Internal documentation
LiveTurning Recurring Tasks Into Documented, Delegable Processes
Using AI to capture SEAR's operational knowledge as written processes — so a two-person company stops depending on either person's memory.
Last updated · Tools: Claude (Cowork), Markdown / shared docs
The problem
A two-person company holds most of its processes in two heads. That works until one person is on a plane, mid-surf-trip, or simply busy — and then a VAT filing step, a carrier quirk or an Amazon case procedure has to be reconstructed from memory. Our Amazon escalation saga made this vivid: months of case history that lived nowhere except an inbox.
The previous manual process
There wasn't one, which is the point. Documentation was the task that always lost to every other task. Knowledge transfer between me and Andrea happened by conversation, repeated as needed, forever.
The AI-assisted workflow
The insight that made documentation finally happen: you don't have to write the document — you have to explain the process once.
When a recurring task comes up, I talk (or type) through how it's done, messy and unstructured. The assistant turns that into a clean process document: steps, exceptions, who does what, what to check before finishing. It asks clarifying questions where my explanation had gaps — which is exactly where the documentation would have failed later.
Documents live in a shared folder, and get corrected whenever reality disagrees with them.
Human review required
Each document gets a real-world test: the next time the task happens, we follow the written version literally and fix what's wrong. A process document that hasn't been executed once is a hypothesis.
Outcome
The recurring tasks that used to depend on memory — shipment prep checks, escalation procedures, publishing conventions — now exist in writing. The pre-shipment barcode verification checklist alone justifies the whole system, given what the lack of it cost us.
What failed or remained difficult
Keeping documents updated is still discipline, not automation. AI removed the blank-page cost of creating documentation; it can't notice on its own that a carrier changed their process. Stale documentation is more dangerous than none, because it's confident.
What I would change next time
Start each document with "when do you use this" rather than the steps. Early documents explained how perfectly and when not at all.
Reusable lesson
The bottleneck for documentation was never writing skill — it was activation energy. AI collapses that cost to near zero, which means the real question becomes cultural: do you actually stop and explain the process once. We had to make it a habit.