The Studio The flow Present

One human CEO.Ten AI departments.

An AI studio built the way people build companies: a Chief of Staff to run the day, and ten departments, each with a head, specialist AI workers, each with written instructions for its job, and the outside tools a real department would use.

Walk through the departments

Built like a company

One person decides. One desk keeps the day in order. Ten departments of specialists do the work.

  1. 1

    The CEO

    One human. Answers the questions, reviews every mock, approves every gate.

  2. 1

    Chief of Staff

    Routes every new issue to the right department, keeps the day in order, and sets up each new product's repository and folder.

  3. 10

    Department heads

    Ideation to Marketing. Each head plans the work and asks the CEO.

  4. 46

    Specialists

    Claude sub-agents, each on the model that fits its job.

  5. 69

    Skills

    All the studio's written playbooks: how each head and specialist does the work.

  6. 11

    Outside tools

    Godot, ComfyUI, Fastlane, Firebase and more: what a department would otherwise buy in.

The flow

From an issue on GitHub to the app on the CEO's phone, in ten steps.

  1. The CEO lodges an issue

    An idea, a bug or a feature goes into GitHub as one issue, in plain words. No labels, no template.

  2. The Chief of Staff routes it

    It rewrites the issue into a clear title and checklist, then labels who owns it and who needs to know.

  3. The head asks the CEO

    The department head breaks the work down and asks quick multiple-choice questions, so the CEO never has to write a brief.

  4. Mock first

    A mock goes to the CEO to refine. Nothing is built until the CEO is happy with it.

  5. Specialists build it

    Specialists build it once, for Android first. The same code can later ship on iPhone and Steam.

  6. Departments tell each other

    Design tells Marketing what changed; Game mechanics tells Balancing. The right desk hears it without the CEO passing notes.

  7. Quality checks

    Build, tests, lint and review run on every change, and all of them have to pass before anyone asks the CEO.

  8. The CEO approves

    The CEO sees the finished change and approves it, or sends it back with one more question answered.

  9. Straight to the phone

    The test version lands on the CEO's phone over Wi-Fi, no cable. Once the store account exists, every build also goes to Google Play's private testing.

    Play ConsoleInternal testing, once the store is set up

  10. On the CEO's phone

    The app opens on the CEO's own phone, and they try it the way a user will.

Step 1 of 10

Volumes

Three weeks of real work.

issues and pull requests
1,648
pull requests merged
858
issues closed
531
commits
1,732

Across eight apps and the studio itself, from 15 Oct to 12 Nov 2026.

Pull requests

  • Remove an item and re-rank today's outfit

    Wardrobe app · #173 · merged 8 Nov

  • Account deletion page on Firebase Hosting

    Fishing app · #292 · merged 8 Nov

  • Shorter tower range, more and cheaper towers, kill gold x1.5

    Game · #476 · merged 7 Nov

  • App Store listing: attach the newest build to the editable version

    Day-out app · #190 · merged 7 Nov

Issues

  • Fallback rows get a source tap-through and licence link

    Fishing app · #295 · opened 8 Nov

  • Location plugin is never built or packaged, so approximate location can't work on Android

    Wardrobe app · #163 · opened 8 Nov

Where they stand

Eight apps, two stores.

How many of the studio's apps sit at each stage, on Google Play and the App Store.

Android
2
1
2
3
iPhone
4
0
2
2

Counted from each app's store tracks on 12 Nov 2026.

From people using it

Less context. More projects at once.

“The aiceo tools and workflows are redefining AI development. By pulling everything into GitHub, it's easier and faster to plan and work on my projects, and it's made it possible to actively work on several at once without blowing up my usage limits.”

Jared ChaytorUses aiceo

“Most AI setups drag the whole company into every task, and you pay for it in tokens. aiceo only sends the department whose job it actually is, with the few rules that matter right now. Everything lives on GitHub issues, so a session reads one issue instead of relearning the whole project.”

Zane RastrickUses aiceo

How a session runs

Pick the departments. Clear. Start again.

The CEO doesn't talk to ten departments. Every session starts with the Chief of Staff, and every session ends clean.

  1. Start with the Chief of Staff

    It routes new issues, then asks the CEO everything waiting on them, one multiple-choice question at a time.

