Learnings From An Early-Stage YC Startup in 2023

When I first joined as Design Lead in September 2022, Sphere was a corporate education startup where vetted leading experts ran live, cohort-based courses for professionals around the world on topics such as machine learning, data science, and sales. 

Backed by Felicis Ventures and Y Combinator (W22), Sphere had upskilled learners from over 700 businesses and had annual contracts with companies like Stripe, Flipkart, Tinder and Accenture. 

I started by designing wireframes, user interfaces, and marketing materials, but over time my role expanded to include interviewing customers to plan new product features as well as working with a team of offshore developers to manage product releases.

New skills: I already had UX skills from past education and internships, but applying them to real world use cases over an extended period of time was very exciting. I was also new to product management which I really came to enjoy despite how daunting it was at first!

Startup lesson: I gained firsthand experience moving fast, understanding how startups work, and wearing more hats than I’d ever worn before (little did I know what was to come next…)

Slowing growth: Although Sphere was doing very well the summer and fall following YC, growth started to slow down that winter as companies continued to cut their L&D budgets, sending fewer learners to our courses. 

Difficulty scaling: It was at this point that the founders realized that the existing product would also be very difficult to scale, as developing and running high-quality live courses was very time and resource intensive.

Generative AI boom: OpenAI’s release of GPT 3.5 was also taking the world by storm, which inspired us to think about ways we could use generative AI to make our corporate education product more scalable. 

Designed an MVP for an asynchronous learning portal including an AI search assistant and advertised the offering to our existing customers.

New skills: I spent this time learning as much as I could about generative AI without an engineering background. I did a lot of reading, attended panels and events around San Francisco, and had many insightful discussions with my CTO. I went from knowing absolutely nothing to having a solid understanding of generative AI and how it works.

Startup lesson: Unfortunately, the AI-powered learning portal generated a lot less interest than we had hoped for, which is how I learned the hard lesson of knowing when to walk away from a business idea. 

At this point, the founders made the difficult decision to pivot out of the education space completely, laying off everyone (about ten full and part-time employees) except for the offshore developers and––you guessed it––me. Although it was jarring at first, I was ultimately honored that the founders valued my contributions and believed in my resilience enough to keep me on board. 

We spent the next month conducting many, many interviews with engineers, product managers, and sales professionals in an effort to surface problems they faced that we felt we could build a good solution for. This involved a lot of LinkedIn cold outreach, Zoom interviews, affinity maps, ideation sessions, pitching ideas to each other, and scoring them on business potential.

New skills: I had conducted exploratory interviews before, but I really honed these skills and learned so much more than I expected about the professions of sales, product management, and software development!

Startup lesson: How to pitch and evaluate business ideas for potential using criteria such as retention, scalability, competition, market potential, founder/market fit, and feasibility. 

After evaluating the ideas we generated from our research, we decided to tackle a manual, repetitive, time consuming task we discovered was experienced by many data engineers: Building data integrations. 

We began a design partnership with Airbyte, a Series B open-source data integration engine that helps data engineers consolidate data in warehouses, lakes and databases. Our goal was to build a system that automatically created low-code Airbyte connectors using only an API’s online documentation. 

I planned the UX and designed the UI for our low-code integration system, which we built and delivered to Airbyte over the course of a month. We also explored some tangential ideas such as an AI Assistant trained on online API documentation to help admins and data managers implement software in their organizations, which I also planned and designed an MVP for.

New skills: Previously, I didn’t even know what a data integration was! Through this project, I gained an understanding of how they work and how to design for them.

Startup lesson: Focus on pursuing ideas you have genuine interest in––throughout the automated integrations project, we realized we were not passionate enough about the subject to continue to improve and grow it into a scalable business.

At this time, we became quite interested in the idea of helping content creators monetize their content and protect it from being freely scraped by generative AI models. Many artists were concerned about protecting their work from being used to train AI image generation models such as Midjourney, Stable Diffusion, and OpenAI’s DALL-E. Reddit’s announcement that it would charge for its API to prevent large language models (LLMs) such as ChatGPT and Google Bard from freely training on its data also caused a lot of controversy.

We decided to build an infrastructure for content owners (such as publishers, universities, and enterprises) to block unauthorized scraping, structure their data for LLM use, and generate licensing revenue from it. I designed a landing page to explain the product to potential customers.

New skills: Through research and design projects, I learned about how LLMs and the creator economy work, as well as how to design for them.

Startup lesson: As excited as we were about the idea, we learned through meetings with publishers and content owners that although our potential customers were concerned about protecting their content from being freely scraped, most of them were not ready to spend money on a solution. Through this, we learned that if the demand is not strong enough, the idea is not going to work no matter how interesting or topical it is.

Designing for generative AI applications such as agents and vision models. Exploring new verticals such as sales, product, finance, and HR.

Stay tuned for more details. Reach out to ashley.zhang@columbia.edu to learn more.

Thanks for reading!