Transforming Collaboration in Digital Product Teams with AI

By Chris Alden, Technology Partner at Work & Co

So much of the conversation around AI centers on automation—taking tasks away from people and giving them over to software. But with this mindset we’re overlooking the truly transformative potential of this technology — the potential to create new sources of dialogue—with broader fields of inspiration, with the audience we’re serving, and with each other.

I’ve been working hands-on with AI for more than a decade now. In that time I have seen that for all the deeply technical work it entails, harnessing AI in product development is a critically human process, and one that requires thinking less about execution and more about collaboration. This north star guides how we work with AI, work on AI, and consider AI’s role in product development at Work & Co.

I’m sharing some of the ways this manifests within our team in hopes that it will inspire you and your teams to collaborate more effectively with AI. Here are four tips uncovered from our experience in the field:

Ground your practice in observed human needs

This has been really critical to the success of both our internal AI tooling and the AI products we build for clients.

One of our first explorations using Generative AI to enhance our creative processes was the development of a tool that allows our designers to use natural language prompts to generate images on a Figma canvas. We carefully designed this plugin to optimize the generation of multiple variations of an idea, to most effectively infuse the external inspiration that image generation models provide into our specific design process, in which developing a wide range of concepts is the seed for further refinement. AI opens up a kind of raw creativity, but asking how AI can be attuned to a specific human need turns it from a mere tool into a kind of collaborator.

We applied this same mentality to our partnership with the Planned Parenthood Federation of America. Planned Parenthood got in touch with us with a broad question: how could we work together to use technology to ensure that critical sexual and reproductive services remain accessible to at-risk communities? Working directly with health educators and teens—the target user group—we learned that, when engaging with this charged topic, they need a source of information they can trust and that will not judge them for the questions they ask. Grounding our work in this need led us to the solution of an AI-powered chatbot trained to be friendly, factual and non-judgmental. The technology opened up a world of possibilities, listening to the needs of the audience is what made it impactful.

Channel productivity with testable hypotheses

To harness AI’s productive power reliably we need to apply a degree of scientific rigor. That means defining a hypothesis about how a system will respond to different inputs and how you will evaluate your assumptions. This has forced a change in our process, moving us from a traditional Software Development Lifecycle of planning, design, implementation, deployment, and analysis to one of ongoing collaboration and evaluation that starts with inspiration and then moves into observation, forming hypotheses, experimentation, measurement, refinement and cross-pollination.

Working with AI means constantly forming hypotheses about how a non-human intelligence will behave and, quite often, being proven wrong. This forces us to clearly define our thinking across technical and non-technical teams, further breaking down silos of execution.

Optimize for ongoing experimentation and refinement

Working with AI has forced us to get comfortable with probabilistic thinking. We see in the generative AI products that we are building with our clients, that you can no longer test a flow, know with 100% certainty how it will behave, and feel confident that we are done. Rather than rely on a simple “definition of done” we engineer methods of continuous and automatic evaluation against thresholds of success. Building up these processes are key to taking these technologies from initial explorations into robust solutions ready to solve problems in production.

Apply Human Craft to Sharpen Focus

AI models make it incredibly easy to experiment if your process flexes to accommodate empirical thinking, but taking the insights you gain and refining them into something useful means leaning back very deeply into our human expertise.

An application of this technology we have invested in sits directly at that intersection of design and engineering. In our work we have seen again and again how quickly the success of design ideas depends on the realities of the data that will power them. This realization led us to build CodeSail, an AI solution for data-driven product development.

In CodeSail, we embed our collective expertise and sense of craft as software engineers into AI agents that help us accelerate the process of building data enablement layers. These focused agents contribute to a persistent project that we can collectively refine across teams, so that it is more than just an extra shot of productivity, but a tool that helps accelerate delivery of not only ideas but fully functional applications.

Moving beyond productive machines

We have a powerful opportunity to help shape this moment. As we adapt to take advantage of AI, shifting away from this focus on automation can also free us from ways of working optimized solely for output.

When we work in ways that look to magnify our human senses of intuition, experimentation, and craft to guide machine processes that extend so far beyond ourselves, we can actually make the way we work more human.