Our Thesis

we are building the autonomous research agent for the physical economy's decision layer.

as a team of engineers who have worked extensively on making capital markets more efficient, we bring the lens of probability to physical industry. it is only fitting that some of the most traded commodities, oil and gas, are our vertical beachhead. we attack the way determinations have been made in physical industries for decades–a messy, political pseudoscience with unthinkably low efficiency. with pure empiricism, axiomatic thinking, and careful analyses of confidence and uncertainty, we are turning this process systematic.

the physical economy is littered with decisions that were obviously wrong in retrospect. wells drilled because a single interpretation became consensus. mines acquired because the upside case was easier to narrate than the uncertainty was to quantify. infrastructure projects greenlit because risk lived in disconnected memos instead of one shared probabilistic model. in each case, the failure was not a lack of data. the data existed somewhere: in a report, a map, a field note, a lab result, an old consultant deck, a sensor stream, or an expert's head. the failure was that the evidence never became a system.

in 1998, kent bowker walked into the break room at mitchell energy headquarters to buy a coke. bowker had just come over from chevron and had been assigned to the barnett shale, the rock formation that most of the industry still thought was a waste of capital. he saw bill stevens, mitchell energy's president and a former exxon executive, and introduced himself. bowker started telling him he was excited about the barnett. stevens cut him off. according to bowker, stevens physically put his hand in his face and said, "stop right there." the message was not subtle: the company had enough barnett shale; bowker should find something else to work on. stevens had called much of the acreage "moose-pasture land," and he reportedly went on cnbc calling the barnett a "black tombstone." eight competitors did the exact same, and the market consensus was that the fair value of the barnett shale was near zero. the market was political, inefficient, and misallocating capital.

then bowker ran the experiment anyway. he measured the gas in place and found the barnett held almost four times as much gas as the company had believed. at the same time, nick steinsberger was fighting a separate internal battle over slickwater fracking. engineers told him it was a stupid idea; executives thought he was wasting time and money; some wanted him fired. but the economics started to flip: slickwater cut frac costs by hundreds of thousands of dollars per well, and one well began producing at rates the old model said should not happen. mitchell energy had spent roughly $250m over years trying to make the barnett work, but the breakthrough turned marginal acreage into the template for the shale revolution. devon bought mitchell energy in a deal reported around $3.1b in cash and stock, plus debt assumption. the competitors got barely anything from the barnett shale, an asset mispriced so badly that many industry players lost out on 9-figure revenue streams.

this is the hidden tax on physical industry: decisions are made as one-off acts of judgment rather than repeatable, instrumented processes. a team spends months collecting fragments, reconciling contradictions, defending assumptions, and assembling a story that capital can act on. by the time the decision is made, the work is already stale. worse, the reasoning is rarely reusable. the next project starts over, with a new room of experts, a new stack of documents, and a new fight over what the evidence means. physical industry has had world-class experts for decades. what it has lacked is memory, scale, and mathematical discipline.

natural language is the missing layer. the most important evidence in physical industry has never lived cleanly in databases. it lives in pdfs, lab reports, drilling notes, inspection comments, regulatory filings, geological descriptions, emails, maintenance logs, and investment memos. before modern language models, this information was effectively dark matter: visible to humans, invisible to computation. nlp changes that. it lets an agent read the messy evidence layer, extract assumptions, connect entities, compare claims, surface contradictions, and compile unstructured knowledge into structured reasoning. language is not the end product. it is the bridge that makes the physical world machine-readable.

once that evidence becomes machine-readable, the law of large numbers begins to apply. sum probabilities multiplied by their utility, and at scale, the expected values become accurate. a human team can evaluate a handful of opportunities deeply. a machine research agent can evaluate thousands consistently. it does not need perfect certainty on every individual asset to transform outcomes across a portfolio. it only needs to kill bad ideas earlier, surface hidden upside faster, and route human attention toward the few uncertainties that actually change the decision. this is how systematic finance won: not through omniscience, but through disciplined probabilistic reasoning applied across enough bets. we believe the same transition is coming to atoms.

this transition to likelihood-based physical business is inevitable. we won't even sit here and bore you writing about some quant finance underdog story; how the exact same happened in markets. but once efficiency comes to an industry, there will be disruption, and it won't be subtle. the players of the past will not be able to compete without adapting.

we are building the company that makes that transition happen. oil and gas is the first proving ground because the stakes are massive, the data is rich, and the workflow is broken enough for a new system to matter immediately. but the deeper company is not petroleum software. it is the autonomous decision layer for the physical economy: an agent that turns messy evidence into auditable conviction, propagates uncertainty to financial outcomes, and decides what information should be gathered next. today, it compresses appraisal and capital allocation in oil and gas. tomorrow, it becomes the research engine for mining, geothermal, carbon storage, infrastructure, defense, agriculture, manufacturing, and every industry where physical uncertainty determines where capital should flow.