Problem/Solution Report

Problem – Across industries, R&D leaders frequently know how to evaluate progress but not which approach will achieve the best result. Traditional R&D relies on humans to survey literature, select methods, implement, test, and iterate — exploring only a small fraction of the solution space. This makes R&D slow, expensive, and prone to missing superior solutions that lie beyond human bandwidth. YC summarizes this succinctly: “You know what success looks like, but you don’t know how to get there… This isn’t a tooling problem. It’s a search problem — and humans are the bottleneck.”

Solution – Aemon positions itself as “the AI R&D engineer.” Given a problem description and an evaluation metric, Aemon reads state‑of‑the‑art research, then generates, tests, and evolves thousands of approaches at machine speed. It is expert‑steerable — teams can constrain the search, prioritize lines of attack, or redirect away from dead ends — while Aemon performs large‑scale autonomous exploration. The outcome is a validated solution against the customer’s evaluation function, benchmarked against public SOTA.

Evidence of efficacy – Aemon reports setting a new world record on a classic NP‑hard optimization benchmark (circle packing), beating Google DeepMind’s 2025 AlphaEvolve result with less than $10 of compute. On its site, Aemon also shows a legal‑document retrieval pipeline improving Recall@10 from 0.540 to 0.912 (+68.9 %), demonstrating the agent’s ability to discover and justify production‑ready pipelines.

Customer fit and workflow – YC’s profile highlights initial target users — CTOs, Heads of R&D, AI/ML teams aiming to stay SOTA, and computational teams in quant/finance, biotech, logistics, or materials science. These environments have clear evaluation functions but ambiguous paths to optimum, providing high leverage for autonomous exploration guided by in‑house experts.

Market and Competitors Report

Market context – Aemon sits at the emergent intersection of “AI agents for research and engineering” and enterprise AI development tooling. The category is catalyzed by (a) massive, rising R&D budgets; (b) the maturation of LLM‑based agents that can read, write, code, and run experiments; and (c) evidence that agent‑discovered algorithms can unlock compounding efficiency gains. Adjacent markets include AI platforms (>$70 billion by 2026) and rapidly scaling MLOps (projected $16.6 billion by 2030), indicating procurement budgets exist for solutions that translate into measurable performance or cost outcomes.

Competitors and analogs

• Google DeepMind AlphaEvolve – An evolutionary coding agent that discovers algorithms; reported real‑world production impact (e.g., 0.7 % compute recovery via improved Borg scheduling, 23 % kernel speedups). AlphaEvolve validates the enterprise value of agentic algorithm discovery, though it remains largely an internal Google capability.
• Sakana AI “AI Scientist” – A fully automated pipeline for end‑to‑end scientific paper generation and discovery, another vision for automating research workflows at scale.
• Devin (Devin.ai / Cognition AI) – Markets itself as the “AI software engineer” capable of planning and executing complex engineering tasks; public case studies (e.g., Nubank migration) show enterprise adoption for software work.
• Elicit (elicit.com / Ought) – An AI assistant for researchers that automates literature search, summarization, and data extraction. While narrower than Aemon’s end‑to‑end R&D loop, it competes for researcher time in the early stages of the workflow.

Aemon’s potential advantages

1. Verified beyond‑SOTA results – World‑record circle‑packing achievement with <$10 compute and publicly available verifier.
2. Domain‑agnostic, evaluation‑driven search loop – Integrates expert steering while remaining applicable across biotech, finance, logistics, materials science, etc.
3. Production‑oriented outputs – Demonstrated measurable improvements (e.g., Recall@10 uplift) that directly translate to business value.

Constraints – Requires well‑specified evaluation functions, integration with customer infrastructure, and head‑to‑head validation against strong baselines to earn trust.

Overall, the combination of rising R&D spend, proven ROI from algorithmic improvements, and a nascent ecosystem of AI‑agent tooling creates a receptive enterprise market where Aemon can position itself as an “R&D force multiplier.”

Founders and Background Report

Aemon was founded by twin brothers Ray Xu and Richard Zhou, who serve as the company's co‑founders. According to Y Combinator, both founders published at top AI venues (ICLR, EMNLP) before turning 20 and left university early to build Aemon full time — Ray from UIUC Computer Science and Richard from the University of Waterloo Computer Science. The company is based in San Francisco and is part of YC’s Winter 2026 batch.

Ray Xu – Ray’s profile, as summarized by YC, highlights early research productivity — “published at top AI conferences like ICLR and EMNLP before turning 20” — and a focus on cutting‑edge machine learning. He left UIUC CS to pursue Aemon full‑time, indicating a founder profile oriented toward rapid R&D and productization in state‑of‑the‑art AI systems.

Richard Zhou – Richard complements this with a competitive and research pedigree — YC notes he is an international medalist in mathematics and robotics — paired with Waterloo CS training (also a dropout) and early publication experience. This combination underscores depth in algorithmic problem solving and applied robotics/optimization, mapping directly onto Aemon’s thesis of using autonomous AI agents to push beyond state‑of‑the‑art solutions for complex engineering and scientific problems.

Collectively, the founders present strong signals of research excellence (early publications at tier‑1 AI venues), algorithmic intuition (math/robotics distinctions), and willingness to move quickly (dropping out to build the company). YC’s materials also place Aemon squarely in the “AI R&D engineer” category, with initial customer conversations targeting CTOs and R&D leaders across domains where objective evaluations exist but optimal solutions are unclear — exactly the kind of environment that matches the founders’ backgrounds.