Reason, act, observe, repeat — written rather than taken from a framework, so every step is inspectable.
Autonomous Social Media Agent
A ReAct loop that researches, reads its own history so it does not repeat itself, writes a strategy memo, critiques its own drafts, publishes, and learns from every rejection.
The problem
What it was like before
Most 'AI content' tooling is a prompt with a schedule attached. It produces plausible posts, in volume, with no memory of what it said last week and no view on whether any of it worked.
We wanted to find out what a genuinely autonomous loop requires — not to save time writing posts, but because the same structure is what an operational agent needs.
How it looks
The loop
The two orange steps are the ones most implementations skip.
What we built
What it does
Research, history, drafting, critique, scheduling and publishing, each a discrete call.
It reads what it has already published before deciding what to say, which is what stops a content agent from repeating itself into irrelevance.
It writes down what it intends to do and why, before drafting anything. That memo is the thing a human can disagree with.
Drafts are reviewed against the strategy by the agent itself, and revised.
A rejected draft is not discarded. The reason feeds the next cycle.
Every call, argument and result is recorded. An agent you cannot watch is one you cannot debug.
Where it got to
In production
15Tools in the loop
ReActReason and act cycle
FullTool-call observability
Our view
Autonomy is mostly observability
The hard part of an autonomous loop is not getting it to act. It is being able to tell, afterwards, why it acted as it did — which tool it called, what came back, and what it concluded.
Everything we learned building this went into the agent work that pays: the strategy memo, the self-critique step and the rejection feedback are the same three ideas a compliance agent needs, with different stakes.
Built with
Tech stack
- Node.js
- Express
- Claude
- Playwright
- Prisma
- PostgreSQL
The work behind it
Delivered under AI Agent Development.
More work
Other things we have built.
Sanctions Screening Agent
Watchlist alerts arrive all day and most are false matches. The agent clears the obvious ones against a rule engine backed by an LLM and sends every disagreement to a person with the evidence already assembled.
Read itLaw-Enforcement Request Intake
Requests arrive as letters with scanned attachments. The agent reads the mailbox, OCRs the attachments, verifies identifiers against core banking, classifies the intent and routes it — with an audit trail behind every step.
Read itInternal Policy Assistant
Staff ask a policy question in plain language and get an answer drawn strictly from approved documents, with the source named. Nothing leaves the bank, and the assistant says so when the documents do not answer.
Read itCTR to goAML Conversion
A compliance team spent hours a week turning Currency Transaction Reports into goAML XML by hand. Parser, XML builder, schema validator and audit trail, behind a interface officers actually use.
Read itRegulatory Reporting Engine
A registry of every periodic return a bank owes. The engine pulls the data, builds each report against the mandated template, then schedules and submits it behind a maker-checker step.
Read itMulti-Agent Marketing System
Two agents on one platform with a person approving everything before it ships. One researches and takes a position; the other plans a week of posts reviewed as a batch.
Read itWhatsApp Booking Agent
Patients book on the number a clinic already advertises, at any hour, in the language they normally write. Only genuinely free slots are offered, and anything clinical goes to a person immediately.
Read itHave something like this?
Describe the process and we will come back with whether it is worth automating, roughly what it would take, and what we would build first — or tell you plainly if it is not a job for us.