When an agent writes code, the development framework acts as the control system for a non‑deterministic worker. Since 2025, spec‑first, agent‑driven frameworks have gained rapid traction; notable examples are GitHub Spec Kit, obra/superpowers, BMAD, GSD, and the newly introduced Consort. All of them capture intent up front via a specification or durable planning artifact, but they differ in how they enforce the engineering discipline that keeps agent‑generated code clean, correct, and maintainable. We categorize existing enforcement strategies into three modes: persuasion—prompt‑based discipline that the model may ignore; front‑loaded structure—strong specs followed by a trusted build; and immutable controls— a deterministic orchestrator, human‑approved gates, immutable tests, and a green result that must pass against a live, branched database.
Consort adopts the third mode. A deterministic orchestrator drives separate role agents through a spec‑first design lane and a test‑driven build lane on a live database branch. Specifications are authored first and stored persistently; thereafter each role agent operates only within the orchestrator’s controls, generating code, running unit tests, and performing integration checks that must succeed on the actual database branch before any merge. By hard‑coding tests and gates inside the framework, agent‑written code is forced to remain verifiable, while specialized roles—including human reviewers—provide maintainability. We frame these claims as a pre‑registered, testable hypothesis and plan to validate them experimentally.
Review