Designing an Autonomous AI Sales Platform for Online Creators
Overview
Let AI sell while you sleep - transforming a manual proof-of-concept into a scalable, self-service platform
Constant Closer is an AI-powered email marketing platform that provides online creators with a fully autonomous AI sales agent to handle their entire sales cycle.
When I joined the team, they had a validated proof-of-concept generating real revenue, but everything was managed manually—creating a clear ceiling on growth.
My role was to transform this validated AI concept into a scalable, user-friendly product that could serve customers without constant manual intervention.
The AI product context
Unlike traditional digital products, AI products require technical validation first before design can scale them.
The scaling roadblock
Founders personally onboarded customers via Slack, engineers configured AI systems manually, and everyone scrambled when human intervention was needed. This worked for a dozen customers but created a clear ceiling on growth.
Design's strategic entry point
They had already proven their autonomous agents could sell via email. Now they needed design to push the product to the next level of scalability and usability without losing the AI's effectiveness.
My starting point: Understanding the bottlenecks
I spent time with the CEO and engineering team to see firsthand what worked, what didn't, and where smart design could make the biggest difference in this autonomous AI system.
Phase 1: Getting to self-service
Solving the initial bottlenecks that were blocking growth
Mapping the bottlenecks
Time-intensive manual onboarding
Every customer required personal setup through Slack conversations, preventing the team from scaling beyond a handful of users.
Zero visibility into AI activities
Customers had no way to see what their AI agents were doing, creating anxiety about trusting an LLM with sales conversations.
The information overload problem
Their solution at the time was dumping all agent activities into Slack channels via bots. This flooded client channels with a mix of auto-generated logs and human conversations, making it impossible to track what mattered.
Customers couldn't distinguish important updates from routine messages, defeating the purpose of visibility.
The first solutions
Self-service onboarding flow
Designed a guided setup process that replaced manual Slack configurations, allowing users to configure their AI agent independently while maintaining the same level of customization.
Organized activity feed
Created a chronological activity feed organized by intention (Engage, Follow-up, Support, Conversion), giving users clear visibility into their AI agent's actions without information overload.
Impact: Fully self-service platform
These features launched successfully, making the platform completely self-service and removing the manual bottlenecks that were limiting growth.
Phase 2: Rethinking the AI architecture
Advocating for user-centered complexity management
The multi-agent problem
After achieving self-service, we regrouped to discuss the long-term agentic AI architecture.
Initially, the CEO and developers wanted to frame the product like other AI agent builders—users would create multiple agents for different tasks, manually setting up system prompts, external memories, RAG sources, and tools for each one.
The user reality check
I pushed back on this approach. Our users aren't developers or SaaS founders familiar with technical concepts. They're online creators who want results, not complexity. Asking them to understand and configure multiple agents with different tools and memories would be too much cognitive overhead.
The super agent approach
I proposed a fundamentally different approach: one super agent created during onboarding that handles everything.
Instead of managing multiple specialized agents, users would simply ask their single agent to perform different tasks. This eliminates the complexity of agent management while maintaining the same functional capabilities.
Strategic alignment achieved
The team agreed this approach was both user-friendly and technically feasible with minimal changes to our existing infrastructure. This decision became the foundation for our next design phase.
Phase 3: Designing the workflow system
From concept to interface - abstracting AI complexity for creators
Designing new workflows
Based on the super agent approach, I started designing the creation of new workflows.
The UX became laser-focused: help people accomplish tasks with minimal setup, without forcing them to understand complex AI configurations.
Technical collaboration
I explored various workflow creation flows and gathered feedback from developers about technical considerations and constraints. This iterative process helped balance user simplicity with system capabilities.
The final solution
After several iterations incorporating technical feedback, I created the final workflow design solution. The interface abstracts away the complexity of agent configuration while giving users the power to create sophisticated automated sales processes.
Implementation phase
The designs are currently waiting to be implemented in the testing environment, representing the next major evolution of the Constant Closer platform.
Phase 4: Direct code contribution
Design through development - bridging the gap between vision and execution
Learning to code for direct impact
To have more direct impact on the product UX, I started using GitHub Copilot to do front-end coding.
This gave me the freedom to shape the product experience directly at the code level rather than just through handoffs.
Hands-on improvements
I've been modifying the front-end code using native Bootstrap components and custom SCSS properties. For example, I simplified and polished the side navigation bar, making immediate improvements to the user experience while the larger workflow system is being developed.
Design-dev hybrid approach
This direct coding capability allows me to implement UX improvements faster and more precisely, bridging the gap between design vision and technical execution.
Key learnings
What I learned about strategic AI product design
Advocating for user-centered complexity
The biggest lesson was learning to advocate for hiding complexity from users. While the technical team naturally thought in terms of exposed AI capabilities, I had to consistently push for abstractions that matched our users' mental models and capabilities.
The power of product philosophy
The shift from multi-agent to super agent wasn't just a UX decision—it was a fundamental product philosophy that affected our entire technical approach. Good design thinking can reshape not just interfaces but entire product architectures.
Technical skills amplify design impact
Learning to code has significantly increased my design impact. Being able to directly implement improvements rather than waiting for development cycles has made me a more effective designer and collaborator.
AI product design is still evolving
Working on an AI product has shown me how rapidly this space is evolving. The patterns and best practices for AI UX are still being established, creating opportunities for designers to help define the standards.
Looking forward
What's next for the platform and my continued growth in AI product design
The workflow creation system represents a major evolution in how users interact with AI agents. Instead of technical configuration, users will have a streamlined interface for creating sophisticated automated sales processes.
As the platform grows, I expect to keep refining the balance between AI capability and user simplicity. The super agent approach provides a foundation that can evolve with advancing LLM capabilities while maintaining ease of use.
This project has transformed my approach to AI product design. I've learned to think beyond interfaces to product architecture, advocate for user-centered complexity management, and contribute directly through code.
To be continued...
This case study represents an ongoing journey in AI product design. As Constant Closer continues to evolve, so does my understanding of how to create effective interfaces for autonomous AI systems that truly serve their users.
My takeaway
• AI products require a different design approach. Start with technical validation, then layer on user-centered design to create scalable, intuitive experiences.
• Advocate for hiding complexity from users. The most powerful AI capabilities should feel effortless to use, not technically intimidating.
• Learning to code amplifies design impact. Direct implementation capabilities allow for faster iteration and more precise execution of design vision.