Rethinking an AI platform through Agentive Design
People
Tools created to increase productivity
Planet
Corporate workers sharing knowledge and streamlining workflows
Profit
Increase in usability
Big picture
A well-known AI platform was in its early development stage when Bros&Co was brought in to help find its product-market fit through a complete redesign.
Over six months, we rebranded and rebuilt the solution to assist 20,000 knowledge workers with task-specific AI tools.
Hurdles
AI utilises Large Language Models (LLMs) to process and translate vast amounts of text data into understandable human language. The original system relied on a single LLM, which limited its capabilities and introduced issues like tool duplication and technical debt.
This single LLM approach allowed high-order tasks to be expedited with guidance and structure from the user’s context window. However, the limited context window, combined with a one-size-fits-all LLM solution, created a significant gap between high-order and low-order tasks.
These challenges resulted in low user satisfaction and a usability from users, with a score of just 32%.
Game plan
Our approach went beyond simple rebranding. We transformed the tool from a single LLM (OpenAI)-based solution into a multi-LLM system with an agentive framework, empowering users to select the most suitable model for each task—for instance, Mistral for coding and DALL-E for image generation. This multi-agent architecture, combined with a holistic service model, reduced duplication, minimised technical debt, and supported third-party plugins.
The win
Our initiatives drove significant growth, transforming the platform into a vibrant community and increasing usability from 32% to 85% - a 265% boost. Tool usability tripled, leading to the creation of 1,500 tools, with 100 shared publicly with 20,000 corporate knowledge workers.
The new service model also enabled valuable partnerships, such as a Copilot plugin for Microsoft Copilot users.
“It's incredible to see collaboration and innovation in action within an agile environment at this scale. I’m excited to see this ripple through our organisation.”
- Innovation & Engagement Leadership
Hurdles
AI utilises Large Language Models (LLMs) to process and translate vast amounts of text data into understandable human language. The original system relied on a single LLM, which limited its capabilities and introduced issues like tool duplication and technical debt.
This single LLM approach allowed high-order tasks to be expedited with guidance and structure from the user’s context window. However, the limited context window, combined with a one-size-fits-all LLM solution, created a significant gap between high-order and low-order tasks.
These challenges resulted in low user satisfaction and a usability from users, with a score of just 32%.
Game plan
Our approach went beyond simple rebranding. We transformed the tool from a single LLM (OpenAI)-based solution into a multi-LLM system with an agentive framework, empowering users to select the most suitable model for each task—for instance, Mistral for coding and DALL-E for image generation. This multi-agent architecture, combined with a holistic service model, reduced duplication, minimised technical debt, and supported third-party plugins.
The win
Our initiatives drove significant growth, transforming the platform into a vibrant community and increasing usability from 32% to 85% - a 265% boost. Tool usability tripled, leading to the creation of 1,500 tools, with 100 shared publicly with 20,000 corporate knowledge workers.
The new service model also enabled valuable partnerships, such as a Copilot plugin for Microsoft Copilot users.
