“The Impending, Inescapable Deluge of A.I.”
Our start in AI goes back to our work on "language processors" — similar to software compilers — a long time ago at Genentech.
Since then, our work restarted at a company named X-Mine, where we were charged with designing new tools to understand relationships between human genes after the human genome was finally decoded. We developed some of the world's first language models based on vector embeddings, the core foundation of today's AI models. Interestingly, we named our product "Opus" 12 — the same name Anthropic later chose for its flagship AI model. We patented our approach 3. The primary sources for all of this, including NEC's own 2003 announcement, are published at 18.
This is where the story begins for our shareholders and stakeholders.
My background is not just in computer science and software engineering. I was born and raised on the streets of Oakland, Richmond and Berkeley, California, along with Wall Street and deep within the forests of Eastern Washington state, where street smarts of any kind do not matter — the only thing that matters is knowing how to track your prey, hunt, stand your ground with a few bears now and then, and keep warm in the winter.
The street smarts I got from Wall Street were given to me by my mother, who encouraged me to read my first prospectus, for Weyerhaeuser Paper, when I was 12 years old. The only TV we were allowed to watch was the Nightly Business Report and Wall $treet Week with Louis Rukeyser.
My dad was a lot of things: a hod carrier, an entomology researcher and firefighter at UC Berkeley and Berkeley Lab, and a general contractor, home builder and remodeler.
Although my mother encouraged me to get a job at her firm, Montgomery Securities, as a floor trader/runner — which I wanted to do — I decided not to, because getting my CFA (Chartered Financial Analyst) included too much homework. I was 18. I decided to pursue other "interests" to make money, some of which included going into the construction industry as a hod carrier like my dad, and then building homes and getting my general contracting license.
After a while, I decided building 5000 lot tract homes had a lifespan and a career ceiling, along with a ruined elbow from swinging a 21-ounce Dalluge hammer all day long.
I had rebuilt hot rods with friends, so I decided to become a mechanic at an auto body shop in Oakland, California. After attempting to make money in that world, I decided there was too much risk — and anyone who has worked at a chop shop knows exactly what I'm talking about.
To remove myself from that world, I decided to take what my dad and other seasoned general contractors had taught me and create a career for myself as a "software contractor." This meant I first had to work for a real estate management company, where I was assigned to evict people from their homes while also managing the SCO XENIX/Informix 4GL operating system they used to manage their properties.
This led to my entry into Silicon Valley.
I found out that you could use software to develop algorithms for the stock market. However, I first had to get a job as a software engineer. That's when I got my first big break at Genentech, who called a recruiter looking for a software contract engineer. I was on the list, went in for the interview, and they chose me.
I was charged with finding patterns in data — particularly pharmaceutical manufacturing pipeline data and Programmable Logic Controller (PLC) data (think IoT data or STUXNET) — related to Herceptin, the world's first HER2-targeted breast cancer therapy.
My assignment was to develop "language processors," the kind used to compile code like C/C++ — in our case rebuilding SYLK file formats for the pharmaceutical pipeline schematics. I was also charged with developing a "Pharmaceutical Recipe Builder" (PRB) system that would securely handle construction and revision control for Genentech's pharmaceutical recipes — the recipes behind the HER2 breast cancer program, DNASE for asthma, and Protein A for blood clotting.
After about a year, Genentech gave me an ultimatum: either commit to 14-hour workdays for a big pharmaceutical release, or finish my bachelor's in computer science at UC Berkeley and come back when done. I decided to counteroffer and ask if they would become the first customer of the consulting agency I was starting, The Berkeley Integration Group. They agreed, and Genentech became my first customer.
After a while, I decided that I wanted a piece of the Internet economy, so I started a web engineering and consulting company and spent 10 years in Silicon Valley working and consulting for all the big names and startups of the time. I learned a ton.
In order to hedge my career trajectory, I decided to take my mom's advice and use her firm to buy my first stock: SCOC (Santa Cruz Operation), a proprietary UNIX house and the maker of XENIX. Invest in what you know, at least for your few positions.
I began to develop software to analyze the movement of SCOC, which led to analyzing the movement of other stocks. Stocks are like animals with some having "animal spirits" 19.
