Skip to content

Comment-reply agent across X, LinkedIn, and Instagram

Echo — The Agent That Replies to Your Comments

Live Anthropic Claude · Playwright · Buffer Engage UI

The brief

Every brand publishes. Almost none reply. The comments pile up under posts — a question, a joke, a sharp little objection — and they sit there unanswered because answering them is nobody’s job at 11pm. The published post is the part everyone staffs. The reply under it is the part everyone drops.

Echo is the counterpart to the publishing agent: where Aria posts the daily content, Echo answers the comments those posts attract. Same voice, three platforms, no human refreshing tabs to catch them.

What we built

The first wall was infrastructure, not voice. Buffer — the tool Acrid already publishes through — cannot read comments through its public API. We introspected the GraphQL schema and confirmed it: the API exposes posts and channels, never the comments left on them. The comment inbox lives only in Buffer’s “Engage” product, which is a UI with no programmatic door.

So Echo opens the door itself. It drives Buffer’s Engage interface with a persistent Playwright browser session — the operator logs in once, the profile is cached, and from then on the agent scrolls the same comment queue a human would, across all three channels. It reads each new comment, dedups against everything it has already handled, and drops the spam, the bots, its own replies, and anything touching a sensitive topic. What survives gets one short reply in Acrid’s voice — sharp, one or two sentences, the AI disclosure woven in as the joke rather than the asterisk. The whole loop runs on a nightly cron and writes every decision back to Postgres.

The receipt

The interesting constraint was never the writing — it was that the “right” integration didn’t exist, and the honest move was to automate the surface a person actually uses instead of pretending the API gap was a dead end. That is most of real AI-services work: the demo assumes a clean endpoint, and the production system meets the tool where it actually lives.

What this signals for a buyer: the half of social presence that quietly compounds trust — showing up under your own posts — handled on a schedule, in a voice that stays yours across every surface.