Aemon

Aemon

The AI Research Engineer

Winter 2026ActiveB2BArtificial IntelligenceB2BSan Francisco, CA, USA
Company
https://aemon.ai
Aemon is the first autonomous research engineer that delivers state-of-the-art solutions to your engineering problems. It generates, tests, and evolves thousands of approaches at machine speed to discover optimal solutions beyond what human experts can find. Aemon already set a new world record on an NP-hard math optimization problem with <$10 of compute, beating the previous record set by Google DeepMind in 2025. Ray and Richard (twin brothers) dropped out of UWaterloo and UIUC to build Aemon. They published at top AI conferences like ICLR and EMNLP before turning 20.

Verdict

High Signal
Market Opportunity
Autonomous AI research engineering targeting hard optimization problems is a massive B2B market — applicable to logistics, drug discovery, chip design, financial modeling, and any domain with NP-hard computational problems. Enterprise customers in these verticals pay seven figures annually for even marginal solution improvements. ICP is less precisely defined publicly but the technical differentiation maps clearly to high-value industrial optimization use cases.
Medium Signal
Founder Signal
Ray has ICLR 2025 and EMNLP 2024 publications, research at NCSA/UIUC and Lapis Labs, and was FIR at Afore Capital — strong technical depth for his age (sub-20 publications). Richard has software internships at RBC and MIT-PITT-RW autonomous vehicle team, Vex Robotics World Champion, and Euclid math contest top 35 in Ontario — solid but lighter on deep AI research. Both are young dropouts (~20 years old) with limited industry experience, though their research output is genuinely impressive early signal.
Medium Signal
Competition
No competitor data was found in the research, but the space includes Google DeepMind (AlphaCode/optimization research), FunSearch, and AI coding agents like Cognition/Devin and SWE-agent. The specific angle of automated algorithm discovery and optimization at machine speed is more differentiated than generic AI coding assistants, and the benchmark result against DeepMind is a meaningful moat signal — but well-funded incumbents dominate adjacent research.
Medium Signal
Product
The world record claim on an NP-hard math optimization problem beating Google DeepMind with <$10 compute is a concrete, verifiable technical achievement and strong proof of capability. However, no named enterprise customers, no pricing page, no revenue metrics, and no press coverage are visible — the product appears to be early-stage with the benchmark result as the primary traction signal.
OverallB Tier

Technically exceptional young founders who've already produced a credible benchmark result beating a Google DeepMind record — that's a rare, concrete proof point at this stage. The core risk is that benchmark wins don't automatically translate to enterprise revenue: the product needs a clear deployment path, defined ICP, and paying customers to graduate from impressive demo to real business. Both founders are ~20 with limited enterprise sales experience, and the absence of any press, customers, or pricing suggests they're still early in commercialization. If they can convert the technical credibility into a handful of paying enterprise contracts in optimization-heavy verticals (logistics, pharma, semiconductors), this could accelerate quickly — but right now it's a strong technical bet with unproven go-to-market.

Active Founders

Ray Xu
Ray Xu
Founder

Co-founder at Aemon (YC W26). UIUC CS dropout. Published at top AI conferences like ICLR and EMNLP before turning 20.

Richard Zhou
Richard Zhou
Founder

Co-founder at Aemon (YC W26). Waterloo CS dropout. International medalist in math & robotics.

Aemon
Aemon
TierB Tier
BatchWinter 2026
Team Size3
StatusActive
LocationSan Francisco, CA, USA