At Atomic, we’ve been exploring the fast-evolving world of AI tooling, trying to make sense of what’s useful, how best to use it, and where the leverage truly lies in our daily work. With new tools and practices emerging almost hourly, it can feel overwhelming to keep up. But every so often, something sticks and becomes truly useful.
One such tool for me is ChatPRD, a custom GPT that helps manage and document software projects. It’s changed how I kick off projects, organize thinking, and share understanding with clients and teammates.
Revisiting ChatPRD with Fresh Eyes
When custom GPTs first launched about a year ago, I came across ChatPRD and was intrigued. At the time, it felt promising, but it didn’t click with my workflow. Looking back, I probably didn’t understand how to use it effectively.
Recently, I decided to give it another try on a new client project. Since then, my AI-related skills have matured—and this time, it stuck. I was able to tap into much more of the tool’s potential. While I haven’t tried the standalone pro version yet, I’ve been making great use of the free custom GPT version inside ChatGPT.
Turning ChatPRD into My Project Manager Assistant
A few ways that I found using ChatPRD very helpful include dictating ideas to a note-taking app and then transferring that information into the tool. I added many pages of thoughts about the product, who the users are, what they care about, key steps in their workflow, and important moments in the experience. I also noted what we need to be careful about and fed all of that information directly into ChatPRD.
Additionally, I gave it initial meeting notes and product details, which helped populate the tool with useful context and allowed it to begin driving the creation of the documents I mentioned earlier.
It’s important to note that ChatPRD isn’t building the application for you. Instead, it serves as a partner in the process—helping you think through challenges, organize content, and remember different states and descriptions you’ve provided. It supports formatting and structure, questions assumptions, and helps refine workflows.
ChatPRD has become my assistant during project kickoff. It helps me quickly collect, structure, and maintain key project knowledge in a way that scales as the project evolves. I now create a dedicated instance of ChatPRD for each project, essentially turning it into a single-threaded knowledge base.
The current limitation is collaboration. Since GPT chats are not shared, only I have access to the active ChatPRD instance. However, by working in structured documents, I can share results with clients and collaborators more easily.
My AI-Generated Documentation System
Here’s how I use ChatPRD to generate a shared vision and structured understanding of a project:
- Trailhead (Landing Page)
A central index that links to all other project documents. It’s a navigational tool, useful for LLMs and agent-based tools like Cursor. - Product Requirements Document (PRD)
A high-level overview of the project’s purpose, goals, and scope. - Proto-Personas
Outlines of potential users, their needs, and behaviors. - System Structure
A conceptual map of entities and their relationships—how parts of the system interact. - Roles & Permissions
A document detailing what each role can access or modify in the system. - Allowed Actions Matrix
A table format view of role-based actions—more granular and easier to scan. - Core Data Model
A draft technical schema showing entities, fields, and relationships. - User Journeys
Each journey is documented in a dedicated file and framed using the Journey–Step–Story model. This approach helps detail how each user type engages with the system, what steps they take, and the specific stories that illustrate their experience and goals.
Each of these documents is generated and iterated on collaboratively with ChatPRD. We diverge and converge quickly, prototyping ideas and clarifying assumptions. We also have plans to add in Visual/UX documentation along with high-level Engineering Specs to round out the picture and further align stakeholders and the team.
Exporting & Sharing the Work
One of the great parts of this workflow is that I can export each document in Markdown format. I can export by asking the tool to output a section or a full final version of the document in markdown. This is easy to use in the code repo or to import into a Google Doc by pasting as markdown. These files are easy to share with clients (via Google Drive) or embed in GitHub repos for engineers and LLMs to reference during implementation.
Keeping Things Synchronized
Like any documentation system, keeping things updated can be tricky. ChatPRD sometimes forgets or drifts from previous responses. It generally does a good job of remembering suggested changes, but it doesn’t always output the content in the same way, so providing additional context has been helpful. To manage this, I’ve developed a rhythm:
- Copy/paste the current version of a section or document into the prompt.
- Ask for specific changes or iterations.
- Let ChatPRD return a revised version with those edits.
This method works surprisingly well and allows me to batch-update terminology or workflows across multiple documents with a single prompt. ChatPRD can even identify which documents need updates based on a change you describe.
Observations on User Experience
While using ChatPRD, I’ve noticed a few quirks and delightful surprises that stand out. One minor annoyance is when the CustomGPT prompts me—the human—to continue by saying something like, “Tell me ‘Let’s Build Journey 3’ to continue.” These types of prompts can feel a bit forced or robotic.
On the flip side, one unexpectedly delightful feature of ChatPRD is its use of positive reinforcement. As you work with it, catching its mistakes or suggesting new ideas, it responds with encouraging feedback. These small affirmations help make the experience more enjoyable and engaging.
Reflections on Speed and Collaboration
ChatPRD has allowed me to move materially faster than before. On the last project I used it for, I saw time savings of at least a week, if not two or more. This acceleration came from having a structured partner that could keep pace with my thinking, organize documentation quickly, and reflect changes back to me with minimal friction.
That said, another collaboration issue that emerges as the product matures is determining where the true source of truth should live. ChatPRD is extremely helpful in the early phases—before much is known about the product and before the team is aligned on what’s being built. But as things become more defined, the need to go back into ChatGPT to make changes becomes less appealing. The friction increases when documentation lives in ChatPRD, but the actual product definition lives in the codebase or an external repo. It raises questions about ownership and maintenance of that knowledge.
Final Thoughts
We’re all living through a remarkable and transitional moment in software development. AI tools like ChatPRD are evolving quickly, and our ability to adapt and experiment is going to pay off in the long run. I’m excited — and optimistic — about what lies ahead. This feels like one of those early inflection points in a phase shift that we’ll one day look back on and wonder: how did we ever work the old way?
ChatPRD has earned a spot in my toolkit. It keeps me focused, helps me communicate systems thinking clearly, and accelerates the process of shaping ideas into usable artifacts. While there’s still room for improvement (collaboration, stability, and memory), it’s already made a meaningful difference in how I approach early project definition and planning.
If you’re kicking off a new project or simply looking for ways to bring clarity to complexity, ChatPRD is worth a spin.