Proposal: AI Working Group (WG-AI)
Nobody has claimed this yet.
Assessment
- Difficulty
- 5/5
- Estimated time
- Over a week
- Newbie friendliness
- 25/100
- Issue type
- Feature
- Clarity
- Clearly specified
- Activity status
- Active
- Tech stack
- github
- Domain
- documentation
Research direction
Start by reading the proposal and its linked TSC charter guidance. The requested work spans policy drafting, operational pilots, tool evaluation, documentation guidance, and a SIG-AI charter, with decisions and ratification assigned to the TSC. There is no single bounded implementation task or completion test for a first-time contributor to take on.
Written by the indexing model from the issue text.
Description
Submitted to: O3DE Technical Steering Committee
Proposed by: colinb[APMG] (SIG-Platform chair)
Leadership Nominations: Proposed 3 leadership seats up for vote
Date: 2026-10-27
Requested TSC actions:
- Formally except formation of the WG-AI.
- Vote on WG leadership, one week for nominations, one week for voting.
- Assign the resources in resources requested.
Proposed lifespan: 6 months from formation, ending with a formal proposal for the formation of SIG-AI and charter proposal
Summary
Linux Foundation has a Generative AI Policy, however O3DF has no written position on AI. Contributors are already sending AI-generated pull requests. Maintainers are already using AI to review them. Both are happening without shared rules. The TSC has raised the topic in most meetings since August 2026 and has agreed that it needs a policy "in writing". It has also agreed the topic is too large to settle in a regular TSC agenda slot.
This proposal forms a short-lived working group to do three things:
- Draft the project's official AI policy for TSC ratification.
- Pilot AI in project operations, such as code review, issue and PR triage, and SIG labeling, and measure whether it helps.
- Write the charter for a permanent SIG-AI to own this area once the WG ends.
Background
The TSC has discussed AI at most meetings since August 2026. It hasn't decided anything yet.
| Meeting | What was raised |
|---|---|
| 2026-08-11 | Should AI review code, with one AI and one maintainer per PR? The bottleneck is getting a PR from zero to one review. Proposal: every PR gets an AI review, whether or not it counts toward approval. |
| 2026-09-01 | Recent PRs are AI-generated, very large, include prompts, and take a long time to review. |
| 2026-09-22 | Velocity is low and reviewers are scarce. Could AI help with review? The Linux Foundation is pushing into AI and may have resources. |
| 2026-09-29 | "AI policy discussion: need it in writing." Options named: human in the loop, blind, or "don't ask, don't tell". Keep the policy small and clear. Ideas raised: PR size limits, an exception for mechanical changes, requiring an issue for every PR, a CodeRabbit-style reviewer, and free AI credits for open source projects. |
| 2026-10-06 | More people find projects through AI tools, and AI tools can't read our Discord, so they don't know about the project. |
| 2026-10-20 | Long discussion. Proposed: "bring your own AI"; the person who merges is responsible for the change. New contributors need a solid policy. The focus must stay on quality for users, not on making AI contribution easier. Conclusion: "We need a meeting like the TSC meeting, but where we speak about it specifically." |
| 2026-10-27 | Ulrick28 wants to create an AI WG but is stalled because there's no WG creation process. JT: a WG needs a write-up, a paper trail, and assigned resources. |
This document is that write-up.
Why a working group
- AI cuts across every SIG. Every SIG reviews PRs, triages issues and applies labels. A policy owned by any one SIG wouldn't bind the others. Under the charter, the TSC may create "sub-committees or working groups to focus on cross-project technical issues and requirements".
- A WG is the lighter first step. Earlier TSC guidance, from the Mobile WG discussion, is that a WG is quicker to stand up than a SIG, has a fixed purpose and lifespan, and can grow into a SIG. A SIG needs much more policy and structure up front.
- We don't know yet what a SIG-AI should own. The WG's research answers that question before a permanent body is created.
Mission
Define how O3DE uses AI as a project, and how contributors may use it, then hand that work to a permanent SIG-AI.
Scope
In scope
1. AI contribution policy. One page, small and clear, ratified by the TSC. It covers:
- Accountability. The human who signs the DCO owns the contribution, whatever tools produced it.
- Whether and how contributors disclose AI assistance.
- PR size and focus rules, with an exception for mechanical changes.
- Whether contributors must understand and be able to explain their change.
- Tool licensing and third-party content, aligned with the Linux Foundation Generative AI Policy.
- What reviewers may do with a PR that doesn't meet the policy.
2. AI in project operations. Evaluate and pilot:
- Code review. AI review on PRs. Decide whether it is advisory only or counts toward the required approvals, and which findings block a merge, for example only criticals and blockers.
- Triage. Applying SIG labels (
sig/*), triage labels and duplicate links to new issues and PRs, so SIG chairs spend less time on routine sorting. - Housekeeping. Finding stale issues and PRs, flagging PRs with no linked issue, and drafting release notes.
