In a world of agent-generated capabilities, the unit of value is not the model. It is the construct.
A construct is a git-resident unit of GPU capability: a kernel, a skill, a piece of equipment. Construct economics is the study of how these units are created, valued, shared, and composed into larger capabilities — and how the agents that produce them are incentivized to keep producing.
Traditional software economics is based on packages, licenses, and labor. A human writes code; a company sells access. Open source inverts part of this model: code is free, but support, hosting, and integration are paid.
Construct economics is a third model. Constructs are:
- Generated by agents: Human-authored constructs are valuable, but agent-generated constructs scale exponentially.
- Versioned semantically:
SemVerenables compatibility contracts and dependency resolution. - Composed recursively: A construct can depend on other constructs, creating a supply chain of capability.
- Executed on demand: Constructs are loaded, compiled, and deployed only when needed.
- Verified by tests: Every construct carries evidence of correctness, not just trust.
Value in the construct economy flows through four layers:
- Raw compute: GPU hours, memory bandwidth, network transit. The commodity layer.
- Constructs: Reusable capabilities that turn raw compute into specific outcomes.
- Compositions: Pipelines, agents, and applications assembled from constructs.
- Outcomes: The end-user value produced by compositions.
A mature construct economy drives value toward the top of the stack. Raw compute becomes cheaper through competition; outcomes become more valuable through composition.
Constructs can be priced in several ways:
- Free and open: Anyone can load, modify, and redistribute. Sponsored by reputation, grants, or upstream services.
- Freemium: Basic constructs are free; advanced constructs require payment or subscription.
- Usage-based: Pay per invocation, per GPU-second consumed, or per byte transferred.
- Dependency royalty: When construct A depends on construct B, a fraction of A's revenue flows to B's maintainers.
- Staking and reputation: Agents stake tokens on the quality of their constructs; bad constructs lose stake.
The SuperInstance ecosystem is designed to support all of these models without mandating any of them. The registry format includes identity, signature, and dependency fields precisely so that economic contracts can be layered on top.
The construct economy has powerful network effects:
- Complementarity: The more constructs exist, the more valuable it becomes to add new constructs that compose with them.
- Standardization: Widely used constructs become de facto standards, attracting more users and more contributors.
- Specialization: Agents develop comparative advantage in specific domains, producing higher-quality constructs over time.
- Trust accumulation: Constructs with long histories of successful invocation build reputations that new constructs lack.
These effects can create virtuous cycles of capability growth — or lock-in around early winners.
- Construct dependency graph analysis: Crawl a registry of 1,000 constructs and measure in-degree distribution, clustering coefficient, and most central constructs. Expected: power-law in-degree, high clustering, a small set of "standard library" constructs with outsized influence.
- Revenue simulation: Simulate three pricing models (free, usage-based, dependency royalty) under different agent population sizes. Measure total ecosystem revenue and Gini coefficient. Expected: usage-based maximizes short-term revenue; dependency royalty maximizes long-term diversity.
- Quality vs. quantity trade-off: Compare agent-generated constructs at different temperatures. Measure acceptance rate (passing tests) and uniqueness. Expected: low temperature → high quality, low diversity; high temperature → low quality, high diversity.
- Reputation convergence: Run a staking game where agents rate each other's constructs. Measure how quickly reputation rankings converge and whether they correlate with objective test scores. Expected: convergence in O(log n) rounds with honest majority.
- Agent marketplaces: Platforms where agents buy, sell, and compose constructs.
- Federated R&D: Organizations contribute constructs to a shared pool while monetizing proprietary compositions.
- Open science: Research groups publish reproducible experiment pipelines as constructs.
- Infrastructure funding: Critical open-source constructs receive sustainable funding through usage royalties.
- Skill specialization: Agents develop niche expertise and rent it to other agents via construct invocations.
- Anti-trust: How do we prevent a small number of constructs from becoming unavoidable dependencies that extract monopoly rents?
- Attribution: If construct A is generated by an agent trained on millions of human-written constructs, who owns A?
- Quality assurance: Can economic incentives replace centralized review in maintaining construct quality?
- Cross-chain settlement: Should construct payments settle on a blockchain, through traditional payment rails, or via in-game resource exchange?
- oxide-constructs — The technical substrate for the construct economy.
- COST-ECONOMICS.md — The dollar math of running ternary compute.
- VERSIONING-AND-COMPATIBILITY.md — SemVer as the economic contract language.
- SECURITY-MODEL.md — Trust and identity in a decentralized construct market.
- GOVERNANCE-AND-COMMUNITY.md — How the ecosystem decides rules that shape economic outcomes.
- THE-ROADMAP.md — Where the construct economy is heading.
oxide-fleet— The execution layer where construct value is realized.
Inventory every reusable capability in your current project. For each one, create a minimal construct manifest with name, version, type, and dependencies. Publish the manifests to a private registry (even if the code is not yet git-published). Calculate how many distinct capabilities you own, how many you depend on, and what your dependency tree's depth is. Identify one construct you could open-source to start participating in the ecosystem, and one construct you would want to buy or borrow if a marketplace existed.