Notes from an AI-native company
Essays on the ideas behind EventTrader: cryptography, prediction markets, and the long line that connects ancient writing to on-chain money and modern AI.
11 essays so far. New ones go to the feed the day they publish.
Subscribe via RSSHow AI connects to Crypto
One unbroken chain runs from the first scratched inscriptions to today's language models and on-chain money. Follow the dots: epigraphy, ciphering, decipherment, cryptography, mathematics, cryptocurrency — and where AI enters every link.
Vector, matrix, tensor — the waffle-stack explainer
The three words at the heart of AI, in one line: a vector is a row, a matrix is a waffle, and a tensor is a syrup-drenched stack of waffles. Same idea, one more axis each time.
Ranking CEOs on real AI expertise, not hype
The Global CEO AI Expertise Index ranks founders and CEOs on a verifiable record in the core of AI — the mathematics, vector embeddings, the transformer lineage — with research weighted 70% and every score re-checkable from its sources. Popularity counts for nothing.
What people are really using AI agents for in crypto
Sorted by how real each use is rather than how loud: trading bots, agent frameworks, research tools, prediction-market agents, payment rails, social personas, and a genuinely new attack surface. The money is rotating away from 'tokenize the agent' toward 'let agents trade' — and a lot of what wears the word 'AI' is a lure, not a product.
Inside the Tuatara Hedge Fund: a live, multi-strategy book
Our fund page shows the whole machine, not a single number: a multi-strategy fund-of-funds powered by the Tuatara vector engine, with M&A Ripple, FTA, and themed AIB sleeves — long and short, each a live price-weighted index updating in real time.
The financial SaaSpocalypse: what the word means, what the evidence says
Where 'SaaSpocalypse' came from, what happened to legal, tax, wealth and financial-data names when AI agents started doing the work software sells, the strongest case against the panic, and why the thin financial tool is the thing that is actually ending. Every figure is sourced.
Where AI goes in the next ten years — with references for the naysayers
A community question, answered with sources: the destination is human biology, not AGI; AI is already outside 'the box'; it will not top out because nobody has even defined intelligence yet; the money is bigger than oil; and nobody is slowing down. Learn to tell when someone is talking up their own book.
Tuatara Decide vs Jev on BANKING77: a tie at a thousandth of the cost, then a win
We scored a Tuatara Vector Model blend against Jev's published answers on all 3,080 BANKING77 test messages: 91.79% versus 92.40%, a statistical tie, at about 14 cents of CPU per million messages against about $142 in API charges. Updated September 28: an embedding head trained on the same messages now scores 94.09%, ahead of Jev (p = 0.00014), at about $5.70 per million. Jev still leads with no training examples. Every input is public, so you can check it.
What to build when AI takes the work
McKinsey says 11 million US workers will need a new occupation by 2035, BCG puts the jobs AI could eliminate at 25 million, the IMF calls it a slow-moving tsunami over 40% of jobs worldwide, and the WEF expects 92 million displaced by 2030. The strongest case against the panic, and why the answer is to use AI to build the trading tool that pays you, with the question to paste into your agent.
What is a forward deployed engineer?
The forward deployed engineer gets a product working inside one customer's business, in production, and makes the next deployment easier. What the job involves end to end, how it differs from pre-sales and post-sales engineering, why AI companies now lead with it, what it pays, and the four roles we are hiring.
Fund Optimizations: a live ledger of how an AI-native hedge fund learns
Every optimization, fine-tuning, training run, calibration and hyperparameter search behind the Tuatara hedge fund's AI system, datestamped and streamed live, with running totals per stage, a race chart, and which work actually lifted returns on held-out data. One in five optimization names is now served as a BlackBox to protect proprietary algorithm state.