The machines read.
You decide.
We use AI agents only where they earn their keep — and we talk about them only here. One page, the whole account: what they do, what they will never do, and why the last step is always yours.
A formula can only describe where a business has been.
Investing is forward-looking. A model built on historical and current numbers — however clever — is a portrait of the past. Screens are honest about what happened and silent about what happens next; and next is the only thing the price is arguing about.
What the numbers leave out is exactly what matters most: competitive advantage, management quality, industry dynamics, where margins and growth go from here — and the edge cases a screen will always miss.
The numbers are the canopy. The causes are the roots.
Every figure a screen can rank — revenue, margins, return on capital — is the visible crown of something underground: the moat, the management, the competition, the sector's slow weather. Formulas can count the leaves; they cannot see what feeds them. So the agents work underground — tracing which root feeds which branch, and which one is quietly rotting.
The sap rises — and the leaves turn into numbers.
Up from the root tips, through the trunk, out along the branches. Where the flow reaches a leaf, the leaf becomes a figure the record will only confirm later — revenue, margins, return on capital. And the whole current gathers into the two numbers the valuation is still waiting for: growth from here, and margins from here.
Qualitative causes, quantitative outcomes. The tree is the whole argument.
Numbers first. Always.
What receives those two inputs is the foundation: a discounted cash-flow model — owner earnings projected forward, discounted back. Nothing exotic. Most of its inputs come straight from the record. Two do not: where growth goes from here, and where margins settle. Those two assumptions decide almost everything a DCF says — and they are exactly what no backward-looking formula can supply.
This seam is where the agents work — reasoning about what growth and margins are achievable, from the moat, the competition, the management, the sector. Never from the price.
The model states its terms honestly: this is the growth the price requires; this is the growth the qualitative record supports. When required exceeds achievable, the idea waits — no matter how good the story. Reasoning about those two inputs is the agents' entire job. Everything else stays arithmetic.
Three readers, with narrow briefs.
Each agent reasons about one question the screen cannot answer, scores it in the open, and shows its work. They are researchers on retainer — not oracles.
The value read
Is the moat real, is management honest with your capital, and is the price actually demanding too much?
The growth read
Where do margins and growth go from here — the two inputs the DCF is waiting for. How much runway is left, and who is taking share?
The edge-case read
The setups a screen always misses — catalysts, dislocations, the odd quarter where price and reality diverge.
They read all of it. So you don't have to.
Every filing, every footnote, every line of market data — read overnight, in the same quiet window where the briefing is written. Each run leaves a trace: which sources were consulted, in what order, for how long.
The reading is auditable too. If a claim appears in a read, the tool that fetched it is on the record.
A read you can audit — never a verdict.
What comes back is reasoning you can inspect, not a conclusion you must trust. New scores sit beside last quarter's scores. Changes are named in plain sentences. Patterns are tracked across runs, with confidence attached. If a read can't show its work, it doesn't ship.
What changed: membership renewal ticked up; leverage came down. The valuation improves on mix, not on a re-rated multiple.
“The moat is unchanged — scale economies shared with the member. What moved is the balance sheet: net debt is now covered by fourteen months of free cash flow. The multiple still assumes more than the record proves; the entry discipline holds.”
“You are neither right nor wrong because the crowd disagrees with you. You are right because your data and your reasoning are right.”Benjamin Graham
The last step never moves.
The agents assemble the file. You write the thesis and the kill criteria — in your own words, while you are calm. From then on, the briefing tests each day's facts against your sentences, not theirs. The circle behind this page is drawn in one stroke and left open — the machines bring it this far; the closing is yours.
And we mention the machines exactly once — on this page. Never on the briefing. There, a read that matters arrives as one plain sentence, and the machinery stays in the engine room.
The machines stay in the engine room.
The judgment stays with you.
That is the whole arrangement — and the only time we'll bring it up.