Long story short
Long story short
Redesigned the AI agent creation flow in Flametree to make it accessible for non-technical users.
Redesigned the AI agent creation flow in Flametree to make it accessible for non-technical users.
Replaced a code-heavy setup with a natural language interface, reducing support dependency and enabling self-serve agent creation.
Replaced a code-heavy setup with a natural language interface, reducing support dependency and enabling self-serve agent creation.
Context
Context
Flametree is a platform that allows businesses to create and deploy AI agents for customer communication across channels like web widgets, Telegram, and WhatsApp.
Flametree is a platform that allows businesses to create and deploy AI agents for customer communication across channels like web widgets, Telegram, and WhatsApp.
The core product value lies in enabling users to configure and launch their own AI agents — defining how they behave, what they know, and how they interact with customers.
Initially, the agent setup experience was designed with a strong technical bias, making it difficult for non-technical users to adopt.
The core product value lies in enabling users to configure and launch their own AI agents — defining how they behave, what they know, and how they interact with customers.
Initially, the agent setup experience was designed with a strong technical bias, making it difficult for non-technical users to adopt.
Problem
Problem
The flow was overly technical, exposing low‑level concepts and settings, which prevented new and non‑technical users from creating agents independently and severely limited product adoption.
The flow was overly technical, exposing low‑level concepts and settings, which prevented new and non‑technical users from creating agents independently and severely limited product adoption.
The existing setup flow was heavily oriented toward technical users:
Key configurations were defined using Python code
Concepts like chunking, workflows, and conversation logic were exposed directly
The interface included many unclear or low-level settings
As a result:
New users often failed to create even a single agent during the trial period
Users relied heavily on support to complete setup
Non-technical users were effectively blocked from using the product independently
This created a major gap between product potential and actual user activation.
The existing setup flow was heavily oriented toward technical users:
Key configurations were defined using Python code
Concepts like chunking, workflows, and conversation logic were exposed directly
The interface included many unclear or low-level settings
As a result:
New users often failed to create even a single agent during the trial period
Users relied heavily on support to complete setup
Non-technical users were effectively blocked from using the product independently
This created a major gap between product potential and actual user activation.
The original interface was designed with a developer-first mindset, exposing technical settings that created unnecessary complexity for non-technical users.
The original interface was designed with a developer-first mindset, exposing technical settings that created unnecessary complexity for non-technical users.
My Role
My Role
I led the redesign of the agent creation experience as a product designer, working closely with:
an AI specialist (defining how agents are trained and behave)
a customer-facing expert who regularly helped clients set up agents
This collaboration helped ground design decisions in both technical feasibility and real user pain points.
I led the redesign of the agent creation experience as a product designer, working closely with:
an AI specialist (defining how agents are trained and behave)
a customer-facing expert who regularly helped clients set up agents
This collaboration helped ground design decisions in both technical feasibility and real user pain points.
Process: Understanding the problem
Process: Understanding the problem
I started by analyzing the existing user flow and identifying friction points:
Reviewed the current setup experience end-to-end
Analyzed where users were dropping off (e.g., during trial without creating agents)
Used indirect signals such as support requests and onboarding friction
This helped build a clear picture: users were not failing because of lack of need — they were blocked by complexity.
I started by analyzing the existing user flow and identifying friction points:
Reviewed the current setup experience end-to-end
Analyzed where users were dropping off (e.g., during trial without creating agents)
Used indirect signals such as support requests and onboarding friction
This helped build a clear picture: users were not failing because of lack of need — they were blocked by complexity.
No technical knowledge required — users can configure their agent by describing its role, tone, and responsibilities in plain language.
No technical knowledge required — users can configure their agent by describing its role, tone, and responsibilities in plain language.
Process: Research & references
Process: Research & references
Users don’t want to “configure AI” — they want to describe what they need in plain language.
Users don’t want to “configure AI” — they want to describe what they need in plain language.