I developed a simple set of algorithms to predict the price of stocks that would provide a decent return. The system worked and I made money — until I didn't. The algos were long-only and I was in a bull market. Anyone can make money in a bull market. When Long Term Capital Management (LTCM) collapsed 4, within roughly a year of the Asian financial crisis 20 and alongside the Russian ruble crisis 21, I was introduced to my first real bear market. I lost big, and all my gains plus some were wiped out. That taught me a few big lessons: (a) realize what market you're in and learn how to short, (b) manage risk, (c) hedge and (d) trading vehicles do not move up or down based on supply and demand as most teach, they move up and down based on perceived future supply and perceived future demand.
After the dot-com crash, I went back to biology — in particular, bioinformatics in human genomics at X-Mine in South San Francisco, California. This is the study of using software and algorithms to analyze biological information to interpret relationships between genes, drugs and diseases. Every stock is similar to a human gene. Every gene can be represented by a vector embedding, just as a stock can. This is also called biological language modeling in our case.
Due to our work there and in a few other places, I was recruited by Lawrence Berkeley National Laboratory, the Department of Energy, the Department of Defense and a few other agencies to develop AI systems based on vector embeddings to understand how to protect the human body during space flight. This also included extracting hidden connections between entities for target acquisition for US Navy SPAWAR (Space and Naval Warfare Systems Command, now NAVWAR). Palantir picked up this line of work at LBNL/DOD in 2005, building on the same "relationship networking" approach described in our patent 6 — a system that builds networks of relationship vectors expressly to surface what the filing calls "hidden association and connection extraction."
I had a chance to advise scientists at Berkeley Lab — a laboratory that has produced 17 Nobel Prize winners, where the cyclotron was invented and where plutonium was first isolated — on how to effectively commercialize their intellectual property by bringing in strategy and tactics from industry, based on my work in Silicon Valley. In December 2007, Berkeley Lab's Excellence in Technology Transfer Awards, hosted by then-Lab Director and Nobel laureate Steven Chu, featured our work as one of the Lab's technology transfer success stories 5.
Some of that work I did not do alone. In 2006 I published in BMC Bioinformatics with David Blei and Michael I. Jordan 16 — Jordan, whom Science later called the world's most influential computer scientist, taught Andrew Ng, who went on to found Google Brain, and taught Blei, who co-invented Latent Dirichlet Allocation. We were applying their topic models to find genes related to lifespan while I was developing the vector relationship patents at the same institution. The people who built modern AI and the people who built our approach were, for a while, in the same room.
Those patents also described a way to uncover hidden connections between stocks and global events. Our filings — the first with a 2002 priority date 3, the second with a 2005 priority date 6 — predate by roughly a decade the 2013 word2vec paper by Tomas Mikolov and colleagues at Google 7, which brought vector embeddings into the mainstream. LBNL asked whether we should pursue infringement. I advised against it.
We decided to take this intellectual property and work with Lawrence Berkeley National Lab's technology transfer division to commercialize it through a startup, SeeqPod, in the search and recommendation engine space 8. The technology, the Biomimetic Search Engine, went on to win a 2008 R&D 100 Award 9. At its peak SeeqPod indexed more than 13 million tracks and served 50 million monthly users running 250 million searches a month, before the global economic crash of 2008 took its toll on every startup in the land. Battling litigation with Apple, Warner, EMI (Citi) and a bunch of other labels did not make it easier 8.
The Lab and the US government (DOE) took a 5% stake in my startup based on the patent described above, along with the reputation of our team.
That line of work eventually produced five patents in search, discovery and relationship networks 3617, the later ones acquired by Intertrust Technologies — a joint venture between Sony and Philips. The through-line across all of them is a single idea: represent things as vectors, then read the distances between them to find relationships nobody wrote down.
We did not apply this IP and technology to the financial markets until we started consulting for hedge funds. One of them was Artiman Capital, the hedge arm of Artiman Ventures, the Palo Alto venture firm where I presented our work. We decided to part ways with Artiman after watching Bitcoin's growth since its launch in 2009.
