Abstract: My first startup experience: 2024/07-2025/01

Last year in April, I decided to explore building a startup, and in June, I quit my job at Amazon to focus on it. From April to June, I also explored some AI startup ideas and worked on a few projects with people I met, but I didn’t incorporate a company or formally devote myself to building a startup. From July 2024 to January 2025, I incorporated a company with a cofounder and devoted myself to the 0-to-1 stage of building a startup. I transitioned to an advisory role after January 2025. The experience was bittersweet but very valuable, and I learned a lot. I will summarize my learnings and thoughts here. Please note that, per agreement, I will not disclose any details about the company, cofounders, or product, but will share my learnings and thoughts from my experience and those of other first-time founders I know.

Tech Stack

From a technical perspective, I have learned a lot about building full-stack SaaS apps with AI agents from the team and my open-source collaborators. Before this, I had limited experience with building SaaS apps and deploying AI agents into production: I had only built a few small Python fullstack projects using streamlit. Now I have a tech stack for fast prototyping and deployment: (1) Vercel for frontend, backend, and AI backend deployment; (2) Supabase for database, authentication, and subscription; (3) Next.js + shadcn + Tailwind for frontend; (4) Express.js + Node.js for backend; (5) FastAPI for AI backend. If we hit a technical limitation with Vercel, we can deploy to AWS.

Co-founder matching

I also have some takeaways regarding cofounder matching: Start with people you know well, and then expand your network. If you start with people you don’t know well, please take the time to understand their values and principles to check if they align with yours. Don’t just find a cofounder based on their resume and experience. Don’t be too quick to commit to someone (if someone pushes you to make decisions as soon as possible, that is potentially a red flag). You will need time to understand what type of people suit you well as a cofounder, and it will take time to develop your own perspective on this. As a first-time founder, you typically have these options regarding cofounder matching: (1) join a team that has already raised pre-seed funding or been accepted into an accelerator - in this case, it won’t be an equal split, but it reduces risk significantly; (2) start from scratch with another first-time founder, usually with a 50/50 or 51/49 split. Please note that unless the other person commits to putting in more money into the company on day one, you should argue for an equal split, since 51% means the other person can remove you from the board and force you out if you disagree. The other person can also change your equity since they have the voting power, which would put you in a vulnerable position. (3) Work with a cofounder who is significantly more experienced than you - for example, a successful serial entrepreneur or a director+ at big tech companies or VP+ at series A+ startups, while you are a staff software engineer or engineering manager. In this case, you would take less equity, but your chances of success would be higher. (4) Join a team that has already built a product (whether it’s just a prototype, a working product, or one with paid customers), but hasn’t raised any funding and is completely bootstrapped. In my view, if the founder already has some paid customers and is on the right track for fundraising, and you fit into the team, it might be worth taking less equity to join them. However, if the founder has only built a prototype or MVP, then as a builder, it doesn’t really make sense for you to sacrifice your equity. Please note that any cofounder relationship is at-will, meaning you can leave immediately if you want to. However, make sure to work on a separation agreement to ensure the company can continue operating without you.

Principles

I have found it very important to align on certain principles and values when building products with the early team members. Some of my principles: (1) For B2B ideas, speak with potential customers to verify their need for the product before building the product. (2) Build quickly and verify technical limitations before committing: Launch a minimal version to prove the concept rather than spending months building something without certainty of success. For example, if AI is a core feature, test its limitations in the problem space before investing months in full-stack and infrastructure work. (3) First build to ship, then refactor - don’t overcomplicate the tech stack before achieving product-market fit, as refactoring can always come later. (4) It’s ok to start with non-scalable solutions, focus on selling the product before scaling operations.

Community & Study Group

In addition to my former startup, I joined the South Park Commons founders community to meet more people and learn/build with them. I also organized a non-profit research organization called PathOnAI.org (https://github.com/PathOnAI-org) to collaborate with people online on open-source projects. I learned a lot from my peers and mentors in these communities. When there is a shared thesis bringing people together to work on something, they can learn from each other and build momentum more effectively than working alone.

My next steps

I plan to take a break and recharge instead of jumping into another startup right away. I’ll continue focusing on learning, exploration, and building:

Learning

  • Diving deep into embodied AI and robotics fundamentals

Exploration

  • Investigating arc-agi
  • Studying reasoning AI frameworks and their applications
  • Following the progress of DeepSeek R1

Building

  • Developing full-stack SaaS applications powered by AI agents
  • Implementing search and reinforcement learning systems with LLMs

During this time, I’ll wrap up my full-stack learning experience (documented at https://danqingz.github.io/blog/2025/01/20/full-stack.html) and complete my open-source/ research work on web agents. Additionally, I’ll deepen my understanding of reasoning AI, embodied AI, and reinforcement learning.