Position Paper: Not Another Taxi — AIGP as a Two-Sided Governance Economy
Position Paper: Not Another Taxi — AIGP as a Two-Sided Governance Economy
PRIVATE AND PROPRIETARY. Owned by Kanjani AI Research & Causum. See NOTICE.md.
Status: Position Paper Author: Kanjani AI Research & Causum Date: July 2026
Thesis
Every generation of enterprise software eventually spawns a “governance solution” that is really just another bespoke SaaS product — another taxi company with its own fleet, its own routes, and its own medallion system. AI governance is no different: the market is filling with point solutions that each solve one concern, for one framework, in one vendor’s proprietary format.
AIGP is not another taxi.
AIGP is the ride-sharing platform — the infrastructure that connects those who need AI governance (Emitters) with those who have specialized expertise in particular concern areas (Receivers). It does not compete with governance frameworks. It makes them composable, distributable, and economically viable at scale.
The taxi company owns the cars, the routes, and the drivers. The platform owns the matching function — and that’s what scales.
The Problem with Taxis
Every Bespoke Governance Tool Is a Taxi Company
Consider what a typical AI governance SaaS product does:
- It defines its own concern model (its “routes”).
- It builds its own detection and evaluation logic (its “fleet”).
- It hires or trains its own domain specialists (its “drivers”).
- It sells access to enterprises on a subscription basis (its “fares”).
- It covers a limited geography — one regulatory regime, one concern class, one deployment pattern (its “service area”).
This works for exactly one scenario: the customer whose needs perfectly match the taxi company’s routes. For everyone else, you call another taxi. And another. And another.
The result is the AI governance landscape as it exists today:
- Fragmented. Fairness tools don’t talk to privacy tools. Safety tools don’t talk to compliance tools. Each operates in isolation.
- Duplicated. Every vendor reinvents evidence collection, storage, and querying — because there is no shared substrate.
- Incomparable. Results from one tool cannot be compared to results from another, even when they assess the same system for the same concern.
- Non-composable. An enterprise cannot combine “best-of-breed” without bespoke integration for each pair.
- Geographically limited. Most tools know one regulatory regime well. Multinational governance requires stitching together incompatible systems.
This is the taxi problem: captive supply, limited geography, no interoperability, and no network effects.
The Uber Insight
Uber did not build a better taxi. It built something structurally different:
| Taxi Company | Ride-Sharing Platform |
|---|---|
| Owns the vehicles | Connects vehicle owners to riders |
| Employs the drivers | Enables independent operators |
| Fixed routes and zones | Matches dynamically by need |
| Revenue from fares | Revenue from facilitation |
| Quality by hiring standards | Quality by ratings and evidence |
| Scales by buying more cars | Scales by attracting more participants |
The insight was not “taxis are bad.” The insight was: the matching function is more valuable than the fleet.
AIGP as the Platform
AIGP applies the same structural insight to AI governance:
| Bespoke Governance SaaS | AIGP Protocol + Platform |
|---|---|
| Builds its own concern model | Hosts any concern model as a Dialect (RFC-038) |
| Owns the evaluation logic | Enables specialized Receivers to provide evaluation |
| Limited to its concern areas | Open to any Domain of Concern (RFC-034) |
| Proprietary evidence format | Open evidence schema — interoperable by design |
| One vendor’s geography | Jurisdictional governance across any regulatory regime |
| Scales by hiring more engineers | Scales by attracting more concern specialists |
The Two Sides
Emitters — Those Who Need AI Governance
Emitters are organizations that deploy, operate, or consume AI systems and need those systems governed. They are the “riders” in this model:
- Enterprise AI teams deploying agents into production
- SaaS vendors with AI features that enterprise customers demand governance over
- Regulated industries (healthcare, finance, defense) with mandatory compliance obligations
- Government agencies accountable for AI decisions affecting citizens
Emitters don’t want to become governance experts. They want to emit evidence from their AI systems and have that evidence consumed, evaluated, and verdicted by parties who specialize in their specific concern areas. They want governance the way a rider wants a ride — on demand, by qualified parties, without owning the car.
