Diop Daily #105 — September 2026

A Protocol Is a Border

A protocol appears to be a technical agreement: a set of fields, messages, permissions, and responses that allow separate systems to communicate. Its deeper function is political. A protocol decides which entities can be recognized, which actions can cross a boundary, which evidence travels with a claim, and which obligations remain attached to an exchange. Every machine-readable world has a border, even when the border is hidden inside a schema.

Several public signals now place this question in the path of real systems. The W3C and GS1 are convening a workshop on e-commerce for humans and AI agents, with a focus on creating content that agents can use. C2PA has published an implementation guide and continues to develop Content Credentials as an open provenance standard. Google’s A2A work describes secure handoffs between agents, while ADK Go 2.0 exposes graph workflows, human checkpoints, dynamic routing, and resilience. These efforts address different problems. Together they show that machine interaction is moving from isolated model calls toward negotiated participation in shared environments.

The question at a boundary is always the same: who is recognized, under which terms, and with what evidence?

Interoperability is a choice about recognition

People often speak of interoperability as if it were a neutral property. Two systems interoperate when they exchange data successfully. That definition is too thin for agents. An agent does not only move a string from one endpoint to another. It may represent a buyer, a publisher, a public office, a creator, a researcher, or a service provider. It may request an action, accept a result, invoke a permission, or transfer a responsibility.

The receiving system must decide what the message means. Is the sender a person, a delegated software role, or an unverified process? Is a price an offer, a reference value, or a conditional estimate? Is a media file an original, an edited derivative, or an asset whose provenance has been stripped? Is a request authorized for one transaction, a time window, or an entire account? Interoperability therefore depends on a shared vocabulary of identity, authority, state, and consequence.

Without that vocabulary, integration becomes a series of private translations. Each platform builds an adapter for the other platform’s assumptions. The surface may look connected while the meaning degrades at every handoff. A field called “owner” can mean a legal rights-holder in one system, an account administrator in another, and the person who uploaded a file in a third. The network carries the word and loses the institution.

Standards carry the memory of an exchange

C2PA’s Content Credentials work is a useful example because provenance is a relationship rather than a decorative label. The C2PA specification describes a way to bind assertions about an asset to a signed manifest, allowing viewers and systems to inspect how content was created or changed. Its July 2026 announcements include a new implementation guide and a governance expansion involving TikTok. Those facts show continuing work on adoption and implementation. They do not prove that every platform will preserve provenance correctly. They do show why the standard must carry a history of transformation across systems.

The same logic applies to machine-assisted research, software, commerce, and administration. A message should carry enough context for the next participant to determine:

  • Identity: which person, institution, role, or agent is represented.
  • Authority: which mandate permits the request and where that mandate stops.
  • State: whether the object is proposed, accepted, paid, reviewed, published, suspended, or withdrawn.
  • Evidence: which sources, calculations, signatures, or prior events support the claim.
  • Terms: which rights, restrictions, prices, languages, jurisdictions, and time windows govern reuse.

This is the memory of an exchange. It makes a transaction portable because the receiving system can reconstruct what crossed the border and why. It also makes disagreement possible. A creator can contest a rights claim. A buyer can challenge a price condition. An institution can revoke a delegated action. An agent can refuse a request whose authority or provenance is incomplete.

The agent economy will expose weak borders

The W3C and GS1 workshop is significant for a precise reason: product information designed for human readers may require different structure when agents search, compare, and act on behalf of people. A page can be visually clear and machine-ambiguous. A catalog can name a product while omitting availability, jurisdiction, delivery conditions, or the authority behind a claim. An agent needs the conditions of exchange in a form it can inspect before it acts.

That requirement reaches beyond retail. A research archive needs to tell a machine whether a text is public, licensed, provisional, or superseded. A publisher needs to express which territories and formats a creator has approved. A cooperative needs to represent units, quality grades, delivery windows, and settlement status in terms that another system can understand. A public institution needs to expose which service owns a request and which decisions require a human signature.

The risk is unequal visibility. Large institutions can afford custom schemas, dedicated integrations, and legal teams that negotiate exceptions. A small publisher, a local merchant, a community archive, or an African-language service may appear unreliable simply because its meaning does not fit the dominant platform’s fields. When a system cannot represent an entity, the market often treats the entity as absent.

Protocol design therefore determines who can participate in the machine economy. A standard that recognizes only global identifiers, dominant currencies, formal addresses, or one language creates a border that excludes through omission. A standard with extension points, clear semantics, local mappings, and portable evidence can widen participation without pretending that every institution has the same structure.