  2. The CEO picks the departments

    One question: which departments work this session. The busiest unblocked queue is recommended, or run every department with work, in order.

  3. Each head works in turn

    The chosen heads run one after another on their own queues. Between heads, the Chief of Staff writes down where things stand.

  4. Tidy up after every task

    When a change merges, its spare copies and finished branches go. Each product keeps its main line and at most one change waiting for review. Anything unfinished is the CEO's call.

  5. Clear and start again

    With the handoff saved, the CEO clears the conversation and starts fresh through the Chief of Staff. Every session begins clean, and nothing is lost.

Not agentic programming. A company.

Agentic programming
A studio built like a company
The unit of work is a prompt to a coding agent.
The unit of work is an issue, owned by a department.
A developer steers the agent, turn by turn.
The CEO makes the decisions. The Chief of Staff and the heads run the work.
It writes code.
It builds the whole product: idea, design, art, balance, marketing and release.
It knows what the model knows, plus the prompt.
Every role has written skills, the way a company writes down how each department works.
Its memory is one conversation that fills up and drifts.
Its memory is handoffs and GitHub, so every session starts clean.

Every department is on call.

The CEO decides. The Chief of Staff, on call at the lit centre, routes each issue to one of ten heads, and the head runs its specialists. Open a department to see who works there and the skills they use.

Ten departments · 46 specialists · 69 skills

The breakdown, from the CEO to every skill
  1. The CEOThe one human. Answers questions, reviews every mock, approves every gate.
  2. Chief of StaffRuns each session and routes every issue to one department head. For a new product, it checks the CEO's computer has the tools the build needs, then creates the private repository, clones it into a folder on that computer and switches the studio on inside it, all after the CEO approves the plan. It can also sweep every product for anything a finished task left behind.
  • What the CEO seesNo mock: the review round is always a round of pointed questions from the department head, run in the session. The second round shows a quality score and an originality check alongside the draft brief.

    • Head of IdeationOpusstudio-ideation

      Grills the CEO on each idea, has it scored and checked for originality, and hands an approved intent to Spec.

        • studio-ideation:grill-meGrills the CEO one multiple-choice question at a time until an app or game idea has no open branches, then writes the approved intent.
          • Outside toolgrill-me methodOne-question-at-a-time interviews that shape every idea
        • studio-ideation:idea-scorecardScores an app or game idea against the studio rubric and gives a go, reshape or park verdict.
        • studio-ideation:originality-checkChecks an idea against the references the CEO named and the closest store comparables, lists what to steal and what to avoid, and flags name, trademark and copying risks.
      • Decision recorderHaiku

        Writes answers into intent.md and the issue

  • What the CEO seesThe spec outline and milestone list, refined with the CEO before the full spec is written.

    • Head of SpecOpusmobile-product

      Shows the CEO a spec outline and milestone list to refine, then has the spec, plan and backlog written and fans issues out to departments.

      • Outline only

        • mobile-product:product-ownerTurns an app idea into approved intent, spec, plan and roadmap docs, then GitHub milestones and self-contained issues, interviewing one question at a time.
      • Spec writerOpus

        mobile-product:product-owner

      • Milestone graphFable

        mobile-product:product-owner (plan step) across departments

      • Issue formatterHaiku

        Issue bodies and labels per issue-templates.md

        • mobile-product:documentationWrites and maintains an app repo's README, docs index, CONTRIBUTING, CHANGELOG and code docs, and keeps docs matching the code.
  • What the CEO seesThe greybox build on the CEO's device. Before building, grill what the proof must show and what may be faked.

    • Head of Proof of conceptOpusstudio-poc

      Agrees with the CEO what the proof must show, runs engine spikes and greybox builds, and has the CEO try the build on a device.

        • studio-poc:engine-spikeAnswers one risky technical question with timeboxed throwaway builds of 2 or 3 candidates, measured on a device, and writes docs/poc/spike-<slug>.md with a recommendation.
        • studio-poc:greyboxBuilds the first playable or clickable greybox of the core loop, installs it on a device and posts a short recording for the CEO, with docs/poc/greybox.md listing what it proves and what is fake.
  • What the CEO seesFor an app, one function-first direction, built on the design system, which always comes before any screen. For a game, 2 to 3 concept directions, never a safe, bold and wildcard ladder. Placeholder art follows in the chosen direction, and ComfyUI runs only after the CEO picks, starting with a sample batch. Restyled screens arrive as before and after screenshots, only after a visual audit and an independent UX review pass.