- Safety. Screening PRs for security problems and moderating spam before a human reviews them.
3. Tooling and resources.
- Compare candidate tools: hosted review bots, the GitHub-native options, and self-hosted or bring-your-own-AI setups.
- Find free or discounted AI credits through the Linux Foundation, foundation members and vendor open-source programs.
- Recommend how to track cost and usage.
4. Making O3DE readable by AI tools. Recommend agent instruction files for the main repos (for example AGENTS.md), an llms.txt for the docs site, and ways to make knowledge that now lives only in Discord findable. Work with SIG-Docs-Community on this.
5. A SIG-AI charter. Write the scope, membership rules and ownership boundaries for the permanent SIG, and present it to the TSC.
Exploratory, with a recommendation only
- AI features in the engine itself, such as ML gems, in-editor assistants and AI-driven tools. The WG will list existing efforts and recommend whether SIG-AI should own them, share them with existing SIGs, or leave them alone. The WG won't build engine features.
Out of scope
- Requiring any AI tool for contributors or maintainers.
- Changing review or merge permissions directly. The WG recommends; the TSC and the SIGs decide.
- Overriding a SIG's own review rules. Each SIG keeps final say in its area.
- Legal interpretation. Licensing questions go to Linux Foundation counsel.
Guardrails for pilots
Any pilot the WG runs on O3DE repositories follows these rules:
- AI never merges. A human approves and merges every change.
- AI reviews are advisory and don't count toward required approvals, unless the TSC votes otherwise after seeing pilot results.
- Pilots are opt-in per SIG. SIG-Platform will go first. Any SIG chair can turn a pilot off in their area at any time.
- AI actions are visible. Bots post from clearly labeled accounts, and every action they take on GitHub is public.
- Least privilege. Pilot tools get only the permissions they need, such as labels and comments, with no write access to branches. If we have SIG-Security resources then reviews from them first.
- No CI cost surprises. Coordinate with SIG-Build before anything adds load to automated review (AR) builds.
Deliverables and timeline
| Month | Deliverable |
|---|---|
| 1 | WG formed: leadership team confirmed, repos created, WG meetings to take place in #WG-AI Discord channel, weekly meeting cadence. |
| 2 | AI contribution policy v1 presented to the TSC for ratification. |
| 2–4 | Pilots: AI review and triage labeling in SIG-Platform repos, then in volunteer SIGs. Baseline metrics recorded before each pilot starts. |
| 5 | Pilot report with recommendations: what to adopt project-wide, what to drop, and what it costs/BYOAI |
| 6 | SIG-AI charter presented to the TSC. The WG closes when the TSC accepts the charter or decides not to form SIG-AI. |
The TSC can extend the WG by vote if the work isn't finished.
How we'll measure the pilots
- Time from a PR opening to its first review.
- Share of new issues and PRs triaged and labeled within 7 days.
- Label accuracy: the share of AI-applied labels a human had to correct.
- Review usefulness: the share of AI review findings that maintainers acted on.
- Maintainer and contributor feedback, from a short survey at the end of each pilot.
- Cost per month.
Leadership
The WG will be led by a leadership team of three to five. AI touches every part of the project, so leadership should come from more than one SIG and more than one organization. The TSC is asked to confirm the leadership team by vote rather than by appointment, so the WG's authority to speak for the project is on the record.
| Seat | Nominee |
|---|---|
| 1 | colinb[APMG] (SIG-Platform chair) Self Nominated |
| 2 | Ulrick28 (former director O3DE) Self Nominated |
| 3 | JT[SCB_Network] Nominated by colinb[APMG] |
| 4 | Shauna [Genome Studios] Nominated by colinb[APMG] |
| 5 | Astin(Chogan) nominated by ladyfia Iris |
| 6 | Open |
| 7 | Open |
| 8 | Open |
| 9 | Open |
| 10 | Open |
Proposed process, following the [2022 TSC co-chair election] (o3de/tsc#33):
- Nominations. Open a GitHub issue in o3de/tsc for one week after the TSC forms the WG. Any community member may nominate themselves or someone else for the three open seats. Each accepted nomination includes a short introduction, an acknowledgement of the role, and the nominee's goals for the WG.
- Balance. Nominees from SIGs and organizations not yet represented are encouraged. No more than two seats should be held by people from the same employer.
- Vote. At the next TSC meeting with quorum, following the charter's voting rules: a majority of those attending, with at least 50% of voting members present. If there are more nominees than open seats, each voting member votes for up to three nominees, and the three with the most votes are seated.
- Unfilled seats. If fewer than three people are willing to serve, the WG starts with the willing members that it has. The remaining seats stay open, and the TSC can fill them by the same process at any later meeting.
- Roles. The leadership team picks a chair and co-chair from among its members and records the choice in the WG repo. The chair runs meetings, keeps the paper trail and reports to the TSC. The co-chair stands in for the chair. All five leads share drafting, pilot oversight and SIG outreach.
