The week in three numbers: Aug 2 EU AI transparency rules take effect (not delayed to 2027) · 6 tools AI coding assistants whose "approve" box can be tricked · ~1% of S&P 500 firms could quantify an AI earnings impact (Goldman's Q4 2025 review).

This week's question: who actually keeps the money AI creates? For each item, the event is the fact; the money read is our analysis — labeled separately.

Our standing scorecard: named companies' AI value claims vs. what shows in the P&L, scored on one evidence ladder. None of the seven companies currently tracked has reached Stage 4. Until one does, we won't treat a public AI-value claim as evidence of a durable moat without additional proof.

The evidence ladder: 0 · Narrative (management mentions AI) · 1 · Operational (a quantified activity/productivity number) · 2 · Financial claim (a dollar saving or revenue figure asserted) · 3 · P&L-attributed (the AI contribution isolated in reported results) · 4 · Sustained (that attribution holds four straight quarters — the moat bar).

S&P Global and Ecolab are the two firms Goldman flagged as already quantifying an AI earnings impact — our most important Stage 3 cases to verify in primary filings. Next update: the August cluster (Klarna Aug 18) and the ServiceNow/BofA prints.

Why might AI create so much value and let so little reach earnings? The cleanest lens is 40 years old. The economist David Teece (1986) showed that a company keeps the profit from a breakthrough only when it's hard to copy — or when it's tied to a scarce asset the company already controls: proprietary data, distribution, an installed base, regulatory approval, customer trust. High-performing AI models are increasingly available from several providers, so the profit tends to flow past the model to whoever owns the scarce thing around it.

Read the week through that lens. The coding-tool and Hugging Face breaches suggest trusted, audited deployment is becoming scarce. The EU rule may hand an edge to firms with compliance capacity. And on the value chain, near-term revenue is clearest for the chip, networking, power and data-center suppliers; the hyperscalers sit in the middle — earning cloud revenue but also carrying much of the capex and depreciation risk. Hold one honest nuance (Brynjolfsson's productivity J-curve): some missing return may be real value still buried in reorganizations, not yet visible in the P&L. Either way, the board move rhymes: don't bank on AI widening your margins without naming the scarce asset that lets you keep the benefit — and the operating evidence that it's working.

Mostly "anyone can get it" → you're buying table stakes; budget it as a cost. Mostly "only you, tied to your business" → that's where a real advantage can form; fund that, not just the AI.

The live split between serious people, and where we come down.

[How we label evidence: Primary source · Corroborated · Reported · Vendor claim · Analysis.] · Written & edited by Mario Suarez · Independent analysis · Sent Sunday evenings. Every link and date verified before send.

AI Above the Cut is a weekly brief for executives — VP-and-up leaders in strategy, healthcare, and AI transformation who want signal over noise. Each Sunday we read a fixed spine of the field's highest-signal voices — operators, researchers, and independent skeptics like Andrew Ng, Ethan Mollick, Simon Willison, Nathan Lambert, the AI Snake Oil team, Erik Brynjolfsson, Cassie Kozyrkov, and Eric Topol — plus a rotating edge of specialists (Chip Huyen, Jack Clark, Ben Thompson, Robert Wachter, and others) and the primary research, regulator, and lab feeds. We tag every source — vendor, researcher, operator, investor, regulator, economist, or skeptic — and check strong claims across categories, so we curate evidence, implementation, and disagreement rather than celebrity.

The brief comes in two speeds: a fast skim — the single most important development, three concrete moves, and the week's decision-relevant signals — then a longer Synthesis that connects them, takes a position, and links to the primary work. We optimize for quality over influence, link to the source (the paper, the post, the talk) rather than the hype around it, and flag anything unconfirmed. No "10 AI tools you need today."