When everyone has access to the same AI models, what actually makes one AI product better than another?
It's no longer just about the intelligence. It's about how reliably you put it to work.
An agent that responds faster. Makes fewer mistakes. Delivers better outcomes for customers.
That's the problem Niyaz Puzhikkunnath and Faiz Vadakkumpadath are tackling with Papaya (YC F26): continuously improving AI agents already running in production.
After spending more than a decade developing their expertise at Amazon, the founders set out to build a company of their own.
Their first product focused on data infrastructure. But as they began working with customers, they discovered something interesting.
Customers liked the data product, but were especially interested in the AI agent harness the founders had built to make their system more reliable.
Rather than asking customers to migrate their existing data infrastructure, they saw an opportunity to make that technology useful across a much broader range of applications.
So they developed the harness into a standalone product, laying the foundation for Papaya.
Today, Papaya analyzes how customers use production AI agents, identifies issues across prompts, context, caching, and tool calls, and opens pull requests with proposed fixes.
It then tracks whether those changes actually improve quality, latency, and cost.
Because knowing why your agent failed is only half the problem. Actually improving it is what matters.
That's what Papaya is building toward: AI agents that get better over time, without engineers having to manually investigate and fix every issue.
If you're running AI agents in production, check out what the team is building at papaya.fyi.
🎙️ Faiz Vadakkumpadath & Niyaz Puzhikkunnath, Co-Founders, Papaya on Fondo START with Guest Host: Maanav Agrawal, Founder & CEO, Memoir