A standard with a history.

What Velocity contributes

Created by Mike Conant, Velocity captures practices developed in his AI-assisted engineering work. Its focus is the combination of protected acceptance criteria, separated write authority, proof matched to claims and environments, enforcement of touched invariants, and separately governed process evolution.

The method is independent of the AI provider. Use the coding tools and runtime that fit your project, and map their permissions and evidence to the standard. Velocity supplies governance and templates; the project supplies the working integration.

The standard comes first

The standard pages publish the canonical documents in full from a named release. The field guide is explanatory material for applying them. The source repository remains the authoritative home, and project overlays hold local commands, environments, and populated records.

Version 1.4 added optional automation evaluation, transition, and measurement templates. Version 1.5 adds a complete ADR record format and attribution clarifications. The manifesto’s proposed delegation grades remain future direction, including its unresolved timeout-based acceptance idea.

Give credit to the ideas

Velocity draws on separation of duties, change control, V-model verification, architecture decision records, and operational lane discipline. It makes no claim to have invented these traditions. The ADR template credits Michael Nygard’s 2011 account, which describes concise decision records and preserving superseded decisions.

Anthropic’s AI-Native SDLC Playbook prompted the September 2026 review. Its development-agent evaluations, artifact-based handoffs, and measurement feedback informed Velocity’s v1.4 additions. These are credited adaptations, not claims of original invention.

Comparing artifact structures

Anthropic’s proposed artifact flow uses named Markdown files to connect stages. Velocity retains its existing work-item and authority model. This is a reading aid for comparing them, not a required file structure or a claim of exact equivalence.

A mapping does not make two methods identical. Each installation still needs to qualify its actual controls and preserve one authoritative home for every artifact.

Other relevant work

The repository contains a dated, source-linked comparison reviewed September 9, 2026. It is a documentation review, not an execution benchmark. These are the primary sources covered:

• AWS AI-DLC Workflows

  Lifecycle workflow and implementation.

• GitHub Spec Kit

  Specification-driven development workflow.

• BMad Method

  Structured agent roles, review, and development workflows.

• Kiro specifications

  Requirements, design, and task artifacts.

• Superpowers

  Composable development skills and review practices.

• OpenAI harness engineering

  An engineering experience report on agent-oriented development infrastructure.

History and evidence

The first Velocity repository commit is dated May 11, 2026. It records extraction of a method that had already been embedded in a product project. That date establishes the repository’s beginning; the start of the underlying practices remains unestablished by the repository.

Related public work predates that extraction. Velocity therefore makes no “first” or “only” claim. Its adoption accounts are maintainer-reported experience, and its automation example is synthetic. There are no published universal speedup claims or measured pilot outcomes.