# Polish Orchestrator: Small, Safe Code Clean Ups, Ranked and Queued > The Polish Orchestrator is the AI code refactoring coordinator in the Ai1 platform by MyZone AI: it maps clean up candidates in your code, admits only small clean ups that keep your software behaving the same, ranks them by where they help most and never edits code itself. Canonical page: https://myzone.ai/pages/agents/ai-code-polish-agent Part of Ai1, by MyZone AI. Book a code polish walkthrough: https://calendly.com/d/ct6h-tcy-8qf/ai1-demo?a1=Polish%20Orchestrator&utm_source=myzone.ai&utm_medium=agent-page&utm_content=ai-code-polish-agent-final Last updated: 2026-10-05. Reviewed by the MyZone AI team. Small, safe code clean ups, ranked by where they help most. ## At a glance - **The map:** An always current map of clean up candidates, with health signals - **Admission:** Bounded scope, no clash, a proof plan, a rollback and a named owner - **Ranking:** Often changed, complex, widely used or duplicated code comes first - **Feature work:** Always has priority - **Proof:** Automated builds, tests and checks decide; an AI review only supports - **What it never does:** Edit code itself ## What the Polish Orchestrator does - **Maps the code:** It keeps an always current map of clean up candidates, such as duplicated, tangled or hard to read code, with health signals for each area. - **Admits only safe clean ups:** A clean up is accepted only with a bounded scope, no clash with higher priority work, a proof plan, a way to roll back and a named owner. - **Ranks where it helps most:** Code that changes often, is complex, widely used or duplicated comes first. Cosmetic tidying comes last. - **Stays out of the way of features:** Feature work always has priority. If a feature lands mid clean up, the clean up is redone on the new code, never merged by hand. - **Reports weekly:** It ranks the queue again every week and reports metrics on safe progress. ## How the Polish Orchestrator differs from an AI coding assistant An AI coding assistant can edit whatever code it is pointed at, even while feature work is changing it. The Polish Orchestrator never edits code itself. It admits only small, bounded clean ups with a proof plan, a way to roll back and a named owner, ranks them by where they help most and keeps them clear of feature work. Automated checks, not an AI review, decide whether behaviour stayed the same. ## How it works 1. **Propose** (Whenever you spot messy code): A clean up is proposed, by you or from the code map. 2. **Admit** (Before anything is queued): It checks the scope, any clashes, the proof plan, the rollback and the owner. 3. **Rank** (Every week): It ranks the code clean up queue by where clean up helps most. 4. **Hand off** (Once an edit is admitted): Each admitted edit goes to the developer agents with its scope locked and an automated proof check. 5. **Prove** (After every clean up): Automated builds, tests and checks decide whether behaviour stayed the same. ## When to use it - **Two screens have nearly identical code:** It checks the proposed clean up against its admission rules, then queues it or keeps it as a read only mapping for now. - **You want to know what to clean up first:** It shows the ranked queue with the reason each item sits where it does. - **You want to chip away at technical debt without slowing features:** It keeps clean ups clear of feature work, and feature work always wins. - **You want a progress report:** It shares the weekly metrics on safe progress. ## What you get - An always current map of clean up candidates with health signals - A ranked clean up queue, re-ranked every week - An admission check for each proposed clean up - Weekly metrics on safe progress ## Example: weekly code clean up report Example with a fictional company. Names, people and figures are invented to show the agent's output. Any resemblance to a real company or person is unintended. What it was asked: Start a safe, steady clean up of our app's code, tell us where it would help most, and do not get in the way of the feature work we have under way. An illustrative weekly report that follows the steps Polish Orchestrator runs, per its current public guide: map the code, check a proposal against the five admission rules (scope, clashes, proof plan, rollback, owner), rank the queue and report weekly progress. The company, code areas and all figures are invented and no real code was read or changed. Run date: 1 October 2026. ### What it found - The reports area is the clear first target: it changed 41 times in 90 days and 38% of its code is repeated across three reports. - The invoice and grant export clean up is held back because new feature work is changing those files; it will be mapped again on the new code. - No edits have been handed out yet: the queue is ranked and ready, but edits wait until a scope locking service and an automated proof check are live. ## Guardrails - It never edits code itself; the developer agents make every edit. - Every clean up must keep behaviour the same. Automated builds, tests and checks decide; an AI review is supporting evidence only. - A test's expected results are never changed in the same edit as the code. - The Platform Architect owns the standards and gate rules, and it never approves changes to its own rules. ## Frequently asked questions ### How should a software team run a code clean up queue? Admit only small, bounded clean ups with a proof plan, a way to roll back and a named owner, then rank them by where they help most. In Ai1 by MyZone AI, the Polish Orchestrator ranks the queue again every week and keeps it clear of feature work. A behaviour preserving clean up tidies duplicated, tangled or hard to read code and leaves what the software does unchanged. ### How can you pay down technical debt without slowing feature work? Give feature work priority and keep clean ups away from code it is changing. Under the Polish Orchestrator, only one change may work on a given part of the code at a time, and if a feature lands mid clean up, the clean up is redone on the new code, never merged by hand. ### How do you prove a refactor did not change what the software does? Let automated builds, tests, contracts and performance checks decide. With the Polish Orchestrator, each clean up must keep the software behaving the same, an AI review is supporting evidence only and never outranks those checks, and a test's expected results are never changed in the same edit as the code. ### Does the Polish Orchestrator change my code itself, and is it available? No. It decides and coordinates; the developer agents make every edit. It is ready to use: it maps clean up candidates in your code, admits and ranks them, hands out edits to the developer agents and sends the weekly report. ## About Ai1 Ai1 is the AI operations platform by MyZone AI, where each client runs on its own private server. The Polish Orchestrator is included on every Ai1 level, including Developer Core, with no per-agent charge. It works alongside the other Ai1 agents on your account. Pricing: https://myzone.ai/pages/services/ai1-pricing. Security: https://myzone.ai/pages/security.