Diop Daily #038 — June 2026
· 12 min read · agents · protocols · commerce

Treaties Before Market Share: A Constitution for Agents

The market still evaluates AI through the optics of spectacle. Which model is faster? Which demo looks more autonomous? Which assistant writes, codes, or reasons with more apparent fluency? These questions are not trivial, but they are incomplete. A serious economy is not built when isolated systems become impressive. It is built when different systems can recognize one another, exchange structured intent, present usable proof, and act inside shared rules without collapsing trust at every boundary.

That is why several recent public signals deserve to be read together. Google’s June 18 note on the Agent-to-Agent protocol argues that agents require a common language for secure collaboration and handoff. W3C has announced a workshop on e-commerce for humans and AI agents, explicitly asking what content and standards must change when agents become intermediaries in commercial activity. The European Commission’s General-Purpose AI Code of Practice process is translating model governance into a more legible public rule surface. And C2PA’s February release of Content Credentials 2.3 extends provenance infrastructure for digital artifacts. Different institutions, different domains, one direction: machine action is being pushed toward machine-readable trust.

The next agent market will not be won by the loudest demo. It will be won by the systems that can enter reliable relations with other systems.

From model performance to inter-agent order

It is easy to imagine the future of agents as a simple scaling story: more reasoning, more tools, more vertical applications, more tasks automated. But once agents leave the sandbox, the problem changes. A useful commercial agent does not merely answer. It must discover another service, declare its intent, negotiate capability boundaries, carry context without leaking everything, prove where an output came from, and act under permissions that can later be audited. In other words, agency without order quickly becomes expensive improvisation.

This is why protocols matter. Human markets do not run on intelligence alone. They run on contracts, standards, receipts, signatures, ledgers, and institutions that let strangers transact with bounded risk. Agents will require the computational equivalent. A2A is one attempt to formalize how software agents coordinate and hand off work. W3C’s workshop signals that commercial content itself may need to be authored with agent mediation in mind. The EU process signals that providers of general-purpose AI will face explicit expectations around transparency, copyright, and systemic-risk governance. C2PA shows that provenance is slowly becoming attachable infrastructure rather than a luxury add-on.

Why treaties is the right metaphor

I use the word treaties deliberately. A treaty is not friendship. It is a structured agreement that makes cooperation possible without requiring total trust or perfect similarity. That is exactly what the emerging agent economy lacks. Firms are racing to produce agents, but there is still weak agreement on how one agent should identify itself to another, what counts as enough context to proceed, how rights and restrictions are expressed, how outputs are attributed, and what evidence survives after the action.

Without such treaties, every new deployment behaves like a closed empire. It may look powerful inside its own walls, but it produces friction at every border. With treaties, by contrast, specialized systems become easier to compose. One agent can route to another without pretending to do everything itself. One merchant can expose machine-readable inventory or policy. One research workflow can move across tools without turning memory into a pile of screenshots and copied prompts. A treaty layer turns raw capability into coordination.

  • Identity: who or what is this agent, and under whose authority does it act?
  • Intent: what exactly is being requested, delegated, purchased, or transformed?
  • Permission: what is allowed, prohibited, rate-limited, or reversible?
  • Provenance: what evidence accompanies the output or transaction?
  • Liability surface: where can a human institution inspect what happened and contest it if necessary?

These are not decorative questions. They determine whether an agent can participate in commerce, procurement, media, or regulated workflows without forcing the human operator to re-verify everything manually.

Where the investable surface is widening

If this reading is correct, the investable surface is not limited to general-purpose models or front-end copilots. It widens around the trust rails that let many agents become governable economic actors.

  • Agent identity and authorization layers: systems that express delegated authority, policy scope, spending limits, and revocation in machine-readable form.
  • Interoperability middleware: protocol gateways, service registries, and negotiation layers that allow specialized agents to collaborate safely rather than remain trapped inside single vendors.
  • Provenance and rights infrastructure: content credentials, attribution trails, usage rights, and evidence packaging for synthetic and mixed-origin outputs.
  • Merchant and publisher tooling for agent-mediated commerce: structured product, pricing, inventory, and policy surfaces designed for both human readers and machine intermediaries.
  • Audit and dispute systems: records that preserve enough context for regulators, enterprises, creators, and customers to challenge or validate machine action after the fact.

The common pattern is clear. These businesses do not merely help AI do more. They help institutions underwrite AI use with lower uncertainty. That moves them closer to budget, compliance, and durable procurement. In capital terms, they sit nearer the layer where categories become buyable at scale rather than merely admirable in demos.

What this means for African and diasporic builders

African technological sovereignty should study this moment carefully. It would be a strategic error to consume agent interfaces built elsewhere while outsourcing every serious trust layer, policy layer, credential layer, and machine-readable commerce surface to foreign defaults. That would reproduce dependency at a higher level of abstraction. We would appear present at the application edge while remaining absent from the constitutional layer of the market.

The Pan-African task is therefore not to imitate the noisiest consumer agent. It is to build competence in the rails that make agent action legible and negotiable: local identity frameworks, language-aware structured data, provenance systems, permissioning, payment integration, public archives, and institutional memory. A continent that wants sovereign participation in the machine economy must not only build agents. It must also help define the treaties under which agents meet.

Cheikh Anta Diop insisted that historical recovery was only the beginning; the objective was scientific capacity. The same standard applies here. It is not enough to celebrate that intelligent systems exist. One must ask who defines the protocols, who issues the credentials, who stores the proof, whose languages are understood in commercial metadata, and whose institutions can inspect the machine record. Otherwise one remains a user inside someone else’s order.

Rails before empires

The present temptation is to build agent empires before agent treaties. Every company wants the dominant assistant, the dominant storefront, the dominant workflow shell. But markets usually reward the quieter layer first. Payments were not transformed only by beautiful interfaces, but by settlement rails. The web did not scale because each company invented a private internet, but because shared protocols made connection cheaper than isolation. The same logic is returning to the agent economy.

Builders and investors should therefore ask a sterner question than “Which agent looks most human?” They should ask, “Which infrastructure makes machine relationships trustworthy enough to transact, publish, negotiate, and recover?” That is the layer on which the next serious agent companies will stand. Before market share comes admissibility. Before empires come treaties.

Sources