Teardown No. 90 · Marketplace

1,019 Views on One Short: How Christian Anderson runs agent-drafted social media for his one-person Etsy shop, with a Discord veto

Christian Anderson has agents on his homelab run the social side of two small brands: THEC1LAB, his Etsy print shop, and IAMC/TECH, his tech page. Scripts compute facts from a fact bank. Models write captions and jokes, and validators check that numbers and names come only from cited facts. He approves or vetoes posts in Discord, and on TikTok he adds a sound to each agent-made draft and posts it himself. His first YouTube Short got 1,019 views and 15 likes, but as of 10 Oct his 138 active Etsy listings had 205 views ever and 1 favourite.

Christian Anderson · iamc_techVerified Nov 5, 2026Markdown copy

The problem

He writes, “I run two tiny things on my own.” He saw that his TikTok videos do 200 to 740 views each. On 10 Oct, his 138 active Etsy listings had 205 views ever and 1 favourite. In his words, “The problem is the bridge”: a personal TikTok gets no clickable bio link until 1,000 followers. He could not easily switch to a TikTok business account, because that “wanted Companies House details and I’m a sole trader.”

The blueprint (as he describes it)

  1. Keep the brands apart. The rule is “no Etsy CTAs on @iamc_tech, no homelab tips on the shop pages.” Each brand has its own intro, colours and sign-off. He says the rule “still bit me twice.” A planets post on the shop account got tech hashtags. The tech page’s “what’s inside” reels also tried to end on a THEC1LAB poster card, and he had to strip it.
  2. TikTok (@thec1lab): agent drafts, human posts. Posting through the API from an unaudited app gets forced to SELF_ONLY, so only he can see it. He ruled out cookie automation because of shadowban risk. Instead, the agent renders a daily video from the fact bank, sends it as an inbox draft and drops it in Discord at 17:30. He adds a sound and posts it himself.
  3. Instagram (@thec1lab and @iamc_tech): automated with a veto. Posting to his own account through the API is allowed, so this is fully automated. Every post is previewed in Discord about an hour before it goes out, and he can ❌ it. The shop account gets reels, carousels and stories. The tech account gets “Server” reels: jokes, tips and explainers. The agent re-weights the mix every week from insights.
  4. YouTube Shorts: uploader after approval. The uploader takes the day’s approved video at 19:45. It fixes the loudness, uploads the video and checks that it really came out public. Nothing needs a human after approval.
  5. Pinterest: weekly CSV, uploaded by hand. He says “Pins are links.” Every week a script makes fresh pins, 2 a day, one per listing, with words only from the listing itself. It writes a bulk CSV, and he uploads it by hand. Pinterest’s API trial access only makes pins visible to you, and he turned down using his own cookie to automate it.
  6. Threads (@iamc_tech): tech news plus his own posts. Each slot is one Hacker News story with why it matters and a clickable source link, under 500 characters. Since 8 Oct, one slot a day shares his own dev.to posts, “because I don’t self-promote enough.”
  7. Scripts gather facts. Models never do. He calls this “the rule that holds the whole thing together.” The model can write a caption or a joke, but “It does not get to invent a number.” Moon distances, binary puzzles and “which part is this” answers are computed by code from a fact bank. Tech facts carry a source and a verbatim quote. Validators then allow numbers and names only from the cited facts, block hype words, and block “closest” or “only” unless the fact text says so. Validator failures get one repair round. In testing, the validators caught a made-up 360,000 km and a stray London.
  8. Score captions and jokes. A small decision model scores captions. Genuine questions scored as supported at 0.81 to 0.95, and a made-up claim scored 0.05. “Comment YES and tag 3 friends” scored 0.97 for bait. For the tech page, a comic writer gives up to six fresh jokes a day. Filters remove anything with digits or a reused punchline, and a judge asks “would most people smile”. He added the judge because the first fresh jokes “passed every check and still weren’t funny.” Anything under 0.6 falls back to a fixed list. Since 10 Oct, the shop’s talking-Moon banter also gets a fresh judged joke every day.
  9. Keep the human veto. On top of the checks are his ✅ and ❌ in Discord: “The agents do the day-to-day running. I keep the veto.”
  10. What he ruled out.
    • Cookie automation for Etsy: he tried a headless browser with his own Etsy cookies to send buyer messages. The second page returned a 403 (“Access is temporarily restricted”). That was his home connection, so it could have hit his normal browsing too. He stopped, and nothing was sent. He gives the same answer for TikTok and Pinterest cookies: “not worth the ban.”
    • Follow-and-like automation on Instagram: he asked for it and the answer was no, because it breaks Meta’s terms and the API can’t do it anyway.
    • £1 drops: cheap limited drops on TikTok didn’t work, so they stopped.

