Do people hand whole tasks to AI and walk away the same way they chat with it back and forth?
How does delegated workflow adoption differ from conversational chatbot usage patterns?
This explores how people use AI when they hand a whole task to a structured workflow and let it run, compared with talking to a chatbot turn by turn, and where each pattern shows up and breaks down.
This explores how people use AI when they hand a whole task to a structured workflow and let it run, compared with talking to a chatbot turn by turn. The one note in the collection that measures the difference directly is about who adopts. Workers have committed AI tasks to structured workflows mostly in information-intensive jobs, and that pattern follows what the technology can technically do rather than following conversational LLM adoption Where have workers actually delegated tasks to AI?. It also breaks the older prediction that routine tasks get automated first, and wage patterns reverse at advanced degree levels. So delegation and chat aren't one adoption curve moving at different speeds. They are separate behaviors with different drivers.
The shape of chat helps explain why. Conversational agents are passive by design: their training rewards responding to a query, not leading toward a goal Why can't conversational AI agents take the initiative?. That leaves the user to steer every turn. Simulations suggest proactive dialogue could cut turns by up to 60%, yet it is almost absent from AI datasets and benchmarks Could proactive dialogue make conversations dramatically more efficient?. Delegated systems handle this differently by replacing turns with artifacts. MetaGPT-style agents coordinate through standardized documents instead of conversation, and that beats chatty exchange Does structured artifact sharing outperform conversational coordination?. FlowMind builds workflows by orchestrating calls to vetted APIs, so the model never touches proprietary data and humans can still inspect the result Can LLMs generate workflows without touching proprietary data?.
Delegation has its own failure mode. When nobody is talking, nobody is checking intent. Tool-enabled LLMs drift from what the user wanted through silent tool chaining. Conversation analysis offers a fix in insert-expansions, the brief clarifying questions people ask mid-task When should AI agents ask users instead of just searching?. In delegated work, then, conversation doesn't vanish. It shrinks to a few well-placed questions. Chat's weaknesses are different. Its relational appeal decays as novelty wears off Do chatbot relationships lose their appeal as novelty wears off?, and chatbots miss cues like ambivalence that a real back-and-forth is supposed to catch Why can't chatbots detect when users are ambivalent about change?.
The surprising part is that delegation feels lighter without necessarily being faster. In a study of 73 users, AI-assisted chat cut clicks, page navigations and scrolling, but task completion time didn't change Does chat delegation actually save time on task completion?. Effort and speed move independently, so adoption driven by effort savings may not match adoption driven by results. Interfaces also matter. Generated task-specific UIs beat plain chat in over 70% of cases Do generated interfaces outperform text-based chat for most tasks?, but the clarity costs flexibility, because a UI that is easy to use is harder to modify mid-workflow Do generated analysis UIs really work better than chat?. One likely reason is that a chat's context keeps shifting under the user and can't be internalized the way a fixed interface can How does AI context differ from conventional software context?. Structured workflows give that context somewhere stable to live.
Sources 12 notes
Workers have committed AI tasks to structured workflows primarily in information-intensive occupations, following technical capability more than conversational LLM adoption. This gradient differs sharply from routine-task automation predictions and wage patterns reverse at advanced degree levels.
Research shows LLMs including ChatGPT cannot initiate topics, plan strategically, or lead conversations because their training optimizes for responding to queries, not creating dialogue from agent goals. This passivity is reinforced by alignment objectives and masked by fluent-sounding outputs.
Simulations show proactivity—providing relevant information without being asked—cuts dialogue turns by 60% in medium-complexity domains. This behavior mirrors human conversation and Grice's maxims but is almost entirely absent from AI datasets and research benchmarks.
MetaGPT demonstrates that agents producing standardized engineering documents achieve superior coordination compared to conversational exchange. Active information pulling from shared environments eliminates noise and mirrors efficient human workplace infrastructure.
FlowMind demonstrates that LLMs can generate on-the-fly workflows for spontaneous tasks by orchestrating calls to vetted APIs rather than accessing data directly, eliminating confidentiality risks while maintaining high-level human inspection and feedback.
Show all 12 sources
Tool-enabled LLMs drift from user intent through silent tool chaining. Conversation analysis reveals insert-expansions—clarifying intent, scoping responses, enhancing appeal—as a formal framework for proactive user consultation that prevents misunderstanding instead of recovering from it.
Longitudinal studies with Mitsuku show that social processes driving relationship formation decline as novelty wears off. Single-session study findings cannot be reliably extrapolated to medium- or long-term chatbot design.
Testing three major LLMs across 25 health scenarios showed they succeed only when users have established goals but cannot detect resistance or ambivalence. Models miss relapse-prevention strategies even for users in action stages.
A study of 73 users found that AI-assisted chat interaction significantly lowered clicks, page navigations, and scrolling compared to traditional-only or AI-first modes. However, task duration did not differ significantly across modes, showing effort metrics and completion time move independently.
Research shows users strongly prefer LLM-generated interactive interfaces—dashboards, tools, animations—over text blocks, especially for structured and information-dense tasks. Structured representation and iterative refinement reduce cognitive load.
TaskArtisan found that GUI widgets improve clarity and presentation in LLM-assisted analysis but introduce rigidity and prompting overhead. This trade-off between malleability and specification appears unavoidable: easier-to-use UIs are harder to customize mid-workflow, while flexible UIs demand engineering-style thinking from non-programmers.
AI interactions operate on a substrate of constantly shifting context—prompt, history, retrieved data, hidden state—that users cannot internalize like traditional UIs. This structural mutability demands a new design discipline centered on context engineering rather than interface design.
Papers this line draws on 8
The research behind the notes this line reads — ranked by how closely each paper relates.
- Proactive Conversational Agents in the Post-ChatGPT World
- DiscussLLM: Teaching Large Language Models When to Speak
- Proactive Conversational Agents with Inner Thoughts
- TaskArtisan: Designing Composable Generative Widgets for LLM-Assisted Analysis
- Rethinking Conversational Agents in the Era of LLMs: Proactivity, Non-collaborativity, and Beyond
- Generative Interfaces for Language Models
- FlowMind: Automatic Workflow Generation with LLMs
- Delegating or Doing? Understanding User Behavior in Hybrid Human-Agent Interfaces