Receivers — Those with Specialized Concern Expertise
Receivers are parties with deep, specialized knowledge in particular governance concern areas. They are the “drivers” in this model:
- Fairness and bias specialists who know how to measure representational and allocative harm
- Privacy engineers who understand differential privacy, data minimization, and consent models
- Safety researchers who can evaluate autonomous system boundary conditions
- Regulatory experts who know the EU AI Act, Japan AI Promotion Act, or AU Continental Strategy inside and out
- Domain specialists — medical AI evaluators, financial risk assessors, military ethics reviewers
- Cybersecurity analysts who evaluate adversarial robustness and model supply chain integrity
- Cognitive harm researchers who assess dependency patterns and human deskilling
Receivers don’t want to build a complete governance platform. They want to package their expertise as a consumable, subscribable, versionable artifact — a Dialect — and have it distributed to the emitters who need it.
Why This Isn’t Just a Metaphor
The two-sided economy model is not a marketing analogy. It is structural to how AIGP is built:
1. The Protocol Is the Matching Function
AIGP’s core protocol (REGISTER, CHECK, RECORD, TRACE, ANTICIPATE, VERIFY) is the shared substrate that connects emitters to receivers. It is not the governance itself — it is the interface that makes governance composable.
An emitter emits evidence in a standard format. A receiver consumes that evidence through a standard interface. The protocol matches them — not by picking winners, but by defining the interaction contract.
2. Dialects Are the Supply (RFC-038)
A Dialect is the distributable, versionable, subscribable expertise package that a Receiver publishes:
Dialect = Domain of Concern (what class of harm) + Mediation Observation Model (what to measure) + Calculation Semantics (how to calculate posture) + Observer Requirements (who is qualified to verdict) + Default Thresholds (what levels trigger action)This is the Receiver’s “vehicle” — their expertise, packaged for distribution. They publish it to the Dialect Registry (the “marketplace”), and any Emitter can subscribe.
3. Evidence Is the Currency
In the taxi model, the currency is the fare. In AIGP, the currency is governance evidence — the stream of structured, cryptographically signed records that flow from Emitter to Receiver:
- The Emitter produces evidence (analogous to the rider providing the trip data).
- The Receiver evaluates that evidence against their Dialect’s measurement apparatus.
- The verdict flows back — posture, compliance, risk, trust level.
- Both parties have an auditable record of the exchange.
4. The Platform Enables but Does Not Constrain
AIGP does not tell Receivers what concern models to build. It does not tell Emitters which Dialects to subscribe to. It provides:
- Discovery: What governance Dialects exist for my concern class?
- Subscription: I want my AI systems evaluated against this Dialect.
- Distribution: All my governed systems receive the same measurement apparatus.
- Versioning: When the Dialect evolves, I can upgrade deliberately.
- Compatibility: Different Dialects can coexist without conflict.
This is the platform function: enabling a market, not owning the supply.
What This Makes Possible
Scenario 1: The Multinational Enterprise
A company deploying AI across the EU, Japan, and Australia does not need three separate governance vendors. They subscribe to:
eu_ai_act@3.1.0— maintained by European regulatory specialistsjapan_ai_promotion@1.4.0— maintained by Japanese compliance expertsau_continental_strategy@2.0.0— maintained by AU governance analysts
All three Dialects consume the same evidence stream from the same emitting AI systems. No duplication. No incompatibility. No bespoke integration.
Scenario 2: The Specialized Concern Expert
A research group specializing in cognitive harm (AI-driven deskilling, dependency patterns) packages their measurement apparatus as:
cognitive_harm@1.0.0— a complete Dialect with variable definitions, calculation semantics, and observer requirements
They publish it to the registry. Any organization concerned about AI-driven cognitive atrophy can subscribe — without the research group building a full SaaS platform, without the enterprise hiring cognitive harm researchers.
The expert’s expertise scales beyond their consulting capacity. The enterprise gets specialist governance without building in-house. The protocol makes both possible.
Scenario 3: The Defense Organization
A military AI program needs governance that no commercial vendor offers — governance for autonomous systems operating under International Humanitarian Law. They either:
- Wait for a SaaS vendor to build this (years), or
- Build their own from scratch (expensive, non-portable), or
- Subscribe to
ihl_governance@2.3.0maintained by IHL-specialist Receivers on the AIGP platform.
The Dialect bundles the measurement apparatus. The platform distributes it. The emitting AI systems produce evidence against it. No bespoke tool required.
