Skip to main content
RustGrid

IMPLEMENTATION ROADMAP

From durable core
to ADD control plane.

Five proposed phases from the October 2026 paper. Each phase has an exit gate.

TEAM ADOPTION / NOT A BIG-BANG MIGRATION

Change the work stream before changing the whole organization.

The paper separates product implementation from team adoption. A Scrum team keeps each existing ceremony until its ADD replacement produces better evidence.

01

Make “done” executable

Add static, security, and test gates to every pull request. Encode the first ten unwritten team rules.

RustGrid answer: quality evidence and policy foundation
02

Run one spec stream

Choose a low-risk area, approve Given/When/Then criteria, and let agents work through controlled pull requests.

RustGrid answer: projects, tasks, runs, GitHub, activity
03

Add routing and cost

Introduce model tiers, per-spec budgets, failure review, and a senior exception rotation.

RustGrid answer: telemetry foundation; orchestration phase 4
04

Switch the operating model

Move remaining work to the spec queue. Derive state from gates and replace sprint rituals with ADD reviews.

RustGrid answer: spec and gate-state phases 2–3

THE STARTING POINT

Reuse the foundation. Extend the model.

The paper identifies authorization, idempotency, optimistic concurrency, tickets, telemetry, HTTP, and gRPC as existing building blocks. The phases below describe the additional ADD work.

Proposed phaseScopeExit gate
0 · Agent-safe coreAgent principals, OAuth 2.1, task permissions, and leases.No agent write without a token and a lease.
1 · Agent surfaceA stateless MCP server, delegation, and cost recorded on writes.An agent ships a real ticket end to end.
2 · SpecificationsImport spec folders and link tasks to approved specifications.Every task traces to an approved spec.
3 · Quality gatesIngest GitHub checks and derive work state from gate results.Status is evidence-driven.
4 · OrchestrationTier routing and cost aggregation per specification.Cost per spec is reported and tasks are routed by tier.

FIRST MILESTONE

One agent. One real ticket. End to end.

An agent claims a RustGrid ticket through an authenticated API, performs the work, attaches its evidence, and completes the task with an audit trail. This is the first proposed proof point.

Keep the core deterministic. Put model calls outside it. Add task-scoped permissions and leases before expanding autonomy.

Explore the developer foundation

THE NEXT WAY TO BUILD

Give your agents a system to work in.

Start with RustGrid’s execution foundation. Build toward a delivery system governed by specs, gates, and clear ownership.