AI-Run Business: The Six-Week Toolkit
Six weeks of building InAgentic as an AI-run business, distilled down to the tools that actually matter. Eight systems, one shared platform, and a lot of lessons about where automation belongs and where a human still needs to click confirm.
Six weeks of building InAgentic as an AI-run business, distilled down to the tools that actually matter. Eight systems, one shared platform, and a lot of lessons about where automation belongs and where a human still needs to click confirm.
The eight tools
Lead scoring (Agent 1) — reads every inbound message overnight, scores it by intent and urgency, and enriches it with company context like funding rounds and headcount changes, so the hottest lead is already at the top when you wake up.
Content pipeline (Agent 2) — takes a topic from idea generation through drafting, editing, and publishing to Ghost and FileDone in a single automated run, no copy-pasting between tools.
Enrolment agent (Agent 3) — the moment someone joins as a member, this grants Ghost access, creates an MCP API key, and drafts the welcome email, all within 30 seconds and with no human in the loop for the routine cases.
Video pipeline — started as a HeyGen-and-YouTube agent, then got rebuilt on LangGraph as a 10-node graph with four human approval gates (idea, script, images, final video), replacing a fragile n8n workflow that treated approval as an afterthought.
Course chatbot (+ RAG) — a freemium assistant on every workshop page: free static FAQ for anonymous visitors, Claude on Bedrock for signed-in members. Later gained real search over InAgentic's own published articles via pgvector, so it can answer from what the business has actually written instead of guessing.
Booking assistant — a branding-free chat page that takes a plain-English request like "1pm Saturday," checks real Google Calendar availability, and gets the owner to confirm with one click. Built once, reused as-is for a second business (dogrun.ai) with zero new infrastructure.
Curriculum tool (DocBuilder) — turns whiteboard photos and handwritten notes into structured markdown, with an email agent that creates a document just from a photo sent to an inbox, no browser required.
Ops dashboard — the control layer tying all of it together: nine (now growing) agent tabs, centrally configured, later extended to be multi-tenant so FileDone, dogrun.ai, and Pancake Pandas each get their own isolated view of the same shared backend.
This is the last day of the six-week sprint. From here, we'll keep building, but posting will slow down. Keep watching this space for what's coming:
WhatsApp: The Remote Control for Your Business — a two-way WhatsApp channel for the agent stack, so reminders, confirmations, and support replies can happen inside a chat thread instead of an inbox or a dashboard. We already have WhatsApp wired up through Twilio and will keep developing it from here.
Cybersecurity and AI agents — a look at what it actually takes to secure an agentic stack that touches real customer data and real credentials, including getting testing certificates in place rather than just assuming the defaults are safe.
PewDiePie's new free Odysseus AI agent — a look at this newly released free agent and what it signals about where consumer-facing agent tooling is heading. We've already installed it ourselves and are looking at what it means to own your context data and choose to send it to different LLMs, rather than being locked into one provider's context.
Avatars for everything: sharpening the content output — more avatar-led video across presentations and other content, with quality improving gradually as we build out a wider library of templates rather than treating each video as a one-off.