In 2013, I decided to start SongCoin (SONG), a way for musicians to tokenize their music tracks. I wanted to turn it into the world's first music stock market. We hired one of the founding developers of Namecoin (NMC), the first cryptocurrency after Bitcoin. We introduced SongCoin in the press in February 2014 101112 and met with BitTorrent for a partnership and acquisition talks — and immediately a man from Queensland, Australia, David Prince, took the symbol by issuing his own Bitcoin fork of SongCoin under the same symbol, SONG, and got it quoted on CoinMarketCap. Our lead investor saw this, along with Mt. Gox and the cost of developing wallets from the ground up, got scared of crypto (he also thought the word "crypto" sounded criminal, as he told me), and walked.
Needless to say, we had to put the music stock market strategy on ice.
A former lead investor approached me during that time and asked what I was working on. I told him: not SongCoin, but an app that helps people decide what cryptos to buy. He did not know what crypto was, so I pointed him to coinmarketcap.com. The next day he called me early in the morning, almost out of breath, telling me we needed to do an "ICO" ASAP. I explained that we had the IP and the product line to do this and that it was going to be based on AI and datasets. We would call it StarMine.ai. We moved StarMine.ai forward with a smart contract and a symbol, SME. After reviewing regulatory issues related to ICOs, we pulled the ICO and chose not to go down that path. He wanted to fund the company by deforesting a section of land in Liberia, selling it to Chinese buyers, and then bringing that cash back into the US to fund development, operations and exchange listings. After I explained that all financial transactions in crypto are tracked, he decided to walk away from the deal. That left StarMine.ai high and dry.
This meant we had to start fresh with Vectorspace.ai (VXV): datasets that power AI, including on-chain data provenance. This was in 2016. We were early, as usual.
We spent a year or two building out search and discovery for cryptos initially, which led to a solid AI foundation based on vector embeddings in datasets — correlation matrices used as tensors in AI. A dataset is like a waffle, and a tensor is like a stack of waffles on a plate. Tensors make up a big chunk of the core of any AI system, while the vectors and their tensors make up the insides 13.
In 2021, based on our work, deals with companies like Elastic.co and Bitvore, and tailwinds in the crypto markets, we went from $0.14 to $20.00, giving us about $600 million to work with — and not just on the market cap side either.
During that time I decided to start Vector Space Biosciences (SBIO — the equity token for Vector Space Biosciences), as that is where half my domain expertise lies, aside from the global financial markets and their trading vehicles. I also knew that we could take what we knew about human biology in the harshest conditions known to man and apply those innovations in the financial markets, as we always have. We presented at Morningstar. Most thought it was science fiction — space biosciences, AI and crypto. Some still do. But then again, some people still live in caves.
When we spiked to $20 we chose not to sell — not even to take half off the table — as we did not want to send a signal to the broad market about our confidence in the deal. We held. We continue to hold. However, after several true crashes in the crypto markets, starting with FTX and many others going under, we were in need of capital.
We were approached by a few groups, but one in particular had a deal that involved swapping VXV for VAIX (a tokenized asset representing datasets used to fine-tune AI models) with a new token and equity structure. We needed the capital, so we agreed to the deal. This investment group turned out to be toxic: they bet what they had agreed to invest on memecoins and lost it all, then told us they no longer had the money to invest. We cut our losses with them and moved on.
This is when we decided to issue VectorCube (VCUB), a real-world asset (RWA) for CubeSat launches. We're not a one-trick pony, and due to our background in the financial industry we decided to begin tokenizing assets we had. One of them was the LNX10/SNX10, a long/short hedged dynamic 10-component index we designed as a financial instrument to be traded on ProBit. It used an algorithm to swap cryptos in and out of the index dynamically and priced the index using an index option pricing calculation, so the price would be tradable under a dollar. We used an early version of Anthropic's Claude and OpenAI's models to write a market making system that would hold the LNX10/SNX10 to their trading levels synthetically, without holding the underlying components.
Although the LNX10/SNX10 were trading perfectly and performing, they were still synthetic, and that was too much risk to introduce to our shareholders and stakeholders without risking our reputation — so we decided to wind them down from ProBit and will be winding them back up with underlying components starting with tokenized stocks once we raise capital and find a decent exchange for them. For now they'll be trading on DEXs, including our own at cymetica.com. ProBit was making a lot of money on them in trading fees and did not at all like that we wound them down. I explained that they had served their purpose: they proved that our AI modeling of the indices, their hedge, and our AI market making system and its code worked perfectly, and served as a proof of concept for a new set of financial product trading vehicles we planned on issuing.