To define a better approach, I explored:
How similar tools structure AI agent configuration
Patterns in no-code / low-code products
Interaction models in tools like ChatGPT, where behavior is defined via natural language instead of technical setup
To define a better approach, I explored:
How similar tools structure AI agent configuration
Patterns in no-code / low-code products
Interaction models in tools like ChatGPT, where behavior is defined via natural language instead of technical setup
Process: Iteration & design decisions
Process: Iteration & design decisions
1. From code → natural language
1. From code → natural language
Replaced Python-based configuration with structured text fields:
Identity & role
Speech style & tone
Primary task
Welcome message
This allowed users to define agent behavior in a familiar, human way.
Replaced Python-based configuration with structured text fields:
Identity & role
Speech style & tone
Primary task
Welcome message
This allowed users to define agent behavior in a familiar, human way.
2. Clear separation of concepts
2. Clear separation of concepts
Reorganized the experience into 4 distinct tabs:
Instructions — agent behavior and personality
Knowledge Base — sources of information (files, links, Q&A, etc.)
Channels — where the agent is deployed
Conversation Results — structured data extracted from conversations
This reduced cognitive load by grouping related concepts into clear mental models.
Reorganized the experience into 4 distinct tabs:
Instructions — agent behavior and personality
Knowledge Base — sources of information (files, links, Q&A, etc.)
Channels — where the agent is deployed
Conversation Results — structured data extracted from conversations
This reduced cognitive load by grouping related concepts into clear mental models.
The new information architecture grouped agent configuration into four logical sections, reducing cognitive load and helping users focus on one aspect of setup at a time.
The new information architecture grouped agent configuration into four logical sections, reducing cognitive load and helping users focus on one aspect of setup at a time.
3. Knowledge Base simplification
3. Knowledge Base simplification
Introduced a flexible system where users can:
Upload files
Add website links
Create Q&A collections
Use content like YouTube videos
This replaced more abstract and technical data configuration.
Introduced a flexible system where users can:
Upload files
Add website links
Create Q&A collections
Use content like YouTube videos
This replaced more abstract and technical data configuration.
4. Embedded channels configuration
4. Embedded channels configuration
Moved channel setup (Telegram, WhatsApp, web widget) directly into the agent flow, making the experience more cohesive and reducing context switching.
Moved channel setup (Telegram, WhatsApp, web widget) directly into the agent flow, making the experience more cohesive and reducing context switching.
5. Playground for instant feedback
5. Playground for instant feedback
Added a Playground where users can test their agent in real time.
To avoid friction:
Agents are created immediately after naming (no manual save required)
Users can start interacting with the agent right away
Added a Playground where users can test their agent in real time.
To avoid friction:
Agents are created immediately after naming (no manual save required)
Users can start interacting with the agent right away
Instant feedback makes it easier to validate agent behavior, refine instructions, and identify issues before deployment.
Instant feedback makes it easier to validate agent behavior, refine instructions, and identify issues before deployment.
6. Onboarding & guidance
6. Onboarding & guidance
Added onboarding support:
Guided entry points
Links to documentation
Clearer starting points for new users
Added onboarding support:
Guided entry points
Links to documentation
Clearer starting points for new users
Solution
Solution
The final solution transformed the agent setup experience from a technical configuration tool into a structured, human-readable workflow.
The final solution transformed the agent setup experience from a technical configuration tool into a structured, human-readable workflow.
Users can now:
Describe agent behavior in natural language
Add knowledge sources without technical setup
Connect channels within the same flow
Test the agent instantly
The system shifted from “configuring AI” to “describing intent”.
Users can now:
Describe agent behavior in natural language
Add knowledge sources without technical setup
Connect channels within the same flow
Test the agent instantly
The system shifted from “configuring AI” to “describing intent”.
Impact
Impact
The redesign significantly lowered the entry barrier and improved product accessibility.
The redesign significantly lowered the entry barrier and improved product accessibility.
Reduced dependency on support for agent setup
Enabled non-technical users to create agents independently
Increased activation: more users create agents during trial
Existing clients started building agents without assistance
Reduced dependency on support for agent setup
Enabled non-technical users to create agents independently
Increased activation: more users create agents during trial
Existing clients started building agents without assistance
