Michael Royzen made his name training coding models that could outperform GPT-4 on HumanEval, a respected coding benchmark. That was 2023. Now he's applying that same technical obsession to a problem that's confounded quantitative traders for decades: Can you build a hedge fund where artificial intelligence does everything—research, analysis, execution—without a human finger on any trigger?
His answer is Standard Signal, a venture that promises exactly that. No portfolio managers making gut calls. No analysts second-guessing the system. Just AI, all the way down.
Whether that's visionary or hubristic depends on who you ask. And maybe on how the fund performs once it goes live with actual client money.
A Crowded Suddenly
Standard Signal isn't exactly alone. A cluster of similar ventures surfaced in early 2026, each claiming to have cracked the code on autonomous trading with large language models.
On April 24, 2026, Bloomberg profiled Abundance, where Instacart co-founder Apoorva Mehta deployed what the company describes as "thousands of bots" to research stocks and size positions. That same day, Lumenai issued a press release declaring itself "believed to be the first institutional hedge fund built on an agentic AI architecture." Phasic had announced its own AI-managed private fund weeks earlier, in February.
Standard Signal, backed by Y Combinator, also calls itself the "first hedge fund that researches and executes trades purely with AI" in its YC directory profile.
Someone's counting wrong. Or perhaps the definition of "first" has become elastic.
What all of them share is a pitch: Large language models have gotten good enough—at reasoning, at synthesizing messy data, at interacting with APIs—that they can now do what previous generations of statistical models couldn't. They can read an earnings call transcript, form a hypothesis about management credibility, cross-reference that against commodity price movements, and place a trade. All without a human intermediary.
That's the theory, anyway.
The System, as Described
Standard Signal's website lays out the architecture in broad strokes. The fund runs what it calls "frontier financial LLM models" that continuously analyze markets and execute trades within predefined risk parameters. The approach centers on chain-of-thought reasoning—essentially, the AI is forced to show its work, documenting the logic for each trade in what the company describes as a "fully explainable audit trail from signal to execution."
The strategy spans equities, commodities, forex, and derivatives, operating with a market-neutral stance. Minimum investment: $100,000, aimed at qualified purchasers who meet the regulatory thresholds for such things. The site also advertises a "2.0+ Sharpe Ratio," though it offers no time period, no benchmark, and no third-party verification for that figure.
For context, a Sharpe ratio above 2.0 would be exceptional. Renaissance Technologies' legendary Medallion Fund operated in that range, but over decades and with infrastructure most firms can only dream about. An unverified claim from a fund that registered in March—captured online in May—deserves a healthy dose of skepticism.
From Search to Spreads
Royzen's path here isn't entirely linear, but it makes a certain kind of sense.
He co-founded Phind in 2022, an AI-powered search engine that emerged during the post-ChatGPT frenzy when it seemed like every YC batch had at least three companies trying to reinvent Google with transformers. Phind competed directly with Perplexity and others, carving out a niche among developers who wanted fast, accurate answers to coding questions.
In October 2023, Royzen published benchmarks showing that Phind's fine-tuned CodeLlama-34B models outperformed GPT-4 on HumanEval, a respected coding benchmark. It was a genuine technical achievement at the time—before the frontier models leapfrogged again, as they tend to do.
Phind raised $10.4 million in late 2025, according to Axios—separate from Standard Signal's financials. A few months later, Royzen registered Standard Signal as a foreign business corporation in New York. The YC directory lists the team size as one person, though the company is hiring for an ML research engineer (New York or San Francisco) and a quantitative researcher in New York, offering equity, profit sharing, and benefits.
So: a one-person operation, at least for now, trying to stand up a hedge fund run by AI. Bold, if nothing else.
Not Exactly New
Fully autonomous trading funds are not a 2026 invention. Wired covered AI-driven hedge funds like Aidyia in January 2016, highlighting systems that executed trades without human input. Numerai has used crowdsourced machine learning models to drive its meta-model since at least 2016; it raised $30 million in late 2025 to scale operations. Qraft has run AI-managed ETFs since 2019.
What's different now—or what the 2026 cohort argues is different—is the architecture. Large language models fine-tuned for reasoning and tool use can synthesize unstructured data in ways earlier statistical models simply couldn't. A research preprint from January titled "Autonomous Market Intelligence: Agentic AI Nowcasting Predicts Stock Returns" describes a 100% agentic framework that autonomously searches, filters, and synthesizes information into predictions.
It's compelling on paper.
But there's a gap—sometimes a chasm—between promising research and robust live performance. An April survey paper, "A Review of Large Language Models for Stock Price Forecasting from a Hedge-Fund Perspective," catalogs the pitfalls: data leakage, evaluation biases, and the fundamental limits of market predictability. (Markets are, inconveniently, designed to be hard to predict.)
Meanwhile, retail infrastructure is evolving in parallel. MoonPay acquired Dawn Labs in May and launched Dawn CLI, an AI trading agent. Coinrule expanded its agentic framework to U.S. equities and ETFs in February, claiming it trained AI on 1.7 million strategies. These tools target individual traders rather than institutions, but they signal that autonomous execution is becoming, if not mainstream, at least familiar.
The Performance Problem
Here's where things get murky.
Standard Signal's website prominently displays that "2.0+ Sharpe Ratio" alongside the usual disclaimers: the site "is not an offer or solicitation," and "all investments involve risk." No start date. No end date. No methodology. The footer reads 2026.
The YC directory lists Pete Koomen as the primary partner for Standard Signal. The fund's regulatory posture is harder to pin down. Searches of the SEC's IAPD and EDGAR databases didn't turn up Form ADV or Form D filings under Standard Signal, Inc., at least as of the research date. That doesn't mean the filings don't exist—regulatory databases can lag, and naming conventions vary—but it does mean potential investors should verify compliance status independently.
None of this is unusual for a fund in its early days. But it underscores the gap between a sleek landing page and the messy reality of launching a financial vehicle.
The Edge Problem
Assume for a moment that Standard Signal works exactly as advertised. The AI reads earnings reports, parses news, reasons through public data, and generates alpha. What happens when the next ten funds do the same thing?
This is the paradox sitting at the heart of every autonomous trading venture: the more successful these systems become, the faster they eliminate their own advantage. If everyone's AI is reading the same information and arriving at similar conclusions, the edge evaporates. Speed matters, sure—getting there first still counts. But the window shrinks.
Standard Signal's landing page promises a system that discovers "new fundamental truths" before human analysts can. That's a high bar. Whether it's achievable depends on architecture, data access, execution speed, and risk controls—none of which are publicly documented.
What Happens Next
Royzen has demonstrated he can train high-performing models in one domain. Coding benchmarks are one thing. Financial markets—where signal decays fast, where mistakes cost real money, where randomness masquerades as pattern—are something else entirely.
The launch itself is straightforward enough. You register the entity, build the website, get into YC, start recruiting.
The hard part starts when the fund goes live and the AI begins making decisions with other people's capital. That's when marketing claims meet market reality. And that's when we'll learn whether Standard Signal is a technical breakthrough or just another well-intentioned experiment in a long line of them.
