Agent Native Dev Protocols 0726
Research: is anyone building an agent-native protocol for the SDLC (specs, task graphs, progress state) — or is it still markdown all the way down? (July 2026)
About This Project
Agent-Native Dev Protocols is a research project that examines whether anyone is building a protocol designed from the ground up for AI coding agents to produce and consume software specifications, task graphs, and progress state. The central question is whether current AI-assisted development relies on ad hoc markdown files, or whether a standard exists that treats agents as primary consumers rather than retrofitting human project-management tools such as Jira, Linear, or sprint-based workflows with agent integrations.
The repository is organized into three parts: a refined research question with the original verbatim capture, a set of gathered findings and source notes from a deep-research pass, and a Typst source file with the compiled PDF report. The method combined multi-angle web research with adversarial claim verification using a Claude Code deep-research workflow, followed by synthesis into findings and authoring of the report in Typst. The final PDF is attached to the latest GitHub release.
The project is relevant to developers building multi-agent coding workflows, tooling authors evaluating spec and task-tracking formats for agents, and researchers tracking the state of AI-assisted software engineering practice as of July 2026.