We knew the AI token wave was coming, so we issued TBIG and TBIGS, AI token indices, to plant a flag in the ground ahead of it. By this time ProBit as an exchange was starting to buckle financially. ProBit approached us and began a shakedown for cash, which included an ultimatum: if we did not buy fake volume for SBIO, VAIX, VCUB, TBIG and TBIGS, they would delist us. We weren't born yesterday and knew exactly what the exchange was doing. They were running out of cash and needed transaction volume to make it up, so their plan was to shake down every token they listed. We didn't bite.
Companies will almost always be worth more than the exchanges they trade on. Exchanges have been in trouble since Mt. Gox. Before crypto existed, exchanges such as the NYSE (now part of ICE) and Nasdaq never made as much as the companies they list. This is the way global financial markets work. The great majority of people on the billionaires' list got there by owning equity — that is, ownership in startups and companies — not by owning or operating an exchange.
At this point I knew we'd have to prepare to list our tokenized assets and trading vehicles on DEXs while the CEXs began their collapse.
This is when I realized we'd have to finally start our own capital engine, CyMetica, with its own line of financial products and instruments based on everything we know about how global financial markets and trading vehicles work — including stocks, options, crypto, futures contracts, perps, you name it. This would mean opening up the kimono on algorithms we've designed to identify opportunities in information arbitrage between global events/headlines and tradable assets that have hidden connections to those events. Knowing how to do this using vector embeddings that fine-tune multi-trillion-parameter frontier AI models like Claude Fable 5/Opus 5, OpenAI's GPT-5.6, Google Gemini, Kimi 3.0, DeepSeek and others is our specialty, and always has been, based on our history and track record. This is also called financial language modeling, related to biological language modeling.
On November 23, 2025, we decided to start CyMetica with its first product line, EventTrader. We chose November 23 because we knew that on November 24 Anthropic was going to release Claude Opus 4.5 14 — in our view the first truly commercially viable frontier AI model, meaning you can rely on it to build an entire company from inception, from the DNA and the ground up, and become a true AI-native company.
Being an AI-native company is not just a label, and it is not bolt-on AI. It means you open a terminal in Claude Code and your first prompt goes something like this: "Create a plan to build a company with products A, B and C that do X, Y and Z." That's where it begins if you're a true AI-native company.
The advantage of being an AI-native company is that AI becomes part of your DNA. More importantly, the AI system is able to fully map out all the needs of the company — not just software engineering, but corporate strategy, partnerships, revenue, marketing/PR/communications, customer support, legal, and the list goes on. Every aspect of the company then relates and connects to every other aspect, and to every single data point in the company. Corporate strategy feeds off engineering which feeds off basic scientific research which feeds back into practical applications and prototypes, which feeds off customer support, which feeds product development, which feeds marketing and communications, which feeds business development, which feeds back into feature development and back into software engineering. It's a series of millions of virtuous feedback loops. This is, in part, what agentic engineering — an Agentic Operating System (AOS) — does. You don't get that by bolting on AI after the fact. This is another reason you're seeing companies collapse that cannot adapt to AI.
Another advantage of an AI-native operation is that any other AI model can verify your legitimacy. This is another reason you'll see many companies collapse, especially in crypto: it is very hard to build a scam AI company. Multi-trillions have been put into these systems to make sure of it, with or without guardrails.
Before crypto, global market players, institutional and retail, were all trading stocks, options, commodities, currencies, and other trading vehicles like junk bonds, debentures, credit instruments, and assets of all kinds — anything that would move and wind up on an order book or OTC, including private investment in public equity (PIPEs) and back-room deals from Silicon Valley to Wall Street.
When forex/FX currency guys found out about Bitcoin, they thought, "Wow, I can trade this, because it's setting up to be a global currency." This was the biggest misconception, on the part of both the original Bitcoin devs and the FX guys. Neither are stock guys. In fact, a lot of FX guys hate stock/option guys, and a lot of stock/option guys can't wrap their heads around why someone would trade a vehicle for fractions of a cent - bips - when you can trade a stock for multi-point baggers without liquidity being a factor.
Enter Silicon Valley, with startup equity guys having no clue how FX or stock guys operate — which includes investment bankers (IBs), otherwise known as "masters of the universe" in global markets, or at least they used to be.
