Scrape Every Social Platform at Once with AI
In this walkthrough you will turn Claude Code into a social media intelligence engine. With a single prompt, it spins up four specialized agents, one each for X, YouTube, Instagram, and Reddit, scrapes them all at once through Apify, and hands everything to a synthesizer that surfaces what is trending across every platform. The output: raw cross-platform intelligence plus ready-to-twist ideas for your posts and videos.
What you need
- Claude Code, the application where the whole workflow runs.
- An Apify connection. Apify is the platform that gives you the capacity to search through every social media platform out there and find trending information. Connect it to Claude Code before you start; mine is already connected in the demo.
1. What is a dynamic workflow (0:14)
The mechanism behind all of this is called a dynamic workflow, and it is very simple to understand: it is Claude Code using multiple agents at once, in our case one per social media platform. You have a normal conversation with Claude Code, you give it a prompt where you explicitly say you want a dynamic workflow, and it orchestrates the agents for you: X, YouTube, Instagram, and Reddit in parallel, then a synthesizer that merges the four results into ideas.
There are two ways to play it: find what is trending on all platforms at once, or find something trending on one platform and carry it over to the others where it has not landed yet.
2. Writing the prompt (2:00)
One rule matters more than anything else here: you have to explicitly specify that you want a dynamic workflow. If not, it does not work. Here is the exact prompt from the video:
Use a dynamic workflow for the following. Using Apify with a total of four subagents, find trending topics and trending posts on X, YouTube, Instagram, and Reddit. Focus on the AI niche. I want to find out what people are talking about that's trending and any questions they might have about recent topics. Then take those four results, synthesize them, find what they have in common, and ideate some post ideas.
You do not actually need to be that specific. Frankly, with just two sentences Claude Code is smart enough to understand. The keyword "dynamic workflow" is the part you cannot skip.
3. Dynamic workflows vs subagents (3:10)
Once you send the prompt, Claude Code goes through an ideation phase, reasons about the task, then launches the four agents. So what is the difference between this and regular subagents? With dynamic workflows, the agents Claude Code creates are super specialized for the task at hand. With subagents, it uses general purpose agents to accomplish the task. The result: on top of going faster, dynamic workflows also give you better quality.
4. Watching the four scouts run (3:40)
While the workflow runs, Claude Code gives you a visualization. Click it and you see the phases the workflow moves through. In my run, it is in the scout phase: one scout for X, one for YouTube, one for Instagram, one for Reddit, each running for about forty seconds before reporting back. Claude Code returns to the conversation when they finish.
5. Scaling to 99 agents and other use cases (4:11)
Four agents is the floor, not the ceiling. You can run this with as many as one hundred agents, if not more. My personal record right now is ninety-nine. Instead of one agent per platform, you can go per platform and per niche and extract far more information in a single pass.
This pattern is fantastic well beyond content ideas. It works for intelligence in general, seeing what other people are doing on the web. If you have a competitor, you can use a dynamic workflow to scout them out on LinkedIn, X, and beyond.
6. The results: recurring themes per platform (4:57)
When the workflow completes, you can open the raw output of each and every agent if you ever need to verify something. Below that, the synthesizer delivers the good stuff: recurring themes broken down per platform. In my run, one theme showed up on X, YouTube, and Reddit but was apparently not so present on Instagram. Everything around the newest GLM 5.2 open-weight release was super loud on X and Reddit. And that gap is exactly the insight: if a topic is loud on three platforms and absent on Instagram, you can start launching it on Instagram before everyone else. The run also surfaced discussions on loops, agent swarms, and how much people are actually spending on AI, plus a set of post ideas I had not even asked for.
7. How to reuse the prompt (6:20)
The same prompt flexes in three directions. Run multiple platforms at once, as in the demo. Run the same platform across different niches. Or point it at creators: if you follow a list of creators you want to study every week, feed the list into this exact workflow and let the scouts do the reading.
Pitfalls and tips
- Say "dynamic workflow" explicitly in the prompt. Without those words, Claude Code will not set one up.
- Expect real wall-clock time. Setting up and running the workflow takes several minutes; I cut the video twice waiting for results.
- Verify before you publish. The per-agent raw outputs are there precisely so you can check what the synthesizer built its themes from.
- Treat the generated post ideas as raw material. The recurring themes are the real value; twist them into your own angles.
Where to go next
If workflows like this are new to you, start smaller with How to Build Your First Claude Skill. And to put an agent where your team already talks, read Claude Tag: Your New AI Employee in Slack.
