Rosetta
Investor briefPresenter mode

AI value is moving above the model. The layer it moves to is unowned.

Models are converging; the durable advantage is shifting to whoever holds the institution's accumulated state — the evidence, decisions, and governance history behind its AI work. Rosetta is the protocol layer for that state: reviewable before action, tamper-evident after it, portable across model families, and owned by the institution.

The gap, in the market's own numbers
88%

AI-agent pilots that never reach production; top blockers are evaluation gaps (64%) and governance friction (57%).

Forrester / Anaconda, 2026
~30%

Organizations with governance maturity adequate for the agents they already deploy.

McKinsey AI Trust Maturity Survey, 2026
$1.4T

Projected global enterprise AI-agent spend by 2027.

IDC / McKinsey, 2026
Dec 2027

EU high-risk AI obligations, now fixed; transparency obligations began Aug 2026. Fines to €35M or 7% of turnover.

EU AI Act & Digital Omnibus, 2026
The market just said it out loud

Palantir's Alex Karp, channeling what he calls the private fury of enterprise CEOs: customers want control of their models, their data, and their edge — to “own the means of production.” CNBC interview, July 2026

Microsoft's Satya Nadella, more quietly: a company that is only a consumer of a foundation model will struggle to retain the learnings — and the enterprise value — created by using it. Stanford remarks, 2026

Both are buyer-side demand for the same missing object: a record of the institution's AI work that it owns, that any model can read, and that no one can quietly rewrite.

The wedge · audit-native cognition

Everyone else records the agent. Rosetta's review is part of the decision.

Twenty-plus vendors sell observability — a camera pointed at the cockpit, a trail you could delete without the agent noticing. Here the Rosetta Review participates in the decision before action; any prior state is reconstructable, and alteration is detectable at the exact entry. The live case answers the buyer's question —show me the trail behind the decision — in sixty seconds.

The Review is the door. The larger asset is governable state: what the system knew, what remained unresolved, what it learned, and which pressures shaped action. What stays hidden cannot be governed.

Run the answer yourself →
What capital builds

A protocol company, on the SOC 2 pattern

Protocol layers are winner-defined markets: the first adopted specification is the one everyone else conforms to. The commercial motion follows the attestation pattern — reference implementation, conformance criteria, certification — with the protocol as the defensible core: patent filings on record, mechanism under NDA, specifications shared with partners under license.

Technical detail — defensibility, precisely

The public site demonstrates behaviors: pre-action review that changes outcomes (the counterfactual), exact reconstruction and tamper-evidence (the record test), and cross-model portability (the handoff, read from the institution's own database). The mechanisms producing those behaviors — representations, mutation discipline, verification model, arbitration — are deliberately absent from this site and constitute the filed IP. Model vendors face a structural conflict in replicating the layer: its core property, portability, dissolves the lock-in their positions depend on; trust layers historically had to be neutral to be adopted. And an intermediary that holds everyone's institutional memory merely relocates the dependency — an intermediary is a new landlord; a protocol is a deed.

The demonstrations are public behavior. The mechanism is filed and under NDA. If the layer thesis is right, the open question is who defines the layer — and protocol windows, historically, are short.

Request the investor brief & diligence sessionPresenter mode

Above board: all market figures are third-party and named on screen — verify them independently. Named executives are cited as buyer-side demand signals, not as endorsements of this product. Demonstrations show behaviors; scenario content is fictional and labeled; the persistence, verification, and handoff machinery is real.