0620: Creator-Operator Long-Running Agent Case
July 12, 2026 ยท View on GitHub
Summary
This case describes a public-safe creator-operator workflow for a non-technical user who wants a long-running agent to help with content research and planning.
This is an appendix case for product direction. It should not be treated as a frontstage top-card proof until a real user story or approved public evidence exists.
The user is not trying to operate an agent framework. They want a controlled work loop that can keep a creative goal moving:
- detect public trend candidates;
- map those candidates against personal creative preferences;
- extract reusable insights;
- draft article, short-video, or newsletter angles;
- maintain a material, phrase, source, and copy library;
- ask for human taste, risk, or publishing decisions at the right time.
The case is synthetic. It does not contain real platform data, private notes, creator drafts, audience metrics, screenshots, raw browsing traces, internal links, or performance claims.
Situation
A creator-operator has recurring medium-horizon work:
- Check what changed across public conversation spaces.
- Decide which topics fit their own taste, expertise, and audience.
- Turn the selected topics into useful creative angles.
- Keep a library of source snippets, examples, hooks, and reusable phrases.
- Review drafts before anything is published.
Without a control plane, the experience is awkward even if the underlying agent is capable:
- the user has to ask "what did you do?" after every run;
- research, draft, and material-library work blur together;
- private notes and public evidence can be mixed by accident;
- a publishing decision can block the whole loop;
- feedback such as "not my style" is easy to lose in chat;
- the user cannot tell whether the agent is safely continuing side work or waiting for a required decision.
The bottleneck is product experience, not only model ability.
LoopX Behavior
LoopX turns the workflow into visible control-plane objects:
| Workflow concern | LoopX object |
|---|---|
| Long-running creative objective | goal state |
| "Can this be published?" | user gate |
| Research, insight, draft, and library work | agent todos |
| "This is not my style" | feedback signal / preference hint |
| Synthetic demo data versus private user material | boundary note |
| What changed since last check | run history summary |
| Safe work while a publishing gate waits | safe side path |
| Agent identity and lane ownership | soft claim / optional hard lease |
The important behavior is gate-aware continuation. The agent should not autopublish or treat private drafts as public evidence. But while a publishing decision waits, it can still do safe side work: organize synthetic examples, prepare source-attribution checklists, update the material-library schema, or draft questions for the user.
Public-Safe Walkthrough
The following walkthrough uses only fake data.
1. Trend Candidates
The agent proposes three synthetic trend clusters:
| Candidate | Why it might matter | Boundary |
|---|---|---|
| "AI note workflows for solo operators" | aligns with productivity and agent-tooling audience | public-safe synthetic topic |
| "Short-form content from long research notes" | fits creator-operator reuse loop | public-safe synthetic topic |
| "Human approval before agent publication" | matches safety and trust framing | public-safe synthetic topic |
2. Preference Map
The user has compact preference hints:
- prefers practical case studies over broad futurism;
- avoids growth-hack language;
- wants evidence boundaries stated plainly;
- likes examples that show what the agent will not do.
These are preferences, not hard safety gates. LoopX should keep them separate from permission decisions such as publish/no-publish.
3. Insight Board
The agent extracts draft insights:
- "A long-running creator agent needs a dashboard, not another hidden prompt."
- "The user gate is not a failure; it is the product boundary."
- "Safe side paths keep research useful while publishing waits."
Each insight should carry source status: synthetic, public-source summary, private note, or needs review. Public docs may only use synthetic or public-source summaries.
4. Draft Queue
The agent prepares draft angles:
| Draft angle | Status | Gate |
|---|---|---|
| "How I keep a research agent from waiting forever" | outline ready | tone review |
| "What a creator agent should show before it publishes" | idea only | publish policy |
| "Material libraries as memory for creative work" | source map needed | no publish yet |
The user sees what is ready, what is blocked, and what can continue.
5. Material Library
The agent maintains structured public-safe material:
- reusable hooks;
- source summaries;
- phrasing examples;
- rejected angles and why they were rejected;
- boundary notes for private drafts or unpublished ideas.
This library is not a raw memory dump. LoopX should keep it governed by source status and user feedback.
6. Human Feedback
The user can respond with structured feedback:
| User feedback | Control-plane effect |
|---|---|
| "This angle is useful." | reward / preference reinforcement |
| "Too salesy." | preference hint and draft revision todo |
| "Do not use this source." | boundary correction and source removal todo |
| "Publish after I review tone." | user gate with explicit stop condition |
| "Keep researching, but do not draft yet." | todo reprioritization |
The feedback changes the next plan without becoming an invisible chat-only memory.
The public feedback and source-status rules are specified in creator-ops-feedback-boundary-contract.md.
User-Facing Value
For a creator-operator, the value is not "the agent can browse and write." The value is that the work remains legible:
- the user knows what changed since the last check;
- publishing gates stay explicit;
- private material stays out of public examples;
- safe side work continues while gated work waits;
- feedback becomes a durable planning signal;
- the agent's next move is visible before it spends another run.
For a potential LoopX user, the reusable pattern is broader: a long-running agent loop becomes easier to trust when goals, gates, todos, evidence, boundaries, feedback, and next actions appear in one control plane.
Evidence Boundary
This case is a synthetic product case spec. It intentionally excludes:
- real creator notes, drafts, screenshots, or audience data;
- raw social-platform content or scraping output;
- private user preferences that were not rewritten as synthetic examples;
- internal links, private repositories, local paths, credentials, or raw agent sessions;
- claims about reach, quality, engagement, or revenue improvement.
Future demos should use fake data or public-domain sample material and should keep no-autopublish gates visible.
Public Evidence Sequence
- A creator-operator has a long-running research and content-planning goal.
- The agent proposes trend candidates, maps them to preferences, and drafts insight options using synthetic data.
- LoopX separates the publishing gate from safe side work.
- The user gives feedback that becomes structured control-plane state.
- The next agent run starts from visible goals, todos, boundaries, and feedback instead of a hidden chat transcript.
Demo Status
The first public-safe storyboard is available in creator-ops-fake-data-storyboard.md. It defines a synthetic fixture and frontend panel sequence for trend candidates, preference map, insight board, draft queue, material library, human feedback, and controlled replan. No executable demo is included yet. The companion feedback contract is creator-ops-feedback-boundary-contract.md.