Home / AI Lab
A working look at the n8n workflows and AI agents behind the marketing systems — what problem each one solves, what it's built with, and what changes once it's running.
Lab 01 — Content Operations
Reads approved posts from a content sheet on a schedule, filters to what's ready and due, then publishes the same post to LinkedIn, Instagram, Facebook and X (Twitter) at once — and marks it posted so nothing goes out twice.
What it's useful for: teams that plan content in a spreadsheet but currently copy-paste it into four different apps by hand.
Lab 02 — CRM & Outreach
Pulls contacts from HubSpot on a schedule, an AI agent (Google Gemini) decides what each contact needs next and drafts a message, sends it through Gmail, then updates or creates the HubSpot record — so contact follow-up doesn't depend on someone remembering to do it.
What it's useful for: keeping a CRM warm without a person manually reviewing every contact.
Lab 03 — Customer Support
Answers customer messages on WhatsApp using an AI agent with memory and three tools — live inventory, order status and an FAQ sheet — so it can answer "is this in stock" or "where's my order" correctly instead of guessing.
What it's useful for: e-commerce or service businesses fielding the same handful of questions on WhatsApp all day.
Lab 04 — Content Studio
A three-agent pipeline — AI Agent, AI Copywriter and AI Art Director — drafts a post and image together, an AI Quality Control step reviews it, then it's sent to Telegram for a real human to approve or reject before it ever goes live. Rejected drafts get automatically regenerated; approved ones publish to Facebook on schedule.
What it's useful for: teams that want AI-drafted content but aren't comfortable letting it publish with zero human review.
Lab 05 — Operations
The largest system in the lab: task intake from eight channels at once — Slack, Teams, Email, CRM, ERP, Web Form, WhatsApp and Facebook — normalized, deduplicated, checked against a knowledge base, then triaged by an AI risk-assessment step with confidence scoring. High-confidence tasks auto-execute; everything else routes to a human approval queue, with full audit logging, SLA monitoring, error handling and automatic daily / weekly / monthly reporting.
What it's useful for: operations teams drowning in requests coming from too many different channels to triage manually.
Workflow JSON exports and build walkthroughs for select projects are shared on request, or via GitHub for generic, non-client-specific builds. Client-specific automations stay private by default.
If it happens the same way more than twice a week, it's probably automatable.