What can generative AI offer to The Atlantic?
We had some ideas. A hackathon allowed us to shape them in a matter of days.
By Sebastian Podesta
Generative artificial intelligence and Large Language Models (LLM) seem to be everywhere. Across many industries, including journalism, people are studying the possible implications of these technologies, and each organization is approaching this challenge in its own way — some have been enthusiastic, others cautious, and some have been a mix of both.
At The Atlantic, we wanted to explore the possibilities of AI, too, and reflect on its risks. To do that, we iterated on a process we had used before: a week-long hackathon.
Hackathons at The Atlantic help our teams in several ways. The rapid experimentation enables us to think beyond our immediate priorities. Participants can bring their passions, interests, and curiosity. They can try out a new technology and learn how to use it without fear of failure. They can collaborate with people across the organization and stretch beyond their core disciplines. For example, this is the perfect setting for an engineer with strong design skills to flex muscles they don’t often get to use.
In the past, this structure has also helped us address some challenges we face as an organization. Hackathons create a space to quickly learn what’s possible and feasible in an emerging area or technology.
Hackathons also give us a new perspective on how to better organize ourselves by allowing us to break from our standard workflows.
We focused a Generative AI Hackathon on applying recent advancements to Atlantic surfaces and problem areas. The Atlantic has three core values: Spirit of Generosity, Sense of Belonging, and Force of Ideas. In organizing this hackathon, we placed them at the heart of this initiative. We followed these steps:
1) The invitation: The greater the participation, the more likely we were to create a high volume of quality work. We sent out an open invitation across the organization for anyone interested in joining. No one was required to attend, but all were welcome. We were very explicit about trade-offs and multi-tasking. Participants had to carefully assess how to contribute without putting key day-to-day initiatives at risk.
By not requiring participation and allowing people to experiment and learn, we wanted to amplify our Spirit of Generosity. We sought to embody a Sense of Belonging by accepting anybody within the organization.
2) The proposals: Following our Force of Ideas, we wanted a broad set of suggestions from all levels of the organization. We had an open call for anyone, even those not participating in the hackathon, to submit ideas.
An executive panel with representatives from across the organization reviewed the proposals. They had to ensure no ethical, legal, or reputational issues were at stake with a particular application of AI. The panel narrowed the total list of ideas down to seven projects with outcomes that spoke to a range of goals, from editorial and consumer strategy to advertising and internal tooling.
3) The teams: Ideas defined, we then asked participants to select their first, second, and third choices to inform how we staffed each team. Every group would have representation from the largest product and technology areas: product, data science, engineering, and design. Some teams also included members from audience research, editorial, consumer strategy and growth, and sales.
4) The schedule: We set aside five days, during normal working hours, to ideate, build, and present each team’s outcome. On day 1, groups met in the morning to form a plan, brainstorm, and start developing. On days 2–4, they focused on development, using daily standups and Slack channels to coordinate their activity.
5) The talent show: On day 5, each team had 10 minutes to present their proposal and walk through a working prototype. After the presentations, the executive panel met again to select the projects that would be put on a path for production release.
It was a success.
About 60 people signed up to participate. We received almost 70 ideas, from which we selected seven for the teams to take on. We even launched one during the hackathon: an internal tool that helps the editorial team more easily search The Atlantic’s archives for articles related to current events.
Shortly after the hackathon, we piloted another one. We evaluated the technology to offer AI-powered audio versions of some Atlantic stories, helping make more of our journalism accessible to readers in multiple formats. Our teams are also working on two other ideas: one focused on advertising and another on subscriber retention.
There is still a lot to think about when it comes to AI. Some applications are riskier than others. We see a difference, for example, between internal tools that can ease our team’s workflows and projects that could compromise our journalistic integrity or affect our readers’ experience.
Experimentation, failure, learning, and trust are essential for any team or business to grow. Making time for it and requiring that something must be released at the end can help get ideas going. Time constraints force people to come up with simple solutions that work. The Generative AI hackathon allowed participants to do that in a matter of days. It showed us that we can get great, straightforward ideas built quickly.
As an Agile coach, I regularly think of the Agile Manifesto. During the hackathon, one principle came to mind the most:
Build projects around motivated individuals. Give them the environment and support they need, and trust them to get the job done.
If you use a hackathon, give your teams a focused prompt and the space to put forth their ideas and apply their passion. You’ll be surprised by how quickly and creatively they solve tough problems.
- — — -
Aldana Vales contributed to this post.
Additional acknowledgments: Jefferson Rabb.
