AI Agents: Allow AI to respond alongside an assigned human agent (co-pilot mode)
MKT Administración
Business Problem:
Today, AI Agent can only reply to a conversation while it is assigned to the AI. The moment a conversation is assigned to a human agent, AI Agent stops responding completely — even for repetitive tasks it could still help with.
Desired Outcome:
Allow AI Agent to operate simultaneously with an assigned human agent, without needing to reassign the conversation away from that agent. Specifically:
- The conversation stays assigned to the human agent throughout.
- AI Agent can automatically reply based on configured rules/flows (e.g. welcoming the customer, detecting intent, collecting initial info like vehicle make/model/year, answering FAQs) while the agent is unavailable or hasn't stepped in yet.
- The agent sees AI Agent's replies in real time.
- The agent can pause/deactivate AI Agent at any time via a "Take Control" button, stopping it from replying until reactivated.
- The agent can re-enable AI Agent afterward to resume automation.
Use Case:
A sales team where advisors own follow-up on conversations, but want AI to handle pre-qualification, FAQs, and keeping the conversation active when the advisor isn't immediately available — without constantly reassigning the conversation between AI and human.
N
Nabilah Binti Salleh
Merged in a post:
AI Agent Copilot for Live Assistance, Smart Tagging and Conversation Classification
Istikbal Lebanon
Business problem:
Respond.io’s AI Agent primarily interacts with customers, but human agents also need AI assistance while managing live conversations.
Agents currently have to manually identify the customer’s intent, recognize product names, search for the correct tags, update contact fields and ensure all required classifications are completed before closing a conversation.
This is time-consuming and can lead to:
• Missing or incorrect tags
• Inconsistent product and category classification
• Incomplete reporting data
• Duplicate tags caused by terminology differences
• Conversations being closed before required information is recorded
• Lower productivity across large customer-service teams
Proposed solution:
Introduce a configurable AI Agent Copilot that works alongside human agents inside the Inbox.
The Copilot should analyze the conversation, contact fields, advertisement metadata and workspace terminology to recommend or automatically perform relevant actions.
Core capabilities:
Smart tag suggestions
The Copilot should identify keywords, customer intent, product names, categories, brands, branches and complaint types, then suggest the most relevant existing tags.
For example, if a conversation concerns “Bloom Corner,” it could suggest:
• Bloom Corner
• Corner
• Living Room
• Istikbal
• Product Inquiry
Automatic tagging
Workspace administrators should be able to choose whether tags are:
• Suggested for agent approval
• Automatically applied above a selected confidence level
• Restricted to specific teams, channels or workflows
Required classification before closing
Administrators should be able to define mandatory tag groups or contact fields.
Before closing a conversation, the Copilot could notify the agent:
“This conversation is missing a Product or Inquiry Type tag.”
The workspace could either:
• Show a warning while still allowing closure
• Require the agent to select the missing information
• Allow managers to override the requirement
Workspace-specific learning
The Copilot should understand company-specific terminology, product names, abbreviations, common spelling variations and regional language.
It should improve based on agent-confirmed suggestions and corrections rather than learning automatically from every action, which could reinforce tagging mistakes.
Predictive search
When agents search for tags, snippets, products or contact fields, the Copilot should predict relevant results using the active conversation’s context.
Live agent assistance
The Copilot could also:
• Suggest replies and snippets
• Summarize long conversations
• Identify unresolved customer questions
• Recommend the next action or workflow
• Suggest escalation to another team
• Detect complaints, urgency or buying intent
• Recommend relevant knowledge sources
Administrative controls:
• Enable or disable the Copilot by workspace, team or channel
• Select which tag groups it may suggest or apply
• Configure confidence thresholds
• Define mandatory tags and fields before closure
• Exclude sensitive conversations or channels
• Review an audit log of AI-generated actions
• Monitor acceptance, correction and accuracy rates
• Allow agents to provide feedback on each suggestion
Business value:
An AI Copilot would reduce manual work, improve classification accuracy and create cleaner data for reporting, segmentation and automation.
It would also help new agents follow workspace procedures more consistently and allow experienced agents to manage conversations faster without replacing human judgment.
N
Nabilah Binti Salleh
Merged in a post:
Assign a conversation to a Human Agent for monitoring
A
Andrés Fuseau
Currently, within flows, an AI Agent can be made to assign a conversation to a Human Agent. The problem with this is that when a human agent is assigned, the AI Agent stops interacting and responding. My requirement is that a human agent can be assigned per round robin but only for supervision, without letting the AI agent stop responding. I've tried to set this up in the AI Agent instructions but it doesn't work.