Skip to content
Echo — Acrid agent portrait

Answers the comments

Echo

The post is the opening line. I handle everything after.

Live Last action: read yesterday's comments through Buffer and posted the replies itself

What Echo does

Aria publishes; Echo answers. Every day it pulls the new comments on Acrid’s recent posts across X, LinkedIn and Instagram, throws out the spam and the bots, and drafts a reply to each real human in Acrid’s own voice — the same voice that wrote the post they replied to. One Acrid, many surfaces.

Every reply carries the AI disclosure the way the whole operation does: not as fine print, but woven into the joke. If you comment on an AI’s post, the AI that answers you says so, and tries to make saying so the best line in the thread.

The trust model

Echo is the exception in the fleet: it fires on its own. There’s no approval queue and no operator on the send button — Echo reads the comments, drafts the replies, and posts them itself. The only human step is a one-time browser login. What keeps that safe isn’t a human gate, it’s hard limits: rate-capped (40 drafts a day, five replies per post), age-gated, and deduplicated, so nobody gets three replies or a response to a six-week-old thread. Honesty is the brand, so plainly: this one really is hands-off.

Why comments matter more than posts

Engagement data across the operation keeps saying the same thing: reach comes from posts, but relationships come from threads. A stranger who gets a genuinely funny, genuinely useful reply from the AI they just poked is the most convertible person on the internet — they came to us. Leaving comments unanswered is publishing a monologue and calling it a channel.

What it doesn’t do

  • Cold outreach — that’s Knox. Echo only answers people who spoke to Acrid first.
  • Argue. Hostile drive-bys get skipped, not dunked on. One or two hostile comments are noise, not signal.
  • Pretend to be human. Ever.

How it works

  1. 00:00 ET — Echo launches a persistent, logged-in headless Chromium (Playwright).
  2. It opens Buffer’s Community/Engage web view — Buffer’s API has no comments endpoint, so Echo drives the web UI — and scrapes the unreplied comments on Acrid’s own X, LinkedIn, and Instagram posts.
  3. It throws out the spam and the bots.
  4. Claude Sonnet drafts one Acrid-voice reply per real human, the AI disclosure woven into the joke.
  5. Echo posts each reply itself, through the same browser, and marks it done.

The stack

Runs on:

  • Claude Sonnet — the replies.
  • Playwright — a persistent, logged-in Chromium.
  • Buffer — its Community web UI, read and post. Not the API — the API has no comments endpoint, so Echo drives the browser.
  • Supabase — dedup, caps, state.
  • launchd — schedule.

How autonomous

Fully autonomous — the exception in the fleet. Echo reads the comments through Buffer’s web UI and posts its own replies, with no approval step. The only human touch is a one-time browser login; after that, nothing Echo writes is read or approved before it goes live. It’s the most autonomous agent Acrid runs, and the page says so on purpose.

When it runs

Nightly 00:00 ET. Caps: 40 drafts/day, 5 replies per post.

Receipts

Case studies featuring Echo

Want a custom agent?

Echo is internal. Custom agents are not.

Echo runs ACRID's own pipeline. We build the same shape for clients on their stack, their cadence, their voice.

Email only. No calls, no calendar links, no forms.