About this project

What is a “third interface” for Large Language Models?

I researched and designed the prototypes above while working for IBM in the Spring of 2023. It was my first full-time job after college—helping build experimental tools with a small team of designers and engineers.

Like many, we had spent the past few months enchanted by Large Language Models. They were smart, fast, often magical.

The interfaces fascinated us.

At the time, there were two popular ways to interact with LLMs: chatbots and APIs (Application Programming Interfaces).

The chatbots were easy to use. But they were siloed from everyday tools like Slack or Google Calendar. This profoundly limited their scope—a chatbot could describe a meeting, but never schedule one for you.

Chatbots can’t work where you work.

On the other hand, APIs were powerful but required people to know how to write and read code. Most don’t. That was the game: chatbots were easy but limited, APIs were powerful but arcane.

What could a “third interface” for LLMs could look like—more nuanced than basic prompts, less complex than writing software?
Our hypothesis: shapes.

Conveyor’s draggable blocks help people interact with AI spatially and visually, instead of just with words or code. Just as chatbots are designed to answer questions, our tool encourages multi-step, multi-modal thinking.

Visual programming has been around since the 80s. Products around the world use it to make complex processes feel simple. My favorite example is Scratch, a platform that teaches kids to code.

Conveyor was launched internally at IBM in April of 2023. The response was fantastic. Our colleagues ran far and fast with the new tool. They built notetakers, meeting makers, email senders, and even chatbots. Over a hundred automations and thousands of tasks.

"In terms of performance and ease of use, it's incredible."

Senior Software Engineer, Data & AI

“It’s doing something extraordinary”

Front End Developer, Automation

96/100 Net Promoter Score

Likelihood users would recommend Conveyor to a colleague

It was the best feedback we could have hoped for: Conveyor worked well, and people enjoyed using it.

Devflow is a Slack-based coding chatbot powered by IBM watsonx and built with Conveyor. In a field test, it helped developers reduce deployment times by 13.6x.

Conveyor went on to receive the 2023 Red Dot Award for Interface Design.

In the time since, the product’s grown up. It’s moved to a new team, changed its look, changed its name.

Last I checked, it goes by watsonx Assistant Builder, and it’s making life easier in offices around the world.

Fly high, Assistant Builder. It was an honor to be a part of your journey.

Design
Andy Pham
Noah Kawash-Barnes
Gord Davison
Om Gokhale
Vera Wei

DevelopmentRohit Singhal
Connor Colwill

ClientIBM

LinksRed Dot Award Page
IBM Watsonx Code Assistant