ARC: Reinventing Document Review with AI

A Broken Promise by AI

Artificial intelligence, touted as a revolution capable of transforming our daily lives, was supposed to free us from tedious tasks and provide us with more time for enriching activities. The idea was that AI would generate drafts of emails or Slack messages, allowing us to focus on more meaningful work. Tech giants even predicted that work would become an optional leisure activity in the near future.

However, reality took a different turn. The integration of AI into the professional world has led to waves of layoffs, and remaining employees have had to take on more responsibilities. Designers, for example, have found themselves managing the entire product lifecycle, a task that has reportedly increased fivefold according to Meta. Workplace well-being has been relegated to the background, and the frantic pace of work has become the norm, fueled by the constant emergence of new tools and optimization processes.

This increased pressure has had repercussions on workers' mental health, with a sharp rise in cases of stress and anxiety. Job insecurity and workload overload have made fear and anxiety constant companions in the professional environment. Even though AI has reduced some repetitive tasks, the need to process a massive amount of information to make decisions has led to growing mental fatigue and burnout.

The Document Review Overload

As a designer, I have noticed an increase in the number of AI-generated documents to review. These documents, often dry, require careful reading and feedback to make informed decisions about products and business strategies.

Review requests are flooding in, and while tools like NotebookLM can help summarize complex documents, it is often essential to read them in detail to grasp all the nuances. This task, while necessary, can become tedious, especially when spending long hours in front of a screen.

To mix things up, I sometimes transfer these documents to an iPad to read outside, but that doesn't change the fact that I'm still glued to a screen. Printing the documents and annotating them by hand is an alternative, but it complicates reintegrating the notes into a usable format.

An Innovative Approach

That's when the idea struck me: why not listen to these documents instead of reading them? And even better, why not provide my feedback to a voice assistant that would record it directly into a Google Doc? This would allow me to review documents while engaging in other activities, like walking or cooking.

The application I envisioned needed to be able to read a Google Doc aloud, pause on voice command, and take dictated comments. I wanted to work without being tethered to my desk or a screen, and have my hands free to accomplish other tasks.

Although applications like Speechify and Wisperflow exist, they did not allow for conversational commenting on a Google Doc. NotebookLM was limited to summaries without offering the ability to provide detailed feedback. During the development of this application, Google announced voice reading and editing features for Google Docs, but without the ability to comment vocally, leaving a gap to fill.

Developing ARC with Claude Design and AI Studio

I then set out to create ARC (Audio Review Companion) using Claude Design and AI Studio. This project was an opportunity to experiment with multimodal interfaces, inspired by my experience with Thia, an open-source ideation board partner.

The question was where to start: the underlying model or the flows? I chose to develop them in parallel. After a few brainstorming sessions, I crafted a design concept and a brief specification document, then built the model and flows simultaneously. The application was developed in Google AI Studio, while the design and layouts were created in Claude Design.

By using Gemini Flash Live in AI Studio, I was able to integrate the design generated by Claude, including specific screenshots and animations, to create a functional application.

ARC: A Liberating Application

Using Claude Design was a first for me on a project of this scale. While it was effective for generating visuals, inconsistencies remained without clear guidelines, and the token limits were restrictive.

After several iterations, ARC was ready for testing. The application allows for smooth conversational interaction, as if talking to an assistant capable of taking notes. Gemini Flash Live was configured to prioritize speed, ensuring responsive performance.

The advantage of a single model for reading and navigation is that it offers an experience akin to conversing with someone who understands the document and acts as a guide. Voice commands allow for easy navigation within the document, such as asking ARC to read a specific section or jump to a particular part of the text.

Balancing Work and Personal Life

A recent study by HBR revealed that AI tends to intensify work rather than reduce it, with people seeking to accomplish more tasks outside of office hours. While ARC helps me save time away from the screen, it is crucial not to let these tools encroach on personal time.

I have learned to use ARC in a way that optimizes my work time without compromising my relaxation moments. For instance, reviewing a document while walking can be a welcome break, but working during a walk intended for relaxation encroaches on my personal time. I designed ARC to favor the former scenario, allowing me to manage my work on my own terms while preserving my leisure moments.

Conclusion

In a constantly evolving world, we have the choice to be overwhelmed by systems or to use them to regain control over how we work. By designing tools like ARC, we can shape our work environment to fit our needs.

Try ARC

I invite you to test the ARC application, designed to be used during work hours and not on your personal time. Version 1 of ARC was developed to meet a specific need, but I am confident that the UX and engineering community can further improve it. The repository is available open source on GitHub, and you can customize it on Google AI Studio. If you have ideas to enrich the voice experience or improve hands-free navigation, feel free to contribute or start a discussion in the repository.