How I Automate 90% of My Social Media Content with Claude Code
A solo creator talks Claude Code through building a seven-agent content team — then spends five hours making it actually match her brand.
March 14thSandy Lee interviews Lazaros, who took a podcast from zero to 6.18 million views in five weeks by routing research, scripting, editing, emotion-ranked short-form repurposing, and scheduling almost entirely through Claude Code.
One operator ran research, scripting, editing, and scheduling for an entire podcast through Claude Code alone — the leverage came from consistent folder structure and reusable style rules, not from any single clever prompt.
Lazaros grew a podcast from zero to 6.18 million views in five weeks by routing almost the entire production pipeline through Claude Code. Before recording, he has Claude scrape a guest's past interviews into a searchable index to find an angle nobody else has used, then brainstorms by asking for 100 possible angles instead of one. After recording, the same transcript-plus-timestamps method drives both the intro edit and short-form repurposing: Claude finds compelling moments, classifies each by the emotion it triggers — weighted toward fear and FOMO — and ranks them for a four-times-daily posting schedule. A persistent per-project CLAUDE.md file and saved style 'skills' let every session reuse the same captions, colors, and music-matching rules without re-explaining them.
Sign in and you get 23 free chat messages on us — ask for the hook, quote a framework, find the exact transcript moment, generate a markdown action plan. Bring your own key when you want unlimited.
Create a free account →
A compiled highlight-reel trailer (channel grown to 2.7k subscribers and 6.18M views in five weeks) cuts into Sandy introducing Lazaros and framing the episode: how he uses Claude Code to run 99% of his content pipeline, from research to scheduling.

Lazaros tells the origin story: he recognized ex-CIA officer John Kiriakou at a tavern in Cyprus, worked up the nerve to ask him onto a podcast that didn't exist yet, then texted him once a day (bumped at day's end) until the logistics were locked and the episode got recorded.

After the Kiriakou episode did well, Lazaros posted the win inside his Maker School community. Nick Saraev commented under it, so Lazaros used one of Saraev's own outreach techniques — a Loom video — to pitch him, and flew to Canada five days later to record.

Sandy calls Lazaros's confident pitching 'manifesting.' Lazaros reframes it through Think and Grow Rich: writing down who you want to meet doesn't summon them, it sharpens the everyday decisions that move you toward that goal.

Lazaros explains why in-person outreach beats a cold online ask when the goal is a real connection, then admits his first-ever on-camera interview (studio lights, first time holding a mic) rattled his voice — he had three days' notice and no time to overthink it, comparing it to deciding to skydive and jumping the next week.

Lazaros splits any content project into three phases and says preproduction research has never been easier than with Claude Code — the John Kiriakou episode is the example that follows.

For the Kiriakou episode, Lazaros had Claude scrape every past interview transcript into an index file, then queried it to find an angle nobody had used yet — Kiriakou's Cypriot roots and the island's geopolitical split — and to flag overused questions. A mid-roll HubSpot-sponsored plug for a free 'Complete Guide to Claude AI' prompt document runs in the middle of this segment.

Lazaros never writes a word-for-word script — he builds a loose framework (intro beats, a couple of bullet points per segment) and stays reactive to whatever the guest says, threading their last sentence into his next prepared question in real time.

Editing the long-form podcast itself is minimal — a Claude Code skill or repo like video-use or Remotion can auto-switch camera angles by speaker. The real leverage is the intro: it must deliver on the exact promise made in the thumbnail the instant the video starts, so Lazaros a/b-tests multiple intro angles that all still pay off.

Same framework, reused: give Claude the transcript with speaker labels and timestamps, then have it find clips that induce emotion, classify each by the feeling it produces, and rank them best to worst. Lazaros weighted negative emotions like fear more heavily because they land harder than positive ones.

Lazaros screen-shares his real project folder (organized by client and by podcast, each with a CLAUDE.md context file) and drives Claude live: pick the transcript, brainstorm emotional hooks (FOMO, fear, motivation), draft the clip script, then generate a rough vertical cut with video-use.

Asked how to get over the fear of cold outreach, Lazaros says he reframed rejection as a game — like not knowing what card you'll draw in blackjack — and now aims for 100 rejections, confident a win arrives before he gets there. His rule: it's not about avoiding the no, it's about how far you get before it.

