1. Scope

1.1 Purpose

SENAR defines a methodology for software development where AI agents are the primary producers of engineering artifacts and human engineers serve as Supervisors — directing, verifying, and governing the AI-driven process.

1.2 Intended Audience

  • Organizations transitioning to AI-native software development;
  • Supervisors — engineers who direct AI agents;
  • Managers responsible for delivery, quality, and cost;
  • Tool vendors building AI-native development platforms;
  • Auditors evaluating quality practices of AI-native teams.

1.3 Applicability

This standard applies where:

a) AI agents generate a substantial portion of production artifacts (code, tests, configuration, documentation); b) Human engineers direct, review, and approve AI-generated output; c) Traceability from requirements to delivered artifacts is required; d) Quality assurance is enforced through automated mechanisms.

SENAR is tool-agnostic. It applies to any AI coding platform — autonomous agents, IDE assistants, terminal-based tools, or custom pipelines.

1.4 Out of Scope

a) AI model training, fine-tuning, or evaluation; b) Traditional development where humans write the majority of code; c) Organizational change management; d) Specific tool implementations.

1.5 Position and Relationship to Other Standards

Where this standard sits

Two bodies of work bear on the quality of software work. The subject of this standard falls between them.

Established software-engineering and quality standards govern when work is verified, by whom, and to what rigor. IEEE 1012-2024 governs verification and validation planning by integrity level, and its 2024 revision states applicability to systems built on generative AI. ISO 9001:2015 governs the quality management system surrounding the work. ISO/IEC 42001:2023 governs an AI system as a product an organization builds and operates — producer and product are the other way round from the case addressed here. What this body leaves to the adopting organization is where a control is placed relative to the work it governs, and what a verdict is required to be a statement about. It leaves them open correctly: its subject is the plan, not the mechanism.

The second body is agent-native practice — frameworks, protocols, and tooling that made the AI agent the addressee of the instruction. This body supplies mechanism. What it does not consistently supply is a stated criterion for when a check counts as a gate; checks are described as quality gates in work whose own documentation calls them recommended and provides no barrier.

SENAR addresses what falls between: the properties a control is required to have in order to govern work produced by an agent (Section 8.6), together with the units, roles, and measurements that make those properties operable.

What is claimed, and how to refute it

The claim is about properties, not about the absence of neighbours. Any framework, protocol, or platform whose gates have the properties of Section 8.6 satisfies what this standard asks at that point, whether or not it references SENAR. To refute the positioning, name a published work that states such properties as requirements on a control; to refute the standard’s usefulness at this point, name a control governing agent-produced work that has all five and gained nothing by having them.

Basis for the positioning above: sixteen adjacent works were examined in an adversarial field review on 2026-08-11, and a focused re-check was run on 2026-08-17 against standards bodies, provenance formats, and agent tooling.

NOTE: Both examinations have recorded limits. The full normative texts of ISO/IEC 42001 and IEEE 1012-2024 are paid publications; structure, control titles, and public analyses were read, and their clauses are therefore not quoted here. The search was English-language. Neither examination is a proof that no other work occupies this position, and neither is offered as one.

Adjacent subjects this standard does not cover

a) Provenance of an artifact — what produced it, and through which chain of custody; b) The format of an attestation — how a signed statement about an artifact is encoded, exchanged, and verified as a document; c) Management of an AI system as a product — the controls an organization applies to an AI system it builds, deploys, and operates.

These are the subjects of neighbouring work, referenced informatively in Section 2. A gate conforming to this standard consumes an attestation; it does not define the attestation’s format.

Concepts adapted from established methodologies

SENAR extends and adapts concepts from established methodologies:

  • SAFe 6.0 — SENAR reimagines SAFe concepts for AI-native teams. SAFe comparison notes are provided throughout the document.
  • ISO 9001:2015 — SENAR Quality Gates are mapped against selected ISO 9001 clauses by their published numbering and titles; the mapping is not an assertion about clause text, which was not read (Section 0, survey coverage). Organizations should conduct gap analysis against the source for full compliance.
  • ISO/IEC 25010:2023 — SENAR metrics align with the published characteristics of the software quality model, on the same basis.
  • Scrum / Kanban — SENAR borrows iterative delivery and flow measurement while replacing human-centric ceremonies with AI-appropriate alternatives.

SENAR may be extended with domain-specific profiles for regulated industries (medical devices, financial services, aerospace) that add controls required by sector-specific standards.

← Introduction References →