Scenario 4: The Framework Author
NIST publishes the AI RMF. Today, every governance vendor interprets it independently, incompatibly. With AIGP:
NIST (or an authorized party) publishes nist_ai_rmf@2.0.0 as an AIGP Dialect. Every organization that subscribes is using the same measurement apparatus — the same variables, the same calculation semantics, the same thresholds. Cross-organizational comparison becomes meaningful.
The framework author becomes a Receiver — packaging their expertise for consumption rather than hoping vendors interpret it correctly.
Why Bespoke SaaS Cannot Do This
The taxi company cannot become a platform by adding more cars. The structural limitations are inherent:
| Limitation | Why It’s Structural |
|---|---|
| Single-vendor concern model | A SaaS tool embeds one team’s understanding of one concern area. To cover new concerns, it must hire new experts and build new logic. This is linear scaling — more cars, not more participants. |
| Proprietary evidence format | If your evidence only works inside one tool, you are locked to that tool. The taxi determines your route. |
| No third-party contribution | A SaaS vendor cannot distribute expertise they don’t own. They are a closed fleet, not an open marketplace. |
| No composability | You cannot combine two SaaS tools’ governance without building custom integration. Two taxi companies cannot share a ride. |
| Geography is additive, not inherent | Each new regulatory regime requires the vendor to build support. On a platform, a new regime is a new Dialect published by a specialist. |
The platform model does not compete with taxi companies. It makes them unnecessary — because the expertise that was trapped inside vendor organizations is now distributable, composable, and subscribable by anyone.
The Network Effect
Platforms exhibit network effects that point solutions cannot:
- More Receivers → better governance options for Emitters → more Emitters join
- More Emitters → larger market for Receiver expertise → more Receivers publish Dialects
- More Dialects → broader concern coverage → broader adoption
- Broader adoption → cross-organizational comparison becomes meaningful → regulatory endorsement
- Regulatory endorsement → mandatory participation → universal network
This is the cycle that no individual SaaS tool can trigger — because point solutions do not create markets. They fill slots.
The Economic Model
Emitter Side (Free to Emit)
Implementing the protocol and emitting governance evidence is free. The specification is open. The SDKs are source-available. The barrier to entry is two JSON files and fifteen minutes.
This is deliberate: adoption requires zero friction at the emitter tier. You don’t charge the rider to get in the car.
Receiver Side (Commercial Intelligence)
The commercial value is on the consumption side — the intelligence that interprets, evaluates, and verdicts the evidence:
- Dialect creation tools — the instruments that package expertise
- Scoring and posture engines — the math that turns evidence into decisions
- Certification programs — the trust marks that signal quality
- Registry hosting — the marketplace infrastructure
This is where the platform earns: not by owning the expertise, but by enabling its distribution.
Why This Works
The Uber driver pays a commission, not the rider (in economic terms, Uber subsidizes demand and monetizes supply). Similarly:
- The Emitter pays nothing to produce evidence (subsidized demand drives adoption).
- The Receiver’s expertise is monetized through the platform’s distribution infrastructure.
- The platform earns from facilitation — not from owning the governance logic itself.
What AIGP Is Not
| Not This | But This |
|---|---|
| Another governance SaaS tool | The protocol that makes governance tools composable |
| A competitor to NIST, ISO, EU AI Act | The substrate that makes those frameworks distributable |
| A replacement for domain experts | The marketplace that connects domain experts to demand |
| A single concern model | A registry that hosts any concern model |
| A proprietary lock-in | An open protocol with commercial intelligence on top |
The Title, Explained
“Not another taxi” because:
-
The world does not need another AI governance point solution. It needs the infrastructure that makes governance expertise distributable, composable, and economically viable.
-
Taxis solved the wrong problem. They optimized vehicle ownership. Ride-sharing optimized the matching function. Similarly, bespoke governance tools optimize their own logic. AIGP optimizes the connection between governance need and governance expertise.
-
Platform economics create value that point solutions cannot. Network effects, cross-organizational comparability, specialist participation without full-stack investment — these emerge from platforms, not products.
-
The expertise already exists. The world is full of brilliant fairness researchers, privacy engineers, safety specialists, and regulatory experts. They are trapped inside consulting firms, research labs, and vendor organizations. AIGP gives them a distribution channel that scales beyond their headcount.
Conclusion
The AI governance market is following the same pattern as every previous enterprise software market: bespoke tools → consolidation → standardization → platform. Most participants are still building taxis — specialized, proprietary, non-composable point solutions.
