Papaya tunes AI agents that are already in production. You connect traces with a one-line SDK or hand over raw data, and it runs 200+ research-backed analyses across thousands of runs to find what is recurring — prompt and caching inefficiency, tool schema drift, context bloat and retrieval strategy, model selection and routing, retry loops and wasted API calls. Findings come back ranked by impact, each with the specific runs affected, an estimated effect on quality, latency and cost, and a confidence level, delivered into Slack, a dashboard or a PR. The company cites cost reductions up to 52%, latency improvements up to 61% and quality gains of +11%, and offers a free 24-hour audit. Founded in 2026 by Macy Mody, Niyaz Puzhikkunnath and Faiz Vadakkumpadath, Papaya has 3 employees in San Francisco.

Aiops B2B Enterprise-Software YC Infrastructure AI

Cool Company Raises · YC F26

Agent 37 launches at YC (Fall 2026)

Agent 37 runs coding agents as always-on hosted sandboxes, so you host neither the machine nor the state. One POST to /v1/instances provisions an isolated environment with its own URL that you can message across sessions; files, memory and connected tools stay put until you delete it. It runs Hermes, OpenClaw, Claude Code, Codex or a custom harness, in 2-8 vCPU / 4-16 GB sizes, with gVisor kernel-level isolation, encryption at rest and in transit, and SOC 2 Type I. There is no plan fee — compute meters by the minute at $0.0011 per vCPU-hour and $0.00096 per GB-hour of RAM, which puts a continuously running 2 vCPU / 4 GB sandbox near $4.76 a month. Founded in 2026 by Vishnu Krishnaprasad, based in New York City.

Developer-Tools SAAS B2B YC AI

10 Sept

Cool Company Raises · YC F26

Mirrors launches at YC (Fall 2026)

Mirrors builds runnable staging environments for AI agents. It captures production traces from an agent's real sessions, reconstructs a working twin of the internal tools and APIs that agent calls, then replays those sessions against every prompt, model or tool change before it ships — so regressions surface in staging rather than in production. The gap it fills is that agents call bespoke internal services no standard test suite can reproduce, which matters most where a mistake is irreversible: issuing refunds, updating records, sending outbound messages. Founded in 2026 in San Francisco by Aryan Singhal (ML research at UT Austin and UCSB; six AI/ML papers published while in high school, three at NeurIPS). Y Combinator Fall 2026.

Developer-Tools SAAS B2B YC Devops AI

8 Sept

Cool Company Raises · YC F26

ORO AI launches at YC (Fall 2026)

ORO AI runs an open benchmark for agentic commerce: builders submit shopping agents that independent validators score against problems that change daily, the best performers earn rewards on Bittensor, and every run produces training data that feeds back into the next generation. Founded in 2026 in Vancouver by Shardul Bansal (ex-AWS, Crucible Labs) and Seth Schilbe (ex-AWS). Y Combinator Fall 2026.

Artificial-Intelligence Crypto-Web3 Consumer-Products YC B2B

31 Aug

Cool Company Raises · YC F26