    • Head of DesignOpusmobile-design

      Slop first: every plan, mock and build is checked against one anti-slop list before any other review. Every mock the CEO sees carries a UX critique score, and a design gate blocks device installs and pull requests on a restyle until the visual audit and an independent UX review pass. Every screen gets a designed loading, empty, error and offline state. Keeps Marketing and Audio told.

      • Tokens before screens, anti-slop skills picked by situation, Material 3 and Liquid Glass guides, a WCAG 2.2 AA checklist

        • mobile-design:design-systemCreates an app's design system as tokens.json and docs before any mock, with a function-first first direction and a quick-mock mode built on a Material 3 screen kit.
          • Outside toolClaude DesignDesign systems and mockups, first pass of every look
          • Outside skillsSlop.md, frontend-design, UI/UX Pro MaxAnti-slop checks, with a table of which to load for each situation
        • mobile-design:art-directionSets a game's or app's art direction, the asset rules and code-drawn placeholder art, after a CEO-picked Claude Design style board.
          • Outside toolClaude DesignDesign systems and mockups, first pass of every look
      • CEO's Mac

        • mobile-design:comfyui-pipelineTurns approved placeholder art into painted art with the CEO's local ComfyUI, sample batch first, then atlas by atlas into the painted folder.
          • Outside toolComfyUI (local)Painted art from approved placeholders, on the CEO's own machine
        • mobile-design:asset-lintChecks art files against design/asset-rules.json and that painted art mirrors the classic sources with the same silhouettes.
        • mobile-design:design-handoffGenerates Compose and SwiftUI themes from the approved design/tokens.json and restyles the proof-of-concept screens with them, then audits the built screens against the current app and runs an accessibility pass.
        • mobile-design:web-directionSets a website's art direction: concept directions as one-screen mocks for the CEO, then browser checks at three widths once the page is built.
          • Outside skilltaste-skillWebsite art direction and a real-browser taste check
        • mobile-design:ux-critiqueScores screens and flows on a weighted 10-part UX rubric and writes the top fixes. Nothing reaches the CEO below 80 or with a blocker.
  • What the CEO seesAn architecture sketch before structural work, or a screen recording of the change on a device before close.

    • Head of EngineeringOpus

      Shows the CEO an architecture sketch or a screen recording to refine, then has the code written, reviewed and watched through CI. Every screen shows one clear state, loading, empty, failed or ready, gives up waiting after 15 seconds, tries again when the phone reconnects and shows an offline banner.

      The head also draws on

      • local-ciSets up make targets that run the same locally and in GitHub Actions on a self-hosted Mac, with quiet logs and required checks.
      • performanceMeasures and improves app startup, jank, memory and size against budgets and Play vitals, using Baseline Profiles, Macrobenchmark, R8, XCTest metrics and MetricKit.
      • securitySecures KMP apps to OWASP MASVS v2, covering secure storage, TLS, tokens, permissions, privacy manifest and Play data safety.
        • mobile-engineering:architectureDesigns a KMP app's architecture and writes docs/architecture.md.
      • app-factory:implementer when installed, else the skill the issue names (kmp, api, testing, ...)

        • kmpScaffolds and builds Kotlin Multiplatform apps with shared logic, Compose on Android and SwiftUI on iOS.
        • apiImplements client-side networking in shared Kotlin with Ktor, covering docs/api.md, DTOs, token refresh, retries, error mapping and MockEngine tests.
        • testingPlans and writes coordinated unit, snapshot and UI tests for KMP apps, with one shared matrix, shared test tags and journey ids across Android and iOS.
      • Reviewers

        The stage reviewers in quality-loop

      • GitHub monitorHaiku

        CI status and failed-job summaries

  • What the CEO seesA rule sketch with a headless simulation run or a greybox clip, refined with the CEO until it feels right.

    • Head of Game mechanicsOpusgame-mechanics

      Shows the CEO a rule sketch with a simulation run or greybox clip to refine, then has the rules, data and harness built and Balancing told.

