I designed and built the Creative Factory, an AI system that automates creative drafts, fact-checking and compliance checking. I packaged it as a company plugin, so a lean marketing team could publish at agency volume across formats.
The CEO and founder wanted content at scale: blogs, newsletters, social, video, ads and app screens, from a lean marketing team.
Faster drafting alone would only grow the review queue. So I built a system: skills that research, write in each executive's voice, design, edit video and publish, packaged so anyone on the team could install it.
A lean team, agency volume, formats from blogs to in-app video, and each word an SEC-regulated advertisement.
My rule for the system: language models are probabilistic, and business logic is deterministic. Each skill pairs written instructions with tested Python, so file handling, document edits and logs run as code. The AI handles judgment: research, voice and rewrites.
Social posts and videos are repurposed from blog content, so one round of research feeds each channel. Ads run from pain-point research and campaign briefs, and landing pages follow the ads.
I built it as a plugin and AI harness that users import once and that updates automatically. Reviewers open the Google Docs it writes with nothing to install.
I ran a live end-to-end test of the system and wrote up each defect in a detailed hand-off. That became a shared bootstrap for the scripts, a regression guard and a test suite. I also spotted newer models going beyond the skills' instructions in production output, and flagged them for recalibration.
Five stages and three layers. Each pipeline, from a blog cluster to an in-app video, runs on the same skills and ends in the same place.
Topic clusters, keyword and answer-engine research, competitor intelligence and campaign briefs.
Brand and executive voice writers, video scripts and editing, thumbnails, ad creative, social visuals and landing pages.
Each pipeline ends with a fact-check gate and an SEC Marketing Rule gate, then human approval with the audit trail in the document.
The gates, case file 04 →Filed to the shared review drive, then live on the website behind marketing-controlled release flags, in social and in the app.
A competitor watchlist, pattern library and periodic reports, plus a loop that brings what performed on owned accounts back into the skills.
A written procedure per skill: goal, inputs, tools, outputs and edge cases.
The agent reads the procedure and calls tools in order. One orchestrator runs a whole topic cluster: research, writing, verification and upload, article by article.
Deterministic Python in each skill's scripts folder and a shared library, so the same input gets the same handling each time.
Skills in the company plugin, as it grew
Skills I wrote or adapted, plus open-source skills I selected and integrated for marketing craft, HTML-to-video rendering, avatar video and research.
A private repo, a setup script and a quickstart that gets a tester running in minutes.
Users import the plugin once, and each update arrives automatically. A multilayer memory system carries context from one session to the next.
Blogs, executive newsletters, social campaigns, in-app videos, ads, competitor reports and the Financial Decisions Study package, from one plugin.
Files created in the factory's shared workspace
Source: the shared drive listing. Growth is measured against the month it shipped.
Cluster planning, search and answer-engine research, the blog writer and both gates: 46 long-form drafts, about 116,600 words, built around the app's features.
An orchestrator picks the keyword, then the right executive voice writer, with validators for length and hooks, written in the CEO's voice and the CTO's.
A social series on AI in your financial life, a launch campaign and carousels, rendered against the brand stylesheet.
First scripts about a day after the ask and dozens more within days. The first 13 finished videos started the recording pipeline, each with a re-edit guide and a branded thumbnail.
The Financial Decisions Study's flagship report, mini-report, regional and topical briefs, infographics and press kit: 34 assets, each with its own check documents.
I built a competitor and creator watchlist, a pattern library of what wins, and weekly or monthly reports that turn into briefs.
Hands-on use was marketing, the CEO's office and compliance, and colleagues across the firm got access and training.
Marketing ran content through it daily, from blogs and newsletters to ad imagery, video and landing pages.
The CEO and founder sent his own drafts through it for accuracy and compliance, and got access to its shared workspace.
The Chief Compliance Officer used it in review, and an outside compliance advisor also reviewed the process.
Colleagues across the firm got access to the system and training on AI workflows.
This is an important step forward toward our strategic goal to automate the creation of content and code using human-designed systems and criteria.
The CEO and founder
Build the system, not just the content. Encode the voices and the process as skills, run what must never vary as code, and ship it where people already work. Then a small team publishes like an agency.