Trashformas is building a biogas supply chain in Nigeria — from household waste collection to distribution as a clean cooking fuel — and most of it still runs manually today. Your team is four people, and the context lives in your head: meetings repeat themselves, and even the outside marketing firm you hired routes every question back to you. What you want is one place that holds what the team already knows and hands it back at the moment someone needs it.
You hand a task to your outsourced marketing team, with a written guideline and battle cards already in their hands. They open a WhatsApp group — and still direct every question straight to you, so the intermediary you brought on to save your time now spends it. The same pattern shows up inside the team: you explain a decision in one meeting, and a week later you explain it again, because there’s nowhere it stayed.
A place that already holds the business context — so the marketing partner asks the system instead of you, a decision made once stays made, and you can step out of conversations that don’t need you while the team still moves forward.
When the context only lives in your head, every question has one address — yours.
We’d start with capture — getting what the team already knows into one structured place — because the guides and guardrails you described only work if there’s something reliable underneath them. You’ve chosen n8n and you’re already recording with Fathom; the missing piece is the layer those recordings feed.
Start with the meetings, because that’s where your context is born and lost. You run internal and virtual calls through Fathom today, but the output scatters into Gemini, Claude, and ChatGPT and never lands anywhere shared. Pulling those transcripts into one structured store is the first thing that stops you repeating yourself.
Keep the internal layer and the app data on separate tracks at first. Meeting context and the household data from the collection app are two different kinds of information — audio and decisions on one side, images and waste-volume measurements on the other. They’ll meet eventually, but forcing them together on day one is how a store fills with noise.
Decide what “valuable” means before the guides read anything. You said the Passive Project Guides should surface only what matters — not every piece of jargon that enters the vault. That filter is a design choice to make with you, not a default to inherit.
A hypothesis to co-shape, not a fixed plan. The first artifact would likely be a decision log — a single structured record of what the team decided, why, and what it rules out.
Surfaced to the asker directly. You’re not pulled in.
Routes to you or the COO for a call — and that answer goes back into the log, so it’s only ever answered once.
You named the trap most people walk into: it isn’t automation itself, it’s automating the wrong task and then hiring someone to babysit the machine — automate 80% of the work and still do 80% of it by hand. That’s the exact failure you’re designing against, and it’s why you keep insisting the system guide the team rather than replace it. Guardrails, not autonomy — that instinct is the right one to build on.
Two steps, co-owned.