Overview
The client, a founder building an audience across several platforms, needed a steady stream of content for X, LinkedIn, YouTube and a blog. Producing it by hand every week took far more time than the client had, and generic AI tools produced posts that did not sound like them or follow their themes.
Voxen built a content system run by a team of AI agents, with a control dashboard where the client sets direction, reviews every batch and approves what goes out.
The challenge
- A full week, every week. Each platform has its own format, length and style, and all of it had to be produced on schedule.
- On-brand and on-topic. Content had to follow the client's content pillars, reflect their own views, and react to news while it is still current.
- Reliable enough to publish. AI output often has missing posts, placeholder text or posts that are too long. None of that could reach the client.
- The client stays in control. Nothing could be published without approval.
What Voxen built
A swarm of specialised agents.
Built on Google's Agent Development Kit, each agent has one job: research, planning, writing or quality control. Agents hand work to each other in a defined pipeline rather than one model trying to do everything.
A weekly content pipeline.
Every week the pipeline researches topics with citations, plans the week around the client's content pillars, and writes platform-specific posts for X and LinkedIn plus scripts, titles and descriptions for YouTube.
Content pillars with weights.
The client defines their pillars, such as tokenization, fundamentals, news and industry topics, and sets how much of each week each pillar should take. The planner follows those weights.
The client's own voice.
The client records or uploads voice notes. They are transcribed and fed into research, and the client sets how much weight their own notes carry against external research.
News reactions.
A daily pipeline runs on weekdays, picks up breaking stories and turns them into timely reaction posts.
Long-form blogs.
A separate pipeline writes SEO blog posts with branded hero images and diagrams, and publishes them to the client's content system once approved.
Quality control in code.
Every batch is checked for the exact post count, placeholder text and platform length limits. Anything that fails is rewritten automatically, up to three times, until it passes.
A control dashboard.
The client sees today's queue, approvals and next publishing slot, reviews each week's batch by day, approves or rejects individual posts, and exports approved content by platform or by day. Run visibility shows every pipeline run with its status and trace, schedules are set per pipeline, and any run can be triggered manually.
Engineering approach
Scheduling, retries and run history live in the backend, so every run is traceable and repeatable. The system is built to work with multiple AI model providers, so no single vendor is a dependency.
Outcome
One pipeline run produces a complete, review-ready week of content across X, LinkedIn and YouTube, with blog posts ready to publish and every item approved by the client before it goes out.
What it makes possible
New platforms, pillars or content types are added as new agents or pipeline steps, without rebuilding the system.