        • game-mechanics:godot-architectureDesigns a Godot 4 game's architecture and writes docs/architecture.md.
          • Outside toolGodot 4Game engine for rules, simulation and builds
      • Rule sketch

        • game-mechanics:data-driven-rulesTurns a game mechanic from the design document into a rule sketch the CEO plays with, then into game/data JSON, a JSON Schema, cross-file validator rules and a typed engine-light core class.
        • game-mechanics:sim-harnessBuilds and runs a Godot game's headless simulation harness - bot players at set skills, seeded runs over every level, one JSON line per result and a report that balancing and review rounds read.
          • Outside toolGodot 4Game engine for rules, simulation and builds
      • Sim runnerHaiku

        Runs harness batches and summarises the JSON lines

  • What the CEO seesThe simulation scorecard before and after the proposed change, with the levels it moves. Grill the CEO on difficulty intent when the spec doesn't settle it.

    • Head of BalancingOpusgame-balancing

      Shows the CEO simulation results before and after each proposed change, runs the autopilot campaign, and keeps the economy ledger even.

        • game-balancing:balance-specWrites a game's balance spec with the CEO - difficulty intent per level and world, graded simulation targets, the change ladder, never-do rules, a short rules digest and per-level briefs.
      • Campaign plannerFable

        Plans an autopilot campaign across levels and reads its results

      • One fix per session

        • game-balancing:balance-autopilotGrades a game's sim report against docs/balance/targets.json, ranks a balance backlog, then fixes one item per fresh session and shows the CEO the scorecard before and after.
      • Sim runner and scorecard parserHaiku

        Runs batches, grades and ranks the backlog

        • game-balancing:economy-ledgerKeeps a game's economy ledger - every currency's sources and sinks, designed supply against demand per player profile, and what campaign sims actually earn, spend and leave unspent.
  • What the CEO sees2 or 3 short samples per sound or theme. The CEO picks or asks for another round; nothing is produced in bulk before a pick.

    • Head of AudioOpusstudio-audio

      Plays the CEO 2 or 3 sample options per sound or theme to choose and refine, then has the approved cues produced to the mix rules and Marketing told.

        • studio-audio:audio-directionSets a product's sound identity and the cue list, after grilling the CEO and playing 2 to 3 sample options per key sound.
        • studio-audio:sfx-music-pipelineProduces a product's sound effects and music from the approved cue list, with a licence ledger, loudness normalising, platform formats and an audio lint.
      • Loudness and format checkerHaiku

        Checks loudness, formats and the licence ledger

  • What the CEO seesThe draft listing, screenshot set and release notes for the store and track in question.

    • Head of PublishingOpusmobile-release

      Shows the CEO the draft listing, screenshots and release notes to refine, then ships to Google Play, the App Store or Steam behind the CEO's approval.

        • mobile-release:listing-asoWrites store listing text and compliance answers as fastlane metadata, optimised for search.
          • Outside toolFastlaneBuilds, signing and store uploads for Android and iOS
        • mobile-release:store-assetsProduces store visuals into fastlane folders.
          • Outside toolClaude DesignDesign systems and mockups, first pass of every look
          • Outside toolFastlaneBuilds, signing and store uploads for Android and iOS
      • Play, App Store

        • mobile-release:fastlaneBuilds, signs and uploads app binaries with Fastlane via make release.
          • Outside toolFastlaneBuilds, signing and store uploads for Android and iOS
        • mobile-release:steam-releaseShips a game on Steam via make steam-upload.
          • Outside toolSteamworksSteam store pages, depots and releases
      • mobile-release:player-wiki, mobile-release:press-kit

        • mobile-release:player-wikiBuilds a player-facing wiki for a game generated from the game's data files so numbers never drift, with a spoiler policy and per-release updates.
        • mobile-release:press-kitAssembles a press kit for an app or game as a page and a zip journalists and creators can use.
      • Localiser and release notesHaiku

        Listing translations, what's new text

        • mobile-release:web-hostingShips a website on Firebase Hosting's free tier: encodes clips and images, budgets monthly transfer and hands the CEO the deploy command.
          • Outside toolFirebase HostingWhere the websites live
  • What the CEO seesDraft posts per platform and the launch calendar. Grill tone, platforms and what to show when the plan doesn't settle them.