AIGP skips the middle steps. It is built from the ground up as a two-sided economy — not because “platform” is a fashionable word, but because the structural economics of governance demand it:
- Governance expertise is fragmented across thousands of specialists.
- Governance demand is fragmented across thousands of organizations.
- No single vendor can own all the expertise.
- No single enterprise can hire all the specialists.
- The only structure that connects fragmented supply to fragmented demand at scale is a platform.
The protocol is the matching function. Dialects are the supply. Evidence is the currency. The platform enables both sides without constraining either.
Not another taxi. The infrastructure that makes taxis unnecessary.
Build the roads, not the fleet.
Appendix: The Structural Comparison
┌─────────────────────────────────────────────────────────────────────────┐│ BESPOKE SaaS MODEL (Taxi) ││ ││ ┌──────────┐ ┌──────────┐ ┌──────────┐ ┌──────────┐ ││ │ Vendor A │ │ Vendor B │ │ Vendor C │ │ Vendor D │ ││ │ Fairness │ │ Privacy │ │ Safety │ │ EU AI Act│ ││ │ │ │ │ │ │ │ │ ││ │ Own model│ │ Own model│ │ Own model│ │ Own model│ ││ │ Own fmt │ │ Own fmt │ │ Own fmt │ │ Own fmt │ ││ │ Own eval │ │ Own eval │ │ Own eval │ │ Own eval │ ││ └─────┬────┘ └─────┬────┘ └────┬─────┘ └─────┬────┘ ││ │ │ │ │ ││ ▼ ▼ ▼ ▼ ││ ┌─────────────────────────────────────────────────────────────┐ ││ │ ENTERPRISE (must integrate each) │ ││ │ 4 vendors × custom integration × incompatible evidence │ ││ └─────────────────────────────────────────────────────────────┘ │└─────────────────────────────────────────────────────────────────────────┘
┌─────────────────────────────────────────────────────────────────────────┐│ AIGP PLATFORM MODEL (Uber) ││ ││ RECEIVERS (Specialized Concern Experts) ││ ┌──────────┐ ┌──────────┐ ┌──────────┐ ┌──────────┐ ┌─────┐ ││ │Fairness │ │Privacy │ │Safety │ │EU AI Act │ │ ... │ ││ │Specialist│ │Engineer │ │Researcher│ │Expert │ │ │ ││ └─────┬────┘ └─────┬────┘ └────┬─────┘ └─────┬────┘ └──┬──┘ ││ │ │ │ │ │ ││ ▼ ▼ ▼ ▼ ▼ ││ ┌─────────────────────────────────────────────────────────────┐ ││ │ AIGP DIALECT REGISTRY (Marketplace) │ ││ │ fairness@2.0 privacy@1.4 safety@3.1 eu_ai_act@3.1 ... │ ││ └──────────────────────────┬──────────────────────────────────┘ ││ │ ││ ┌────────┴────────┐ ││ │ AIGP PROTOCOL │ ││ │ (Open Standard) │ ││ └────────┬────────┘ ││ │ ││ ┌─────────────────────────────────────────────────────────────┐ ││ │ EMITTERS (Subscribe to what they need) │ ││ │ One evidence stream → multiple Dialects → unified posture │ ││ └─────────────────────────────────────────────────────────────┘ │└─────────────────────────────────────────────────────────────────────────┘Appendix: Related AIGP Artifacts
| Artifact | Role in the Two-Sided Model |
|---|---|
| RFC-010 (Core Protocol) | The matching function — evidence interface between sides |
| RFC-034 (Domains of Concern) | Defines the concern classes Receivers can specialize in |
| RFC-035 (Mediation Vector Profile) | The measurement apparatus Receivers package |
| RFC-036 (Concern Calculation) | The evaluation logic Receivers encode |
| RFC-037 (Observer Accreditation) | Quality assurance for Receivers |
| RFC-038 (Dialect Registry) | The marketplace where Receivers publish and Emitters subscribe |
| DIALECT-001 (AI Autonomy) | The first published Dialect — proof the model works |
AIGP is owned by Kanjani AI Research & Causum. Registered with the U.S. Copyright Office (Case #1-15160968741, #1-15180695311). Patent-pending. Commercial licensing available exclusively through Causum.