At Atomic Object, we thrive on tackling client challenges across many industries. Rapidly getting up to speed is crucial not only to reduce costs but to ensure we deliver final products efficiently. While AI has come a long way in the past two years, there is still a long way to go. While AI can’t replace all user research, at this point, it can significantly boost our domain knowledge and streamline various aspects of our workflow. Today, let’s explore a few examples where AI tools continues to help me as a designer and project lead.
Note: we use Large Language Model (LLM) models that do not train on our clients’ data.
AI as a Rapid Knowledge Booster
One of the challenges we face at Atomic Object is quickly understanding new domains, especially in smaller engagements where a larger portion of the budget is spent getting up to speed. This is where AI comes in. Recently, we used an AI model to analyze data from our kickoff sessions and notes. This approach had a big impact because it allowed us to maximize the value of every hour spent on a project.
We started by feeding the AI model information gathered during kickoff sessions, client interviews, and industry-specific documentation. Then we created a custom GPT or session tailored to the project. This personalized AI resource enabled us to dive deep into the client’s industry, leading to faster delivery of high-quality final products. The ability to swiftly gain expertise saves time and enhances the quality of our solutions.
Another significant benefit of using AI this way is the reduced demand on our clients’ time. Typically, clients must spend a notable amount of time transferring knowledge to our team. With AI, we can start to minimize these demands, allowing our clients to focus on their core responsibilities while we efficiently gather and assimilate the necessary information. While this will likely not go to zero we have noticed an impact. This not only better utilizes everyone’s time but also enhances client satisfaction.
AI as My Rubber Duck
The concept of using a “rubber duck” to debug code is well-known among developers but was new to me before joining Atomic. It’s stuck with me since. Explaining a problem out loud often leads to breakthroughs, as the process of verbalizing the issue can reveal insights. I’ve adopted a similar approach, but instead of a rubber duck, I use an LLM. Though not identical to the rubber duck concept, this method includes additional benefits. Those may includes uncovering unknowns and providing relevant information beyond just verbalizing to solve debugging issues.
Clarity
Interacting with an LLM allows me to clarify and improve my explanations, ensuring my ideas are well-formed and more coherent. After providing the relevant context to the LLM, I work with it as a co-worker to break down the problem further. This involves asking clarifying or explorative questions, which facilitates a back-and-forth exchange. During these interactions, I often need to correct the LLM and set the expectation around where I’m looking for help, even when I’m not entirely sure myself. This iterative process helps narrow down the focus and refine my understanding of the problem.
Efficiency
Using the LLM as a first line of inquiry also saves time when collaborating with other Atoms. When faced with a complex problem, I now spend more time one-on-one with the LLM before bringing my more well-baked thoughts and opinions to my co-workers. This means that by the time I involve my colleagues, I’ve already worked through many of the initial uncertainties and have a clearer, more structured approach to the problem.
This method not only enhances my problem-solving process but also improves the efficiency of team collaborations. By leveraging the LLM to tackle preliminary issues, we can dive straight into more productive discussions, making our problem-solving sessions more effective and focused. Overall, the LLM serves as a valuable tool in my workflow, functioning as a knowledgeable, patient and sometimes quirky partner in navigating complex challenges.
ChatPRD: AI-Assisted Product Requirement Documents
Starting with a blank page when creating a product requirement document (PRD) can be daunting especially early on in a project when constraints aren’t well defined. That’s where ChatPRD, an AI chat on OpenAI’s ChatGPT, comes in. Through a series of conversations with the LLM, I was able to draft a comprehensive PRD. Although it wasn’t perfect, it provided a solid foundation.
ChatPRD could be recreated or customized, but the base version gave an excellent starting point to get my team’s thoughts on paper and organized. I transferred the AI-generated content to Google Docs for further editing and collaboration. This approach kickstarted the process and provided a structured template, making the task much more manageable. This was a 75-80% savings for the two times I’ve used it. This isn’t a massive task for us, but saving any time and applying it better elsewhere drives more value for our clients.
Conclusion
Integrating AI tools into our workflow at Atomic Object has continually added value, albeit not as rapidly or broadly as we initially anticipated. While these tools have not replaced essential aspects of our work (yet), they have, modestly, enhanced productivity and knowledge acquisition. I encourage others to explore how AI tools can complement their workflows and improve efficiency. By leveraging AI in the right places, we can continue delivering higher-quality solutions, ultimately benefiting both our teams and our clients.