    • Head of MarketingOpusstudio-marketing

      Shows the CEO draft posts and the launch calendar to refine, and keeps every post a draft until the CEO publishes it.

        • studio-marketing:marketing-planWrites the marketing plan for an app or game after a CEO review round.
        • studio-marketing:launch-calendarBuilds the launch calendar for an app or game, backwards from the release date, with store milestones, post slots per platform and asset due dates.
      • Masters

        • studio-marketing:platform-postsDrafts platform-native social posts, Steam events, dev logs and community replies from the marketing plan, one master per message with variants per platform, for the CEO to approve and post.
      • Variant writerHaiku

        Per-platform variants and community reply drafts

        • studio-marketing:websiteLeads a studio or product website: grills the CEO on its job and sections, gets the look from Design and hosting from Publishing, and builds the page from the approved mock.

One day, one design pass

The same day-out app, yesterday and today. In between, the Design department built a new design system first, then restyled every screen and checked each one against the app it replaced.

Home, with photos

Yesterday: a photo card with a pale mint score tile and conditions as pills.
Yesterday
Today: the photo card with a solid green score badge, a swim verdict under the name, the photo credit and Directions.
Today · after one design pass

Home

Yesterday: a flat mint score block over a grey card, with conditions as pills.
Yesterday
Today: a drawn water scene with a solid score badge, conditions on one line and Directions on the card.
Today · after one design pass

A place with a photo

Yesterday: a thin score floating over the photo, with a small Directions pill.
Yesterday
Today: a bold score and a rating pill over the photo, a full-width Directions button and bars showing what holds the score up.
Today · after one design pass

A place without a photo

Yesterday: a large score on a plain mint panel with a diagonal cut.
Yesterday
Today: the same drawn water scene as on Home, so places without a photo still look finished.
Today · after one design pass

Settings

Yesterday: a plain list of settings rows on the page background.
Yesterday
Today: the same rows grouped into rounded cards, with a line saying your location stays on your phone.
Today · after one design pass
    1 / 5
    See this app on Google Play

    The design we nearly shipped

    Mockups for a fishing app. The first set was heavy type, a dark hero and orange accents, and it nearly went ahead until the CEO sent it back as AI slop. The redo starts from what someone on a wharf is asking, and answers it in a sentence.

    A fishing spot

    Nearly shipped: a huge clock time on a dark hero block, with a cut-off card of bait and rig details.
    Nearly shipped
    What replaced it: the day's tides in one row, then a large photo of a snapper with its bait, rig and rod.
    What replaced it

    Map

    Nearly shipped: numbered pins, an oversized clock time in a dark card, and orange highlights.
    Nearly shipped
    What replaced it: quiet pins on the map and one plain answer under it, High in 1 h 15 m, with the likely fish.
    What replaced it

    Favourites

    Nearly shipped: a dark hero card over mismatched cards, all shouting clock times.
    Nearly shipped
    What replaced it: two matching cards, each saying when the next high tide is, with a tide alert button.
    What replaced it
      1 / 3

      One app, two platforms

      The latest Android and iPhone builds, side by side. Material on one, Liquid Glass on the other.

      Search

      Android: the list of places scrolled, with the location, filter chips and sort in a solid bar above a beach photo card, and a flat navigation bar below.
      Android
      iPhone: the same list scrolled, with glass filter chips floating over the beach photo and a glass tab bar over the cards below.
      iPhone

      A place

      Android: a beach photo with a score of 86, plain back and heart icons, a Directions button in the page and a flat navigation bar.
      Android
      iPhone: the same beach with round glass back and heart buttons, a floating Directions button and a glass tab bar.
      iPhone

      Choose an area

      Android: Choose an area as a full-height sheet, with a map of New Zealand and the list of regions.
      Android
      iPhone: Choose an area as a half-height glass sheet, with the list and its photos blurred behind it.
      iPhone
        1 / 3

        Every screen is scored out of 100

        A separate reviewer scores each screen on ten parts of good design. Below 80 it goes back to Design. At 90 it can ship to the stores.

        The designer never scores its own work.

        The outcome

        Real footage from all four worlds, recorded straight from the game.