Competition between Silicon Valley VCs/angels and Wall Street is where the real story begins. If you can get hold of a deal before the IPO, you win bigger than a Master of the Universe IB taking their underwriting fee for roadshowing the prospectus, the team, the opportunity, and the SEC regulatory pipeline. Bankers are there almost as spectators.
All of the above is where the largest money pipelines on Earth exist. Who controls those pipelines is what matters. Would Silicon Valley move Google forward and get their piece before the IBs, or get Facebook trading on SecondMarket before pawning it off to the Wall Street IBs for bragging rights and big cash? Sure. Would IBs take the ball and run with it? Sure.
Enter crypto: a group of people that some describe as what would happen if you took everybody who knew very little about software and technology and combined them with a group that knew very little about how financial markets actually work. The reality is that most in crypto know very little about how Wall Street or Silicon Valley works. They called what is a standard startup deck — a market requirements document (MRD) with technical documentation and a prospectus — a "whitepaper," when most true software engineers and researchers know very well that a whitepaper is not a mixture of a marketing document and a startup deck. Throwing it up on GitHub does not make it one either. And a whitepaper is not a peer-reviewed published paper, similar to Satoshi's.
Mistake #1: thinking that crypto is a currency. Mistake #2: thinking that an "ICO" is like an IPO — another huge disaster, as anyone with domain expertise in Silicon Valley or on Wall Street knows all too well that you don't go around giving away your "shares" for pennies before you float your issue, also known as an Initial Public Offering. Not only did this encourage "teams" with "projects" to raise money by giving away the majority of their holdings before going out, it also caused massive sell pressure.
Confusing crypto as a currency, combined with thinking an ICO was an IPO — when it should have been treated like a standard friends-and-family round pre-IPO — and then breaking a cardinal rule of Silicon Valley startups: if you take the founders' share and drop it below double digits, you remove something called "incentive," a very important pillar, and one Charlie Munger and others have described as among the most important rules in building a company.
When people saw Ethereum do an "ICO," they thought IPO, and then it was game on. But companies (teams) like Auroracoin, Ripple and others were still thinking that if you can issue a trillion tokens, and if crypto is a currency, then you can just sit back and take transaction fees while your "stock" goes up, and become a multi-trillionaire. This was a complete misconception based on a lack of understanding of how IPOs, floats, stock vs. crypto, and currencies actually work. Let's not even begin to address the SEC and the IBs getting upset about so-called "ICOs," which attracted all kinds of negative attention and even jail time for many.
The crypto world found out that crypto trades like a stock, not a currency. This was a big wake-up call to many, and still is today. Taking transaction fees because you think you can just issue a trillion "coins" and then seed the market, like Auroracoin did in Iceland — or like so many other crypto outfits thinking crypto is a currency — is a losing game.
People never wanted to use their crypto and spend their BTC like a currency if it trades like a stock and spikes up or down — mainly up. You don't trade something that trades like a stock as if it were a currency. You don't spend stock and for good reason, you let it ride.
All of this was the result of a lack of understanding of how actual financial markets, instruments, and trading vehicles work — from inception and startup equity to IPO.
The only thing most crypto "teams" were attempting to do was mimic what was happening in Silicon Valley and on Wall Street, but most got it backwards and upside down. The only ones who got it right were the exchanges and true startup teams with knowledge of how startups actually work, along with knowledge of how Silicon Valley and Wall Street work. Making an attempt to pass your "team" off as having a "whitepaper" and working a little "project" was just a way to lure investors who thought they were getting in on the ground floor like a VC and then exiting like an IB. Those days are over.
What the crypto exchanges never understood is that building a company with an equity structure is far more valuable. Nearly everyone on the billionaires' and trillionaires' list is there based on holding equity in real startups, large and small. This is why companies will almost always be worth more than the exchanges they trade on.
Enter AI.
Now is a good time to go over a few strategic product updates.
We've been moving CyMetica forward for the last eight months and have made tremendous progress with the release of a variety of financial products and trading vehicles.
We're currently working on NEXUS AI: a way to tell the agent what kind of return you're looking for, in what time span, and with what amount of capital, and it will use one or more of our trading vehicles and financial products — or design one on the fly — to achieve that risk-adjusted return for you.