Handoffs require a jurisdiction, not only a connection

Google’s A2A materials describe secure agent handoffs and collaborative workflows. ADK Go 2.0 adds graph-based composition, human-in-the-loop orchestration, dynamic routing, and resilience. Those runtime features solve an important mechanical problem: one agent can pass work to another without forcing the entire process into one opaque model context.

The institutional problem is harder. A handoff changes who sees the data, who may act, and who becomes responsible for the result. A translation agent may receive a confidential manuscript. A procurement agent may receive a budget ceiling. A public-service agent may receive a resident’s case. The protocol must carry the boundary conditions, not merely the payload.

Every serious handoff should answer a small set of questions:

  1. Which jurisdiction governs the request and its data?
  2. Which role delegated the action, and can that delegation be revoked?
  3. What part of the context is necessary for this participant, and what must remain hidden?
  4. What output state can this participant create?
  5. Who receives the work when the evidence conflicts or the action exceeds permission?

A protocol that cannot answer these questions creates connectivity without custody. The workflow moves, yet responsibility becomes difficult to locate. The result is an architecture that scales technical participation while weakening institutional control.

African sovereignty begins at the schema

For African institutions, protocol design is a sovereignty question because schemas often arrive with assumptions about how a society is organized. A foreign platform may expect a stable street address, a bank-card payment, a single official language, an individual account, and a continuous internet connection. Those assumptions describe one operating environment. They do not define the world.

A cooperative may organize authority through a committee and a seasonal cycle. A creator may work across oral, mobile, and written records. A local service may identify a person through a combination of names, community relationships, phone numbers, and documents. A payment may be initiated through mobile money and settled later. A multilingual institution may need one concept to retain different legal or cultural meanings across languages.

Owning the schema means retaining the power to define these relationships. Institutions should be able to publish local extensions, map them to wider standards, and carry the original meaning alongside the translation. The goal is participation without semantic surrender. A machine may understand a local term through a mapping, while the institution retains the authority to say where that mapping is incomplete.

NIST’s AI Risk Management Framework gives a practical discipline for this work through Govern, Map, Measure, and Manage. Applied to protocols, those functions require an institution to govern its vocabulary, map its boundaries, measure translation loss and failed handoffs, and manage changes when a standard or partner evolves. Protocol governance belongs in the same room as security, legal review, product design, and language stewardship.

Where the investable surface is widening

If protocols become the borders of machine participation, capital should examine the layers that make those borders legible and fair:

  • Conformance infrastructure: test suites, validators, certification services, and monitoring that show whether an implementation preserves a standard’s meaning rather than merely passing a schema check.
  • Semantic translation gateways: systems that map local identifiers, languages, currencies, rights, and institutional roles into shared protocols while retaining the source meaning and the limits of the mapping.
  • Portable identity and delegation: rails for representing people, institutions, software roles, revocable authority, and scoped permissions across agent ecosystems.
  • Provenance-preserving exchange: infrastructure that keeps evidence, transformations, signatures, and rights attached as an asset moves between tools, markets, and jurisdictions.
  • Protocol observability: products that detect dropped fields, altered semantics, unauthorized handoffs, failed translations, and unresolved responsibility at the boundary.

The underwriting question is exact: does the product reduce the cost of crossing between systems while preserving identity, authority, evidence, and recourse? Useful measures include fewer failed transactions caused by semantic mismatch, lower manual reconciliation, faster certification of new partners, higher multilingual completion, fewer provenance breaks, and clean portability when an institution changes providers.

This is a different object from an API directory or an agent marketplace. Those surfaces help systems find one another. The protocol layer determines whether the systems can exchange a meaningful obligation. Its value grows when the network expands without forcing each participant to rebuild its identity and rules for every new connection.

Build the border before opening the gate

Institutions can begin with one consequential exchange. Choose a workflow that crosses an organizational boundary and document the terms before adding more agents:

  1. Name every entity, role, object, and decision that crosses the boundary.
  2. Define which fields are authoritative, which are translated, and which remain uncertain.
  3. Attach permissions, expiry, provenance, and correction paths to the exchange.
  4. Create conformance tests with local language, edge cases, partial records, and disputed cases.
  5. Measure what the receiving system lost, misunderstood, or invented during the handoff.

The strongest protocols are modest about what they know. They expose uncertainty instead of hiding it inside a successful status code. They leave room for a human institution to correct the record. They allow a participant to leave without taking the shared history hostage. These properties are technical, commercial, and political at the same time.

A protocol is a border because it decides what the machine world can recognize. The builders who control that border will shape who can transact, publish, govern, and inherit value across systems. African institutions should enter this work early, with their languages, authorities, archives, and forms of exchange present in the schema from the beginning. A connected future is worth building only when connection carries meaning across the line.

Sources