She Went From "Using AI" to Running a Content Production Line — In One Hour
Last week, someone close to me asked for help building a syllabus for a course she teaches. She’s smart, creative, great at her craft. She’d been using ChatGPT the way most people do — ask a question, get an answer, then go do everything else manually.
One hour later, she had a fully automated content pipeline: structured syllabus, AI-generated presentations, animated visuals, video-ready materials with sourced references embedded. Not a rough draft. Production-grade output.
She didn’t learn to code. She didn’t buy new software. She just stopped using AI as a search engine and started using it as a production line.
Here’s exactly how we did it — detailed enough for you to replicate today.
The Trap: AI as a Smarter Google
Most people’s AI workflow looks like this: open ChatGPT, ask a question, copy the answer, manually build everything else.
Most people fall into the AI productivity trap — using AI faster without changing the workflow. They gain speed but not leverage. Same architecture, just quicker input.
That’s level one. It’s useful. It’s also a ceiling.
The problem isn’t the AI — it’s the architecture. You’re using a jet engine to power a bicycle. One prompt, one output, one manual step at a time.
The leap happens when you stop asking AI to help you — and start asking AI to build the thing.
The Pipeline We Built in 60 Minutes
Step 1: Structure the Knowledge — Claude Projects (15 min)
We took her raw course material — lecture notes, topic outlines, scattered Google Docs, reference PDFs — and uploaded everything into a Claude Project (claude.ai → Projects → New Project).
Why a Project and not a regular chat? Because Projects give Claude persistent context. It reads all your files as background knowledge, so every conversation in that project is grounded in your material — not generic internet knowledge.
What we uploaded: her existing course outline (rough, incomplete), lecture notes from previous sessions, a few reference articles she liked, and a document describing her target audience.
Why this works: We didn’t ask Claude to “write a syllabus.” We gave it all the raw material and asked it to find the structure that was already there. The output is dramatically better because it’s grounded in real content, not hallucinated frameworks.
Pro tip: After the first output, we iterated twice. Two follow-up prompts turned a good syllabus into an excellent one. Don’t accept the first output.
Output: A complete, structured syllabus in markdown — 8 modules, learning objectives, session flows, exercises.
Step 2: Generate Presentations — Gamma (15 min)
With the syllabus locked, we moved to presentation generation using Gamma (gamma.app) — an AI presentation tool that creates slides from text input.
We copied Module 1 from the syllabus, pasted it into Gamma’s “Paste in text” option, and — this is the critical step — wrote a creative direction prompt with specific design constraints: minimal and modern, one concept per slide, large typography, no stock photos.
Why most people get bad results from AI presentations: They type “make a presentation about X” and accept whatever comes out. The AI doesn’t know your taste. You have to art-direct it. Think of the prompt as a creative brief.
Output: 8 presentation decks, 12-15 slides each, with consistent design language, speaker notes, and exercise slides.
Step 3: Source Video & References — ChatGPT with Agent Mode (15 min)
We needed each module to include 2-3 short video references, key articles, and specific timestamps. We used ChatGPT’s agent mode (GPT-4 with browsing enabled) to search, verify links, and check video lengths.
Why agent mode matters: Regular ChatGPT can’t browse the web. Agent mode can search, visit pages, verify that links work, and check video lengths. It’s the difference between hallucinated URLs and real, verified references.
We reviewed the output manually — took about 5 minutes to scan, swap one or two, and approve the rest. AI does the research; you do the editorial judgment.
Output: A reference document with 2-3 videos and 1-2 articles per module, with timestamps and relevance notes.
Step 4: Assemble Video-Ready Packages (15 min)
Final step: combining presentations with reference materials into video-ready packages. If you’re technical, tools like Remotion work. If not, export Gamma slides as PDF, use Canva or CapCut to assemble slides into video with transitions, and record a voiceover using speaker notes as a script.
The point isn’t the specific tool — it’s the pipeline. Structured content → visual production → enrichment → assembly.
The Complete Pipeline
| Step | Tool | Time |
|---|---|---|
| Structure | Claude Projects | 15 min |
| Design | Gamma | 15 min |
| Research | ChatGPT Agent Mode | 15 min |
| Assemble | Remotion / Canva | 15 min |
Total: 60 minutes from raw notes to production-ready course content.
The Three Mistakes That Keep People at Level One
Mistake 1: Using one tool for everything. Each AI has a sweet spot. Claude is exceptional at understanding large documents. ChatGPT’s agent mode is best for web research. Gamma is purpose-built for presentations.
Mistake 2: Accepting the first output. AI gives you a solid first draft — maybe 70% of the way there. The magic is in the iteration. Two or three follow-up prompts transform “good enough” into “this is actually better than what I’d make manually.”
Mistake 3: Thinking in prompts instead of pipelines. One prompt = one output. The breakthrough is thinking in stages: structure, production, research, assembly. Each stage feeds the next. That’s not “using AI.” That’s building with AI.
What Actually Changed
The person I helped isn’t a developer. She’s a teacher who creates courses. In one hour, she went from “I use AI to brainstorm, then spend days building everything manually” to “I direct the AI pipeline and focus on what I actually care about — the content and the teaching.”
That mental shift — from operator to director — is available to anyone. The tools are free or nearly free. The pipeline pattern works for any content type.
The gap was never the technology. It was seeing the assembly line instead of the single tool.
Try It Yourself
Pick a project you’ve been doing manually. Break it into stages — structure, production, research, assembly. Match each stage to a tool. Write real creative briefs, not “make me a thing” but specific direction. Chain the outputs so each step feeds the next.
Start small. One project. One pipeline. Once you see the output difference, you won’t go back to single-prompt workflows.
If you want to take it further and deploy an always-on agent that maintains this pipeline without your input, here’s what happens in production — the real cost model, the infrastructure constraints, what actually works at scale.
I’m Eliran — building systems that turn AI from an assistant into infrastructure. More at elirank1.github.io/blog.