Lazaros walks through his GitHub-backed folder structure (clients, company projects, podcasts) and explains why fear was the dominant emotion tag on his John Kiriakou shorts — he'd scraped Kiriakou's past interview titles and found almost all of them played on fear, tracking his belief that negative emotion registers more strongly than positive.

Lazaros collected songs from reels he liked, had Claude run them through an open-source Shazam-style model to identify and download the exact track, then had Claude tag each song's emotion by transcribing the source reels that used it — so a fear-tagged clip automatically pulls fear-tagged music from a 10-song library.

Every candidate clip gets made and ranked into tiers; Lazaros posts one tier-one clip and backfills with tier-two/three across four daily slots, since YouTube doesn't penalize high posting frequency. He also compares DM-automation tools — dropping ManyChat, which nearly billed him $6,000 in per-contact fees after one viral post, for the free-for-two-accounts Xerneo.

Lazaros describes running the pipeline remotely — drafting a thumbnail from the gym via the Claude mobile app's remote-control feature, keeping his laptop awake with a caffeine utility, and sending Claude a prompt whenever he has five free minutes. He closes by watching captions and music finish rendering live before thanking Sandy.
The leverage in an AI-run content pipeline comes from consistent project structure, reusable style rules, and transcript-driven prompting, not from any single clever prompt.
“I refuse to use anything else.”
“I just texted him twice a day until he showed up.”
“My usual prompt is give me a 100 different angles that I can go for. I read all of them. 99% of them are completely horrible, and then you find one that kind of works.”
“The most challenging part and the most important part, at least in my opinion, is the intro because it decides after people click.”
“What is gonna make me feel the most amount of difference from what I'm feeling right now, which is a brain rot?”
“All of the shorts that you mentioned were all made in a single day and then scheduled for two weeks.”
“It's not necessarily getting the no. It's how far can you go before you get the no.”
“Unfortunately, humans experience negative emotion at a much larger scale than positive ones.”
“If you see that piece of music being used across multiple reels that their transcript conveys fear, it then probably means that song is about fear.”
“If I were to use ManyChat for this specific thing, I would have paid $6,000.”
“If you have the vision and you are able to communicate it with Claude, you'll basically be able to turn it to life with just words. It's like speaking to an editor.”
See every word as it's spoken — crank it to 2× and still catch all of it. The same dual-channel trick behind Amazon's Kindle + Audible.
The video opens on a cut-together trailer of the episode's best lines before the real conversation starts: a channel at 2.7k subscribers with 6.18 million views in five weeks, an ex-CIA officer landed by cold-texting a stranger twice a day, and sixty reels made in a single sitting and scheduled out for two weeks. The thread running under all of it is the same claim repeated at the top: this entire operation — research, scripting, editing, emotion-ranked shorts, scheduling — runs through Claude Code, with no editor and no editing software.
00:00
00:50
02:00
02:58
03:36
04:27
05:19
05:57
06:39
08:17
08:29
09:14
10:10
11:04
11:43
12:39
13:23
14:27
15:06
16:02
17:02
17:35
18:42
19:12
20:26
21:09
21:40
23:09
23:22
24:35
25:03
25:59
26:39
27:51
28:18
29:18
30:21
31:11
31:44
32:37
33:16
34:27
35:10
35:49
36:39
37:25
38:24
39:30
40:26
41:09
41:38
42:35
43:15
44:11
45:04
45:50
46:59
47:30
48:23
49:14
50:09
51:24
51:49
52:31
53:32
54:05
55:18
56:30
57:00
57:23
58:37
59:10
60:15
60:47
61:44
62:34
63:25
64:17
65:05
66:08A solo creator talks Claude Code through building a seven-agent content team — then spends five hours making it actually match her brand.
March 14thA creator hands her raw footage to Claude Code and walks through exactly what it costs, and how to prompt it, to get a fully edited long-form video back.
July 20thA 12-minute live build: one Claude Code skill turns any YouTube video into three platform-ready shorts with AI avatar, B-roll, and auto-scheduling.
June 17thA live, unscripted screen recording where an SEO YouTuber wires Claude Code into a real Shopify store and SEMrush's keyword tool, then watches it draft, publish, and index new pages in real time.
July 16thA systems playbook for Claude Code and Codex: the five-step loop, the instruction files, and the guardrails that separate trustworthy AI output from expensive rework.
July 16thA 17-minute screen-recorded walkthrough of installing, configuring, and running the mattpocock/skills repo on a real codebase — from a vague idea to a reviewed, committed change.
July 16th