01 · hypotheses.md
Hypotheses in writing
Every thesis is stated in writing before it is tested, and every position is traceable to an argument a human can read.
AI hedge fund — in formation
The AI hedge fund that trades in public.
Markov AI is building a hedge fund staffed entirely by reasoning agents — deep researchers, analysts, and monitors working every market, every hour. Every thesis published, every result verifiable.
Staffed by machines.
Markov AI is organized like an investment firm, staffed by machines — a research organization that never closes.
Deep-research agents take the hard questions and reason for hours at a time: reading filings, replicating published results, writing theses in full.
Monitoring agents hold every position and every relevant market in view through every session, in every time zone.
Research teams review, criticize, and reject one another’s work before anything reaches the book.
The organization runs twenty-four hours a day. Its capacity is measured in compute, not headcount.
The best trader that can be built.
Frontier reasoning models now match the analytical depth of a well-trained doctoral researcher — and unlike researchers, they can be replicated without limit.
Each month’s literature makes them stronger; we implement what survives our review. For the first time, a firm’s research capacity is bounded by compute and method rather than by hiring.
We intend to build the best trader that can be built, and then run it at scale.
Held to the standards of science.
02 · controls.md
Measured against controls
Results are measured against controls. Conclusions are held provisionally, and revised when the evidence moves.
03 · risk-limits.md
Risk enforced in code
Risk limits are enforced in code before any signal reaches the market. The first job is to survive every state.
We build in the open.
status: pre-launch — no fund open · no AUM · our own capital only
Markov AI is at an early stage, and we have chosen to build it in the open. We are bringing our agent teams up against live market data.
Every thesis the organization acts on will be published in full, with its outcome — whether we are right or wrong.
No fund is open and no capital is under management. We would rather show a track record than describe one.
Illustrative simulated log of agent research activity. For concept demonstration only — not live trading.
simulation — illustrative research concepts only. No live trading, no real orders, no client capital.
In time, the desk works for you.
In time, the same organization will work for others.
If you would put a research desk of your own on the markets you care about, leave your name — early access will be small.
For future partners.
The fund, when it opens, will be available to qualified investors. We are not raising today.
The people on this list will be the first to read our results — and the first to hear when that changes.
This is not an offer to sell securities or a solicitation of an offer to buy securities.
Founding roles. No ladder.
We are hiring two or three founding researchers and engineers to direct the machine organization: people who read the literature for pleasure, hold their own work to publication standard, and would rather run a thousand experiments than attend a thousand meetings.
Write to us with something you have built.
- founding-researcher/ # markets, ML, or both
- founding-engineer/ # systems that were not allowed to fail
- ladder/ — not found
Get the first letter.
When there is something worth reading — a published thesis, a result, an open fund — it goes to this list first.
One letter when it matters. No noise.