Membership and decisions
- Open membership. Anyone in the community can join and take part.
- SIG liaisons. Each SIG can name one liaison, so policy drafts reflect how each SIG actually reviews and triages.
- Decisions. Lazy consensus by default. If there's an objection, a majority of members attending a WG meeting decides. Policy documents take effect only after TSC ratification.
- Transparency. Meeting notes and drafts are public in the WG repo. Pull requests to the repo are the paper trail.
Relationship to SIGs
The WG works through the existing SIGs and doesn't replace any of them.
| SIG | Involvement |
|---|---|
| SIG-Platform | Volunteers to host the first pilots. |
| SIG-Security | Reviews tool permissions and the security screening pilot. |
| SIG-Build | Reviews CI cost and load. |
| SIG-Docs-Community | Works on making docs and community knowledge readable by AI tools. |
| All SIGs | Can name a liaison to speak for their SIG. Opt in to pilots. Keep final say over their own review rules. |
Resources requested
- GitHub repository:
o3de/wg-ai, for notes, drafts and pilot reports. - GitHub team:
@o3de/wg-ai, maintained by the leadership team. - #WG-AI Discord channel OR The WG could adopt the AI channel Gaian already started, if Gaian agrees.
- Approval to install pilot tools (GitHub Apps or bot accounts) on repositories whose SIGs opt in, after SIG-Security review.
- Help from Linux Foundation and O3DF staff in asking for open-source AI credits.
Path to SIG-AI
The WG will recommend forming SIG-AI if, by month 6:
- The TSC has ratified the AI contribution policy and it needs ongoing ownership.
- At least one pilot has shown value worth keeping as permanent project infrastructure.
- Enough regular contributors exist to staff a SIG's chair, co-chair and maintainer roles.
If those conditions aren't met, the WG will recommend where its outputs should live instead. For example, the policy could live with the TSC and the tools with SIG-Operations.
A note on process
O3DE has no written procedure for creating a working group I could find. We have created a mobile WG in the past, The TSC charter allows WGs, but the steps have so far existed only in meeting discussion. This proposal follows the steps the TSC has described: a written charter, a presentation at a TSC meeting, feedback, and ratification. The TSC may want to adopt this document's structure as the template for future WG proposals. The SIG-leader runbook requested on 2026-10-27 is a natural place to record that template.
Appendix: strawman policy principles
These are a starting point for the WG's first policy draft, not a decision.
- A human is responsible. The person who signs off a contribution under the DCO owns it, whatever tools produced it.
- Understand what you submit. Contributors must be able to explain and defend every change in their PR.
- Disclose meaningful AI assistance in the PR description. Disclosure isn't a reason to reject a PR.
- Keep PRs reviewable. One issue per PR. Large mechanical changes are allowed when labeled as mechanical.
- Respect licensing. Follow the Linux Foundation Generative AI Policy: the tool's terms must allow contributing its output, and any third-party material it reproduces must be properly licensed and attributed.
- AI assists reviewers but never replaces them. AI reviews are advisory, and a human approves and merges.
- Dominant language
- No language data
- Stars
- 6
- Forks
- 5
- PR merge metrics
- No merged PRs in 30d
Getting set up
This project ships no dev container, Dockerfile or contributing guide, so setting up is up to you: start from its README, and see our first-contribution guide for the general steps.
First steps
- Read the whole issue, then the project's contributing guide.
- Comment on the issue to say you are picking it up — it saves two people doing the same work.
- Fork the repository and make your change on a branch.
- Open a pull request that references the issue number.
More from o3de/tsc
-
Difficulty 5/5 Over a week Newbie friendliness 15/100
-
Difficulty 5/5 Over a week Newbie friendliness 20/100
-
Difficulty 5/5 Over a week Newbie friendliness 20/100
-
Difficulty 5/5 Over a week Newbie friendliness 25/100
-
Difficulty 5/5 Over a week Newbie friendliness 25/100
Similar issues
-
Difficulty 1/5 Under an hour Newbie friendliness 82/100
pollen-robotics/reachy_mini_conversation_app#611 ·
Maintainers usually reply within 1 day
-
Difficulty 1/5 1-3 hours Newbie friendliness 91/100
aws-samples/sample-per-model-bedrock#913 ·
Maintainers usually reply within 1 day
-
lesson-submission pending-review
Difficulty 1/5 Under an hour Newbie friendliness 72/100
Ikalus1988/MisakaNet#5115 ·
Maintainers usually reply within 1 day
-
[ISSUE][DOCS]Openneeds-triage
Difficulty 1/5 Under an hour Newbie friendliness 69/100
Maintainers usually reply within 1 day
-
Difficulty 2/5 1-3 hours Newbie friendliness 73/100
Netcracker/qubership-ai-packages#266 ·
Maintainers usually reply within 1 day