During a recent episode of the "Getting to Zero" podcast, Rory Madden sat down with Anthony Maggio, VP of Product Management at Airtable, to discuss the challenges and opportunities of incorporating AI into product development processes.
A paradigm shift
Historically, software development has been deterministic: teams would identify a problem, brainstorm solutions, prototype, and then build toward a defined outcome. While we get new frameworks and device features the underlying technology has not shifted dramatically so our experience and intuition about what is possible is pretty good.
The introduction of large language models (LLMs) disrupts this process entirely because we both don’t fully understand what modern LLM’s are capable of and the outcome can vary significantly depending on the input, the prompt, and the model's ability to reason and respond.
"Building compelling products with AI requires an almost entirely different product development process."
Encouraging Teams to Think Bigger
Initially, teams at Airtable used AI to solve simplistic problems or automate low-value tasks.
We went in thinking that we can get the LLM to consistently output really great results if we give it a lot of context about a specific industry and a specific set of applications and the shape of those applications.
But we withheld ourselves from actually thinking bigger and saying What if we didn't need to do that? What if we just use the corpus of knowledge that these LLMs are already trained on and allow them to think creatively about the needs of an application for anyone and for any industry and for any role.
It was not easy for people to shift their mental models so Anthony started a new approach when talking with customers. He would ask “What if we had a magic wand what would the solution look like?”
While this helped to get a really deep understanding of the customers’ true motivations there was still a problem with helping the internal teams think about how they could use AI to deliver on these ideas.
Space to explore
While AI opens up new possibilities, it also brings complexity. Teams tasked with building AI-powered features must now juggle traditional software engineering knowledge with AI-specific skills, such as prompt engineering and model training.
Because the technology was so new, nobody on the team really knew what the latest LLMs were capable of delivering. Rather than pushing ahead with tight deadlines and pushing the teams to deliver, Airtable went the other way.
We actually asked the team to just stop coding. We said don't code anything for the next two weeks. Go exclusively focus on prompt engineering, on taking specific use cases, and iterating on the responses that we can get out of the AI models.
By giving the teams time to explore it helped to break the old mindset of what was possible and the ideas that the team started to come up with were much more in line with the ambitions of the customers.
Higher quality in shorter time
You would, conceptually think, working with AI will take a really long time to develop. But a lot of what we've found is that the process of building with LLMs and with AI can actually be much faster than a traditional product development process.
One of the most successful examples of this approach is Airtable’s Co-builder. Customers had said that their employees were not always tech savvy. They knew the challenge that they faced and what they were looking for but they didn’t know how to design and build the applications.
Airtable started prototyping the Co-builder product in May and it was ready for release in July. It allows new users to describe a business process or goal, and the AI then creates a customized application, including both the underlying data model and the UI, tailored to their needs.
It truly feels like a magical experience. You're able to design and deliver a custom application for any person in any role, in any function, in any industry, in any language, right on the fly in a matter of seconds.
Human-in-the-Loop AI: Striking the Right Balance
While AI is immensely powerful, it’s not able automate jobs and replace teams (for now). Anthony believes the best way to use the tools is using an approach that he calls "human-in-the-loop" AI, where AI augments human work rather than replaces it entirely.
For example, Airtable uses AI to enhance marketing processes by analyzing past campaign data to suggest improvements for new campaigns. The AI does the heavy lifting by identifying patterns and generating insights, but human marketers still make the final decisions, ensuring that the results are relevant and actionable.
This hybrid approach allows AI to deliver value without introducing the risk of fully automating complex decisions that require human judgment.
Looking Ahead: Scaling AI Across the Organization
While Airtable has successfully integrated AI into some of its products, the journey is far from over. Only two out of Airtable’s fourteen product teams are focused on adopting AI for their products.
Scaling this knowledge and encouraging all teams to think bigger about AI remains a challenge. "Change management is hard," Anthony admits. "It’s like pushing a rock up a hill." Many teams are still adjusting to the new way of thinking, and scaling AI across a large organization will take time, education, and a continued emphasis on experimentation.
How can you think bigger?
The key takeaway from Airtable’s experience is that building AI products is not just about applying new technology—it’s about rethinking the entire product development process. By encouraging teams to think bigger, dedicating time to prototyping, and balancing human expertise with AI capabilities, companies can unlock the full potential of AI and create products that truly transform industries.