Stated results

  • YouTube: first Short got 1,019 views and 15 likes, “better than any single TikTok I’ve done.”
  • TikTok: 200 to 740 views a video. The anatomy smartphone video did 716. Early on, a video with a commercial sound and a question caption got 389 views, and a silent shop drop got 0.
  • Etsy (as of 10 Oct): 138 active listings, 205 views ever, 1 favourite.
  • Instagram shop account: carousels failed 6 of 7 times since 7 Oct 16:00 with “Action is blocked.” Quota use was 1 out of 100 and reels went straight through, so he concludes it is an account-level spam flag, not volume. Reels and stories still publish daily.
  • Instagram tech account: started from 1 follower. No later count stated.
  • Pinterest: first shop pin scheduled for 11 Oct. He planned to check Etsy’s traffic sources around 22 Oct. No outcome stated.
  • Threads: “Too early to say.”
  • Sales or revenue: None stated.

Caveats

  • This is a self-published, first-person post on DEV Community. It is not vendor-hosted, and the post does not sell a product, course or kit. It does relate to his own businesses: the Etsy shop, and a tech page that his diagram says points at dev.to, Gumroad and GitHub.
  • The page footer says the post was “Drafted with AI assistance from my own homelab notes, logs and repos, then reviewed and edited before publishing.”
  • The numbers are tiny and early, and most are dated 10 Oct. The page shows “Posted on Oct 20” without a year. He calls the evidence for Pinterest driving Etsy printable traffic “anecdotal” and says he is not counting on it.
  • The Meta app is still in Development mode, so the comment agent has never seen a single comment on either Instagram account. Strangers’ comments are invisible until he switches it to Live.
  • Audio bug: the reel mixer was built for silent videos. It kept only the music, normalised it to -14 LUFS and silently dropped the talking-Moon voices. He only caught it because the music sounded too loud. The tech page now uses voice at -16 LUFS with music at -30. The shop account kept the old mix for now. TikTok banter originals still come out at about -36 LUFS, which he says is not fixed yet.
  • Silent gaps: the YouTube channel was set up on 8 Oct, but the daily uploader was never built until 10 Oct. In the comments he says, “I hadn’t put together anything that trips an alert when a slot goes empty.” A commenter whose profile says they build and operate custom AI systems for ops teams suggested a check about an hour after each slot that looks at the platform itself, not your own sent log.
  • Handles: X @iamc_tech (Christian Asare Anderson), GitHub github.com/casare, Instagram @thec1lab (shop) and @iamc_tech (tech), Threads @iamc_tech, TikTok @thec1lab, DEV @iam-tech (all verified 2026-11-05). Bluesky, LinkedIn, Mastodon: none verified.

FAQ

Does Christian Anderson approve every post before it goes out?

On TikTok he adds a sound and posts each draft himself. Instagram and Threads posts can be cancelled with ❌ in Discord, and Instagram posts are previewed about an hour before they go out. YouTube takes the day’s approved video, and nothing needs a human after approval.

Has the setup produced Etsy sales?

The source does not say. As of 10 Oct, his 138 active Etsy listings had 205 views ever and 1 favourite, and he planned to check around 22 Oct whether Pinterest pins send any traffic.

How does he stop the models from inventing facts?

Code computes the facts from a fact bank, and tech facts carry a source and a verbatim quote. Validators allow numbers and names only from the cited facts, and in testing they caught a made-up 360,000 km.

Why doesn’t he fully automate TikTok?

Posts from an unaudited app through the API are forced to SELF_ONLY, so only he can see them. He also ruled out cookie automation because of shadowban risk.

Sources

Want agents like these working your distribution? Get It Built. Get new teardowns weekly: subscribe (double opt-in).

Get one teardown a week.

Same format, a new operator every issue. Double opt-in.