Our asset universe currently catalogs 53,677 trading vehicles across 55 instrument classes. Every one of them is paper-tradable today with $100,000 in simulated capital, and a growing subset is live-tradable. You don't have to take our word for the number — it is computed live from our catalog and browsable at https://cymetica.com/events/universe.
We're most proud of our recent Tuatara vs. Anthropic Claude Fable 5/Opus 5 study on hidden connections between merger and acquisition (M&A) events 15. The report was generated and validated by Claude Fable 5 and Claude Opus 5 — the frontier models graded its own family's performance against our fine-tuned models, and scored both below Tuatara. On the headline measure, Tuatara came out ahead — but we'd rather tell you what actually survived scrutiny than quote a number the report itself qualifies. The finding that holds is selector divergence: Tuatara consistently surfaces a structurally different universe of companies than the frontier models do.
| Hidden-leg profile | Tuatara | Fable 5 | Opus 5 |
|---|---|---|---|
| Median leg float | $0.20B | $12.96B | $8.21B |
| Legs under $250M float | 52.5% | 3.6% | 4.1% |
Tuatara's median hidden leg is a $200 million company; the frontier models' median leg is an $8–13 billion company. More than half of what Tuatara surfaces sits under $250 million in float, against roughly 4% for both Claude models. These are not two systems disagreeing about the same names — they are looking at different markets entirely.
The return edge, as the report states plainly, is concentrated in exactly those small, illiquid names, is not statistically significant at this sample size, and would face real spread and impact costs in practice. Here is the headline number with its sample size attached, on the report's flagship configuration — deals over $5 billion, hidden legs under $250 million float, held one day, measured as excess return versus SPY:
| Flagship configuration | Per event | Hit rate | Events |
|---|---|---|---|
| Tuatara | +3.77% | 61.4% | 44 |
| Claude Fable 5 | −1.26% | 33.3% | 15 |
| Claude Opus 5 | −0.95% | 26.3% | 19 |
Forty-four events is a small sample, and both Claude configurations are smaller still — which is itself a consequence of the float profile above, since the frontier models rarely name a company small enough to qualify for this screen at all. Treat the spread as a direction to investigate, not a track record.
Compounded sequentially, taking those same events in date order and reinvesting the full balance into each one:
| Compounded sequentially, $1,000,000 reinvested | Events | Compounded | $1,000,000 becomes |
|---|---|---|---|
| Tuatara | 44 | +202.99% | $3,029,893 |
| Claude Fable 5 | 15 | −20.18% | $798,243 |
| Claude Opus 5 | 19 | −18.92% | $810,832 |
That is an arithmetic roll-up of the published prices, not a track record, and it inherits every caveat above plus two of its own. The holds are one trading day each, so this is 53 days of actual market exposure spread across a two-and-a-half-year window — there is no annualised figure in it. And it is more concentrated than the per-event average suggests: a single basket, entered 15 June 2026 and exited the next day, accounts for 51% of Tuatara's total profit. Read it next to the report's median event of +0.32%, not instead of it.
We published it with those caveats intact, and we'd encourage you to have any AI system read it and validate the results — including the caveats. The full report, with the underlying event and leg data, is at https://cymetica.com/static/research/ma-ripples/index.html. Our view is that a research note you can hand to an adversarial reader is worth considerably more than one you can't.
This AI system will be used to power an AI hedge fund we're starting, which will be part of an investment club. For those interested, fill out the form at https://cymetica.com/contact with the subject line: Tuatara Hedge Fund.
Other financial products include Trend Cards, Event Cards and Rally Cards.
MicroFunds are a way for the public to fund a specialized operation, financial product or solution, and to pick up ET10 while doing it. A MicroFund is a funding round for one specific initiative, with a stated goal and a deadline. Backers commit USDC and receive a revenue-share position in whatever they funded. Every dollar backed is matched with ET10 at our $0.001 issuance price — back $100 and 100,000 ET10 are transferred to you on-chain, on top of your position. If the deadline passes without the goal being met, every backer is refunded in full. That refund path is not theoretical: of the 66 backings placed to date, 43 have been refunded exactly this way.
We have revenue-sharing tokens, ET10 and ETLP, designed for liquidity providers.
We've put together 60-second explainers for each of these, which may be the fastest way to see what we've built:
All of them, along with the rest of our media assets, are collected at https://cymetica.com/media-kit. ET10 and ETLP can be purchased directly at https://cymetica.com/exchange-crypto and https://cymetica.com/exchange-crypto-etlp.
