Founder Fridays: Run towards hard conversations - Eric Liu, Bayes Impact & Ayomi Samaraweera, Canopy

Building data-driven products often starts with a simple question—who the user is and what problem matters—but both Bayes Impact and Canopy founders say the hardest part is resisting the momentum of execution long enough to correct course. Eric Liu describes Bayes Impact as a data science nonprofit working with government agencies and large nonprofits, built from the ground up through Y Combinator in 2024. The nonprofit’s mission stayed consistent—improving decision-making—but its operating model shifted repeatedly, moving from product development to services and back toward more product-like offerings as it learned what actually scaled.

A key lesson from Bayes Impact’s early days was that “talking to customers and selling” can become a trap if it’s done naively. The team began with a broad, ambitious premise: data science (later framed as AI) could help corporate organizations become more efficient. Without deciding whether the endgame was a software product or a tech-enabled services organization, customer conversations and rapid delivery led to a consulting-style business—complete with a growing engineering bench and high-profile hires. As headcount rose, the reality set in: the company risked becoming a large consulting operation rather than a scalable impact engine. Liu credits a later pivot to the courage to “flip” based on intuition—choosing what would be both meaningful and scalable—even when that meant admitting the current path was drifting.

That pivot required a cultural change as much as a business one. Liu emphasizes building a norm of brutally honest feedback early, especially when everyone is busy executing. He recounts an engineer directly challenging leadership with a blunt question: whether the company wanted to be a consulting business at all. That kind of candor, he says, prevents the team from treating discomfort as background noise.

Ayomi Samaraweeraera’s experience with Canopy highlights a different failure mode: when a founder lacks technical depth, shipping speed and iteration can stall. Canopy aimed to function like a “blind” for the creative economy, helping content creators share pay transparency opportunities and engage with management. As a non-technical solo founder who tried—and failed—to find the right technical co-founder, she struggled to move from customer conversations to implemented changes quickly enough to test adoption. Development timelines stretched, leaving a backlog instead of rapid experiments.

Both founders connect these lessons to a broader future for startups. Samaraweeraera argues that today’s tooling lowers the barrier for non-technical founders to prototype and even “vibe code,” then use that understanding to collaborate more effectively with a technical co-founder. Liu adds that AI shifts architecture and forces companies to assume general capabilities via APIs, while still needing a clear wedge and problem focus for vertical or use-case-driven businesses.

Across both journeys, the recurring theme is disciplined flexibility: keep a clear picture of the user and problem, move fast enough to learn, and create an environment where uncomfortable truths can surface before they harden into the wrong business model.