The unexpected story of Coda AI

Why Coda made a big bet on artificial intelligence, and what product teams can learn from our journey.

David Kossnick

Product Lead for AI at Coda

The unexpected story of Coda AI

By David Kossnick

Share this article

Share on TwitterShare on LinkedInShare on Facebook
Blog > AI · 7 min read

We recently launched the latest version of Coda AI, the new work assistant.

We call it a “work assistant" because it can do so much more than generate content. Coda AI drafts and edits, sure, but it also consolidates information across docs, expands and organizes data, turns information into insights, and streamlines entire workflows from start to finish. We've seen the use cases explored by early testers clearly indicate the impact Coda AI has on an individual's or a team's work.
We’re still in Beta, but the feedback we’re getting from users is amazing: saving hours of busywork, improving collaboration, and making a bigger impact on their teams.
Since launching, I’ve had quite a few people ask me, "How did Coda come to make such a big bet on AI?" The answer probably isn't the story you think. Coda AI didn't originate from a top-down executive decision or a singular moment. It started as an idea and a passion project. Here's how it evolved into the product it is today and what your team can learn from our journey. In June 2022, Coda released Pack Studio, which enables anyone to build a Pack (an integration to connect outside apps to Coda docs) through a web-based editor or a developer-friendly command line interface (CLI). Pack Studio quickly became a platform for community invention and collaboration. It was amazing to see people of all ages and backgrounds build and release hundreds of integrations with all sorts of tools, making it possible to connect all their workflows into one surface in Coda. I personally built and released two Packs of my own: Icons8 and QuickChart. Fast forward to a Friday morning in early November: OpenAI released the API for DALL-E, an image generation model, and one of our engineers immediately wanted to explore the potential for integrating it into Coda. Using Pack Studio, he built an OpenAI Pack that could use all of OpenAI's generative capabilities inside his Coda docs (in under 30 minutes!). When he published it, I nearly fell out of my chair! As I started playing with the OpenAI Pack, my mind raced with all the possibilities for how it could help others save time and unlock their creative potential in Coda. We wanted to start bringing these ideas to life ASAP, so I rallied a small group of Codans to make templates using the OpenAI Pack, and we released the Pack on ProductHunt a few days later.
Within a few weeks, the OpenAI Pack became our #4 most-used Coda Pack out of hundreds of integrations, despite the fact that you needed to create an OpenAI account and link your account credentials into Coda.
A few weeks later, on November 30, 2022, OpenAI released ChatGPT, and my mind was blown again. I previously used their large language model (LLM), GPT3, but it required you to write long and detailed prompts to receive a usable output. ChatGPT was much simpler. You could ask a short question, and receive impressive outputs. My four-year-old son and I created bedtime stories with ChatGPT, and as I read some of these tales I started to picture a world where LLMs could transform every tech product. I participated in Coda's quarterly Hackathon in December. Because many Codans were excited about AI, a few of us formed a Hackathon team and started experimenting. What if we took our OpenAI Pack and made it more native to Coda? How fast and responsive could we make AI? What extra capabilities could that unlock for our users? We prototyped a bunch of concepts and ended up winning the Hackathon! When our Q1 board meeting rolled around, our team presented the Hackathon results and OpenAI Pack progress for feedback. The board shared our enthusiasm and encouraged us to assemble a more dedicated AI team. The initial result: a lean team of four engineers, one designer, and one PM (me!) to start up our native AI effort. A few months later, in February, we shared our initial vision externally and released an alpha signup list. We launched the Coda AI Beta in June—just one year after the launch of the Packs Studio, where it all began. While the origin story of Coda AI may be atypical, several aspects of our approach apply to any team interested in exploring an AI product.

Invest in user feedback.

A unique aspect of Coda AI is its emergence out of validated user demand. The OpenAI Pack was extremely popular, so we knew we had potential champions before we even had an official AI team. Throughout the process, we continued to invest heavily in user research. After the Hackathon, our AI team partnered with Coda’s User Research team to gather feedback in 1:1 interviews and in-product surveys. We also offered survey respondents gift cards to record videos of themselves using the product and tell us what they liked or didn't like. No matter where you begin with AI, investing in user research throughout the process is critical to ensuring you're building something people want.

Build a small and passionate team.

One of our Hackathon's best practices is creating teams of five to seven people. Why? It allows every team to have a diverse set of skills while remaining nimble. We used the same logic when building our AI team. Another tip is that when you're building an AI team, find others who are already engaging with AI on their own. Everyone on our team was personally excited about the technology and leveraged it while building our own product: our engineers used Co-Pilot as they coded, and I used OpenAI to summarize our meeting notes (talk about meta). As a small team, our use of AI enhanced our productivity and stretched our imaginations of what we could build Coda AI to do.

Encourage bottom-up innovation.

We celebrate “maker culture" at Coda, which means we believe that great ideas can come from anywhere. It was much easier to make a bet on AI in a culture that encourages us to explore new ideas, no matter where they begin. And this isn’t just true for AI: many of our product’s most successful features have come out of Codans' passion projects at Hackathons, such as Coda Packs and reaction buttons. Take a page from Coda's book and encourage bottom-up innovation; I guarantee your team has a ton of ideas worth exploring. This is especially valuable for an area like AI, where nobody has all the answers—the entire market is learning and exploring, and new capabilities and opportunities are emerging daily. Whether you’re interested in integrating AI into your product or just trying to figure out how to leverage AI in your day-to-day work, we’ve built Coda AI with the same spirit of “great ideas can come from anywhere.” It’s not just a tool for AI enthusiasts—it's for everyone. We've made it easy to incorporate AI into templates, so you can start using AI for your idea with a single button click. And if the story of Coda AI teaches us anything, it's that you never know how one idea can transform your whole product. So, what will you create?

Related posts

Explore more stories about AI
AI at Work Challenge winners

We asked makers to build and submit templates, and the results are in.

How makers are using Coda AI

Meet your new personal assistant.

Why AI work assistants are better than AI writing assistants

Writing assistants can make content. Work assistants make progress.