FrankX.AI

Agentic Creator

Run AI as a crew. Keep the receipts.

Agentic Creators work with a brief, a memory and a record of what ran. Pick the role you want to grow into, run its first win with your own agent, and keep the receipt.

See the nine rolesRead the receipt format

Manifesto

Give the crew a memory.

An agent can write a song today and forget by tomorrow why you chose it. Most people use AI like a vending machine: ask, get something, start over.

An Agentic Creator gives the crew a memory. It reads a brief before it starts and keeps what it learned after the session ends. It leaves a record of what ran, what failed and what changed afterward, so any piece of work traces back to a decision you made.

One person can carry more of a studio's work this way. A songwriter can run a catalog with a release gate. An investor can keep a thesis with its sources attached and its kill criteria written before the money moves. A teacher can turn what they know into a path other people finish without them in the room. A builder can set up a machine, a harness and a fleet of agents that still work on Monday morning.

Run the crew with your own keys, on your own disk. Your briefs, receipts and files stay with you, and you can take them anywhere. Start with one brief, one review and one finished piece of work.

Nine roles are on the map: creator, investor, musician, producer, player, rig builder, teacher, orchestrator and curator. Each is meant to begin with a first win and end with an artifact you can show. Roles go live one at a time, once their first win has been tested with a real person.

One loop, nine materials

Every role runs the same loop on a different material: See, Design, Build, Automate, Compound. The roles differ in what the crew works on. The loop stays the same, which is why a person who learns one role learns most of the next.

  1. See
  2. Design
  3. Build
  4. Automate
  5. Compound

The map

Nine roles, each with a first win.

Only GenCreator has a waitlist today. The other eight roles are concepts. Each page says what already exists behind it and gives no delivery date.

Waitlist open

GenCreator

A maker who ships reviewed work with a crew

First win

Take one idea, write the brief, run one draft and one review pass, export the result and the receipt.

Read the role

  1. GenInvestor

    An investor whose every number has a source

    First winTake a thesis you already have. The agent asks five questions, then writes thesis.md with sources, two kill criteria and a review date.

    Concept

  2. GenMusician

    A songwriter with a crew that remembers your taste

    First winOne song from brief to three iterations, with the taste notes saved as a file.

    Concept

  3. GenProducer

    A producer who takes finished tracks to a released catalog

    First winRun one finished track through the release gate and fix the first failing check.

    Concept

  4. GenPlayer

    A player who builds the game while playing it

    First winOne playable mechanic built with your own agent and played once.

    Concept

  5. GenRig

    A builder whose machine, harness and fleet stay healthy

    First winRun the rig check and fix the top item it reports.

    Concept

  6. GenTeacher

    A teacher whose path others finish alone

    First winWrite one lesson with an exercise and a rubric, then run it yourself first.

    Concept

  7. GenOrchestrator

    A conductor of many agents on one memory

    First winTwo agents, one task, one reviewer, one receipt. The maker never reviews its own work.

    Concept

  8. GenCurator

    A collector whose library keeps its provenance

    First winExport one AI chat into a vault note with its source link and one tag.

    Concept

Method

How a first win works

A first win is one small job, run end to end with your own agent. It ends in a receipt.

  1. Pick a role

    Choose the role you want to grow into. Each one names its first win up front.

  2. Run one small job

    Take the role's brief into your own agent, with your own keys. The crew reads the brief, does the work and gets reviewed.

  3. Export the result

    The output and a receipt come out as plain files that you own.

Three rungs, one artifact each

Education here teaches how, and every rung needs an artifact. A quiz does not count.

What a person has done at each rung and the artifact they keep
RungWhat the person has doneArtifact they keep
First winRan one small job end to end with their own agentA receipt: the brief, the steps, the failures, the output
ShippedExplained the workflow and produced the real thingA workflow packet with a failure trace and a revision
CompoundingSolved a changed problem faster using the first oneA second receipt that cites the first, plus a peer review

Progress is a list of receipts you own as plain files. There are no points and no streaks.

Proof

The receipt format

Every first win ends in a receipt, a plain file you own. One format across all nine roles.

receipt.md
role
<one of the nine>
date
<the date>
brief
<the brief the crew read>
steps
<what ran, in order>
failures
<what broke, and what fixed it>
revision
<what changed after the first critique>
output
<file paths or links>
changed problem
<only on the compounding rung>
agent and model
<what ran it>

Ownership

What stays on your computer

You run the crew with your own keys, on your own disk.

  • Your briefs, receipts and files stay with you, and you can take them anywhere.
  • No account is needed to run a first win, and the files stay on your computer. Your agent still sends your prompts to its model provider, and that provider's terms apply.
  • Sharing a receipt is your choice.
  • Circle membership is a separate, optional step.

Independent project. About Frank