AI has made producing answers inexpensive. The greater economic opportunity is making institutions more capable.
From AI productivity to institutional performance.
AI is making it cheaper to produce answers. Institutions still bear the cost of establishing what is true, what can be done, whose interests and obligations matter, and how work must be coordinated.
Across law, insurance, healthcare, finance, enterprise management, and government, this is where consequential value is at stake. Runcible AI examines what can be warranted. Runcible Oversing connects that knowledge to people, authority, action, and consequences.
AI is making people more productive. Why aren’t institutions gaining as much?
McKinsey Global AI Survey, August 2026 (1,719 respondents). Different self-reported outcomes; not a failure rate. Source and methodology.
McKinsey’s 2026 survey found that 80% of respondents reported better individual productivity from AI, while 37% attributed a positive enterprise earnings impact to it. These are different measures, not a project failure rate—but the gap identifies a consequential unsolved problem.
AI can produce code, contracts, claims, strategies, and recommendations quickly. An institution must still establish whether the premises correspond to facts, what obligations and costs follow, whose interests are affected, who has authority, and how people coordinate the resulting work.
Generating an answer is not the same as qualifying it, resolving the problem, or carrying the result into institutional action.
Source: McKinsey, The State of AI 2026. BCG also finds enterprise AI value is increasing; this is an execution opportunity, not a claim that AI has no return.
Courts expose the problem. Every major institution faces a version of it.
Courts have documented fabricated legal authorities and unsupported AI-generated material. A convincing argument is not an existing precedent; a precedent is not necessarily controlling authority; and an applicable rule does not establish the facts of a case.
Those same qualification burdens recur in insurance claims, healthcare administration, financial compliance, contracts, public procurement, government determinations, and enterprise management. In every case, the institution must establish what holds, reconcile conflicts and obligations, and coordinate a defensible response.
The opportunity is not simply producing more AI content. It is reducing the recurring cost of knowledge qualification, cooperation, and institutional problem-solving.
The AI Hallucination Cases Database documents court-reported cases; it is not a rate across all legal filings. Explore the institutional AI opportunity →
The Other Half of AI
Generating possibilities is not the same as establishing which survive.
Foundation models have made generating hypotheses, plans, and explanations inexpensive. But a plausible answer is not proof that its claims can actually be constructed from the operations and constraints of reality.
Think of geometry: a valid construction must follow from permitted elements and operations. Runcible is designed to apply this discipline to language: deriving the necessary examinations, testing what survives, and identifying what remains unresolved.
Foundation models compute. Runcible qualifies. Oversing operates.
Two complementary Runcible offerings build on existing foundation models. Information moves upward into qualified judgment and action; observations and consequences return to inform further examination.
Runcible Oversing
Institutional operation · Authority · Action · Memory
Runcible AI
Epistemic qualification · Tests · Recursive closure
Foundation models
Semantic computation · Interpretation · Inference · Generation
Connect decisions to accountable work.
From AI productivity to institutional performance.
For enterprise and government leaders who need to know what a recommendation rests on, who may act on it, and what happens next.
Runcible AI examines information and develops supported options for resolution. Runcible Oversing connects that problem solving to the people, authority, assignments, and outcomes of the institution.
Examine the recommendation. Coordinate what follows.
Inspect the grounds before deciding.
Runcible AI curates information, produces a proof with findings and limits, explains the result, develops options for resolution or settlement, and recommends an action. Its protocols are designed to require the relevant examinations and distinguish supported findings from unresolved questions.
You can inspect what the recommendation rests on and where its limits remain. Runcible AI can also integrate into other software independently of Oversing.
Explore Runcible AI →An operating system for institutional cooperation.
Connect the problem, participants, obligations, decisions, assignments, and outcomes in a continuing institutional record. Leaders can trace work to its purpose; contributors can understand their responsibilities; later reviewers can recover the grounds and history.
People act within assigned authority. Any delegation to AI requires an identified scope and permissions.
Explore Runcible Oversing →Follow one problem through the institution.
A compensation proposal makes the sequence concrete. The AI analysis below was recorded using fictional facts and supplied agreements; the Oversing decision, work, and review steps are a proposed continuation.
Identify the proposed change.
Management wants fairer compensation across unequal territories and proposes pooling commissions already earned.
Examine the obligations.
Runcible AI distinguishes earned payments, future terms, and voluntary contributions, then develops alternatives under the supplied agreements.
Establish authority to proceed.
Participants would inspect the findings, develop terms, and obtain the authorization required for the chosen course.
Assign the implementation.
Oversing would connect the agreed terms and responsibilities to the administrative work and its dependencies.
Retain what happened.
The decision, commitments, work, and observed result would remain available for later review and correction.
One underlying capability. Different institutional opportunities.
Extend models beyond assistants.
Explore whether the semantic capabilities of existing models can be made more dependable through external protocol execution and qualification—without retraining the foundation model.
For foundation-model partners →Connect decisions across teams.
Evaluate how the products connect an operating proposal to its evidence, responsible people, authorized work, and recorded outcome.
For enterprises →Make decisions reviewable.
Keep evidence, procedures, responsibilities, authority, decisions, and consequences connected so that judgments and actions can be examined and contested.
For government partners →Infrastructure for institutional AI.
As generating candidate work becomes less expensive, qualification and accountable execution become increasingly consequential economic functions.
For investors →The next step isn’t another assistant. It’s making knowledge construction executable.
We use language to identify objects, distinguish differences, describe actions, reconstruct causes, and communicate consequences. Mathematics, formal logic, and programming offer powerful methods where their required representations can be sufficiently specified. But real-world innovation and adaptation often demand that we preserve causal particulars those narrower representations cannot yet economically capture.
Our Natural Law research treats language as a system of measurement and develops a constructive method for examining empirical claims through identifiable operations, evidence, and tests. Foundation models create an opportunity to apply such examinations at computational scale.
Runcible’s ambition is to make that method executable—not merely instruct a model to sound careful.
The technical paper on constructibility from first principles specifies the operational hierarchy, derived test obligations, evidence requirements, and falsifiable experiments. The earlier Other Half of AI section introduces the distinction; this research explains the computational method beneath it.
Working software. Continuing engineering and scientific validation.
Runcible AI and Oversing are operating, early-stage software systems. Current work is strengthening non-bypassable protocol execution, recursive examination of closure, evidence qualification, and authorization of adjudication. Enterprise readiness and comparative reliability remain subject to further testing.
We distinguish implemented functionality from independently demonstrated performance. Our immediate research question is how reliably existing foundation models can execute a general constructive epistemic methodology under external control.
Start with one consequential problem.
Bring a recurring decision, agreement, or proposed change. Identify the people involved and the result you need to evaluate. A walkthrough can establish what the current products do, how they would fit your work, and what further implementation is required.
Technical evaluators, enterprise and government teams, and prospective partners are welcome.
Go deeper.
The homepage is the introduction. Explore the products, architecture, applications, research, and people behind the work.