Interest in how to build an AI company has been growing recently, but there's a shortage of concrete know-how on how to actually start and operate one. In this article, the author explains in detail and chronological order the actual process he went through when shutting down an existing analytics product and building a new AI-based product line. The core of this article is presenting practical methodologies from the perspective of someone just starting an AI company.
1. The Most Important Thing in an AI System Is 'Context'
Many people tend to focus solely on prompt engineering when working with AI, but the author emphasizes that properly handling context comes first.
"Most people are too obsessed with 'prompt engineering.' But when thinking in the AI world, context comes before that."
When building an AI system, you need to systematically consider questions like:
- Where is the context stored?
- How much context is needed for the given task?
- How can context be delivered effectively?
Based on these considerations, the author emphasizes that the most important context when building an AI-native company is 'business context.'
2. Organizing Context with the Business Model Canvas
When starting an AI company, it's helpful to use a framework like the Business Model Canvas to organize the core elements of your business. This process greatly helps in collaborating with team members to refine ideas.
- Bring the Business Model Canvas into a collaboration tool like FigJam and brainstorm with your team.
- Fill in each section of the canvas, visually organizing the overall business context.
The completed canvas can then be fed into an AI tool like ChatGPT to extract each section into structured markdown text.
"Please extract all the information from this canvas. Use each box as a heading, and organize all cards into more complete sentences."
This gives you the raw context of your business in text form.
3. Managing Context as 'Living Information'
Many people make the mistake of leaving this context only in a ChatGPT conversation window, or continuously adding to the same conversation. The author points out that doing this kills the context.
"If you just leave context in ChatGPT, the context dies."
Instead, the author proposes using a git repository.
Why and How to Use a Git Repository
- Create a new repository on a platform like GitHub.
- Install a tool called Windsurf and connect it to the repository.
- Give Windsurf the following instruction with the markdown text extracted from ChatGPT:
"Create a new folder called 'business-canvas' at the project root, and create a file for each heading from the text above. All files should be in markdown format, with filenames in kebab case."
This creates organized files for each section in the repository. You can then review each file to check and correct whether the information was properly captured.
4. Treating Context Like Code and Collaborating with AI
Now that context is organized in a repository like code, you can manage it the same way you manage code. For example, you can ask AI:
"What do you think about customer relationships and how they're served? Brainstorm some ideas, and if I like them, update the file."
By repeating this process, you can continuously evolve the context by creating new files or merging context.
"Now that context is stored in a software-native protocol, you can manipulate it like code!"
5. Integration with AI Code Editors and Organizational Culture Change
This organized context repository can be integrated with AI code editors. In other words, when AI writes code, the business context is already embedded, eliminating the need to repeat business explanations each time.
- All context is committed to the git repository, making change history trackable.
- Storing context in Notion or Google Docs makes version control difficult and risks losing context.
- Establishing a code-first (context as code) culture within the organization is crucial.
"When business context is codified, changes in thinking are automatically recorded through git commits."
This way, you can track how business context has evolved, just like a codebase. Additionally, not just engineers but marketing and sales teams can easily share the information they need.
6. Organizational Change and Challenges for an AI-Native Culture
Finally, the author emphasizes the importance of a 100% AI-native culture. Every member must actively adapt to AI tools and new ways of working for a company to truly become AI-native.
"The most important lesson from this AI transition is that you need a 100% AI-native culture. Without full buy-in and voluntary exploration, it's hard to build such powerful systems."
However, this process can feel uncomfortable for people unfamiliar with git or code editors. There's still resistance from those who say "AI is just a passing fad" or "Why should we use a tool that's wrong 60% of the time?"
"There are still many people who resist these new approaches."
7. Closing: Building Companies in New Ways, and the Open Opportunity
The author says that an entirely new path to building companies the AI-native way is open. The article closes with warm encouragement hoping it will be helpful on your journey.
Key Concepts Summary
- AI-native company
- Context management
- Business Model Canvas
- Git repository, Windsurf
- Code-first culture
- Integration with AI code editors
- Organizational culture change and adaptation
This article chronologically outlines the first step toward building an AI-native company, centered on treating context like code. If you want to try new ways of working with AI, the methodology in this article is well worth referencing