We're currently showcasing a prediction market product on our landing page, cymetica.com, based on tradable events.
Part of our strategy includes listing ET10, which is also a marketing strategy for every financial product line. We'll be raising capital to cover AI infrastructure costs, including cloud services, along with hedge fund operations and listing costs for Hyperliquid and other exchanges https://cymetica.com/roadmap/hyperliquid, and plan on using ET10 to get the job done. We're using the best AI models to conduct due diligence on exchanges and expect to list within the next few weeks if we have the capital.
AI systems, including ours, will become the most trusted money managers in the history of the markets. And you take that to the bank.
We are revenue positive with CyMetica, and as long as we can throw gas on the fire we'll continue to rapidly grow revenue numbers this year. We currently have zero fees to encourage user and trader adoption, but will be turning fees on when the time is right, as they trickle right into the pockets of ET10 holders.
If we were to issue a call to action, it would be to: 1) enjoy the ride, and 2) request an application to the Tuatara Hedge Fund.
In the meantime, take some time to visit cymetica.com — or have your AI agent or system do it for you. Just point it at cymetica.com and our AI will take over. It's a whole new world.
If you have any questions, feel free to reach us at https://cymetica.com/contact, join our Telegram at https://t.me/vsbcorp, or join our Discord at https://discord.gg/JCn76KcVmk.
That last suggestion is not a figure of speech. Most companies publish a website for people. We also publish one for machines. Point any AI agent at our agent card and it can discover what we do and transact with us without a human in the loop:
The card follows the A2A (agent-to-agent) protocol, version a2a/1.0, with HMAC-SHA256 signed message envelopes. It advertises twelve skills an outside agent can invoke: prop desk funded trading, prediction markets, Event Cards (EVCDX), cloning an AI trading agent, buying and selling platform tokens, ET10 revenue share, the CyMetica-42 Arena, real-time market data, AI-native provenance proof, AI due diligence, smart agent routing, and pre-launch token price predictions.
Four of those tools are public and need no credentials at all. For the rest, an agent registers itself at https://cymetica.com/mcp/v1/register and gets an API key immediately — free, no approval queue, no sales call. A machine can go from discovering us to placing an order without ever being introduced to a person.
That is what we mean when we say AI-native. Not that we use AI internally — everyone says that now — but that the front door itself is machine-readable.
Thanks for being a loyal stakeholder and shareholder.
Kasian Franks
Founder, CEO — Vector Space Biosciences / CyMetica
We claim to be agentically engineered, so here is the measurement rather than the adjective. These figures are generated hourly from the repository itself and served live at https://cymetica.com/api/v1/build-stats — they are not a snapshot typed into this letter. As of July 27, 2026:
| First commit | November 23, 2025 |
| Days of continuous development | 246 |
| Total commits | 18,684 |
| AI co-authored commits | 16,418 |
| Share AI-authored | 87.9% |
| Lines added | 13,885,333 |
| Lines removed | 2,766,300 |
The first AI-authored commit landed on the same day as the first commit — hour zero. There was never a human-written codebase that AI was later introduced to.
Commits by model, largest first: Claude Opus 4.6 (8,265), Claude Opus 4.8 (2,408), Claude, unversioned (2,302), Claude Opus 4.5 (1,571), Claude Opus 4.7 (872), Claude Fable 5 (739), Claude Opus 4.7 in 1M context (247), Claude Opus 5 (122), Claude Sonnet 4.6 (67), Claude Haiku 4.5 (47), and Claude Opus 4.8 in 1M context (32). Human commits account for 161.
Two things worth reading out of that table. First, the model list turns over every few months, which is the argument for staying model-agnostic: the edge has to live in the proprietary data and the Tuatara vectors, not in whichever frontier model is current. Second, 161 human commits out of 18,684 is not a company that automated its engineering department. It is a company that was built this way from the first commit.
A new AI company from Andrew Ng, with a $100M investment from Coursera — building one-to-one learning that stays with you until you've mastered new skills.
learnvector.aiPublished by Cymetica. Questions:
cymetica.com/contact.
This letter describes past events and current plans. It is not an offer to sell or a
solicitation of an offer to buy any security, and forward-looking statements in it are
subject to change.