Developer-tool project / Submitted to OpenAI Build Week

Wingspan Command Center: A clearer way to direct an AI team

Run many Codex tasks as one visible, accountable AI team.

Wingspan Command Center is a local interface for founders and advanced Codex users coordinating multiple tasks. It brings active work, decisions, evidence, conversations, and bounded usage attribution into one place while keeping human authority visible.

Wingspan Command Center promotional still with the title Wingspan Command Center, the line One accountable AI team, and a fictional Founder dashboard.
YouTube demonstration still showing the judge-safe build with fictional sample data.

The business problem

Capable work can still be difficult to direct.

As multiple Codex tasks move at once, status can scatter across conversations, decisions can lose their context, and a founder can spend more time relaying updates than making choices.

The design question was not how to make agents look busier. It was how to give one person a calm, evidence-aware place to answer four questions.

Four operating questions

  1. 01What is moving?
  2. 02What is held?
  3. 03What needs a human decision?
  4. 04What evidence supports the next step?

The design response

One local shell. Nine connected views.

Each view has a distinct job. The interface connects them without pretending that every action belongs on one dashboard.

  1. 01

    Founder

    See outcomes, active work, pending decisions, and exceptions.

  2. 02

    Operate

    Continue direct task conversations and keep work context together.

  3. 03

    Decide

    Separate recommendations, decisions, approvals, and receipts.

  4. 04

    Mission Control

    Review work, ownership, progress, attachments, and local diffs.

  5. 05

    Discuss

    Run attributed multi-agent roundtables without erasing disagreement.

  6. 06

    Lean Tools

    Frame bounded tests and improvement work.

  7. 07

    Social

    Organize channel work with readable source timing.

  8. 08

    Usage

    Compare bounded evidence for successful turns started in the demonstration.

  9. 09

    System

    Keep operating state and the path back to native tools visible.

How it was built

Kevin set the authority. Codex helped build the system.

Kevin's role

Kevin chose the product structure, the executive shell, the human approval boundaries, direct task continuity, the four-room discussion limit, and the explicit path back to native tools.

Codex and GPT-5.6 Sol

Codex helped implement and test durable operating state, task conversations and recovery, decision receipts, attachments and diff review, handoff controls, four simultaneous roundtable rooms, the isolated judge adapter, and repeatable source-lock verification.

Local verification only

A tested demonstration, with a deliberately narrow claim.

These checks establish what passed in the source-locked local build. They are not customer, market, security, accessibility, or production-reliability evidence.

69/69
application tests passed
47/47
judge smoke checks passed
10/10
responsive captures passed
177
submission-boundary checks passed
Zero
attempted non-local browser requests recorded

What it is, and what it is not

A working local demonstration, not a production claim.

What the build demonstrates

  • A local interface with nine connected operating views.
  • Visible separation among recommendations, decisions, approvals, evidence, and handoff.
  • Bounded usage attribution for successful turns initiated through the Command Center demonstration.
  • A founder-narrated walkthrough using fictional sample data.

What the build does not establish

  • The judge-safe build does not connect to native Codex, local files, repositories, accounts, or external services.
  • It does not establish production readiness, equivalent native-tool behavior, customer validation, market demand, or provider-wide usage visibility.
  • Local checks do not establish security, privacy, accessibility conformance, or production reliability.

Wingspan Command Center was submitted to OpenAI Build Week. That statement describes the submission only; it does not imply selection, award, endorsement, sponsorship, partnership, or official OpenAI product status.

See the work

Watch the project, inspect the submission, or bring us your workflow decision.