStarting from a multi-system integration experience, this essay distinguishes creation, task efficiency, organizational productivity, and business value—and asks where value and cost actually come from when AI enters a complex system.Read on →

    same column · shared 2 tags

  • WritingWhere Exactly Is the Cache? Locality in a Multi-Layer Execution ServiceUsing a multi-layer execution service as an example, this article breaks down the multiple levels of locality across request identity, edge routing, execution resources, and backend caches, then presents reusable approaches to routing, invalidation, failover, and observability.Read on →

    shared 2 tags

  • WritingA Data Center Is Not a Report: From Data Production to Business JudgmentStarting from the different questions asked by analysis, operations, and engineering, this article organizes how a data center for a multi-source business platform should divide Tabs, define metrics, structure read and write paths, and handle billing, reconciliation, growth, and cost.Read on →

    shared 2 tags

© 2018–2026 Liyuk. Built slowly, published openly.

ElsewhereGitHub ↗X ↗LinkedIn ↗Email ↗Links ↗Favorites ↗RSS ↗
CC BY-NC-SA 4.0