Barnicle Productions
Learning Log
Working notes from building governed AI systems for real businesses — the audits we run on our own tools, the methods we borrow and credit, and the forks in the road, including the paths we didn't take.
Explore five complete AI computer setups, from a practical business desk to a shared company knowledge hub. Compare local model limits, widescreen monitors and real operating costs, then download each tier's product list and buying sheet.
Follow water-restoration research through RepoOps, a Grow Wild content experiment and a preserved allies-list import. Open the workspace map and see how sources become useful project questions.
Open the master vault, explore three layers of folders, document roles and system modules, and follow one idea from saved source to useful work. A plain-language map of what each part does and why it belongs.
The plain-English guide we ship inside every system we build: the two brains and the crew, how a thought becomes trusted knowledge, the three backup copies, the key, and which rules are yours to bend.
An evidence-led comparison of operator memory, governed business knowledge, agentic workflows, frontier and local models, costs, safety rails, and the hybrid architecture that lets them work together.
Capture broadly, keep knowledge separated, and control how work becomes action. The operating system behind my businesses, client work, intake system, governance rules, and what I am still building next.
The three jobs AI is genuinely good at, the ones to skip for now, and the order of operations everyone gets backwards. No jargon anywhere.
It isn't lying — it was never told the truth about your business in a form it can read. Why longer prompts don't fix it, and what does.
Our robots.txt said Allow: / for everyone. The file Cloudflare actually served blocked GPTBot, ClaudeBot and six more — for three weeks, on the site of a company that builds AI systems.
We ran our personal AI vault through a published audit spec and wrote up every finding. The real lesson: know which kind of system a spec was written for — and prove your fixes didn't cost capability.
What Interpretable Context Methodology actually is: the 5-layer model, workspaces, pipelines, and why folders plus plain text beat agent frameworks for most real work.
Same DNA, different jobs: a builder's workspace compiler versus a governed knowledge layer for businesses with confidential data. What we integrated, and why we didn't rebuild.
Most AI pipelines are black boxes: files in, answers out, no way to stop it. The five-stage, human-gated process we run client material through instead.
Next step
Start with an audit. We map your workflows, assign privacy lanes, and tell you honestly what not to automate.
Book a call
Book a free audit callPick a time on Monday, Wednesday, or Friday morning — you will get a calendar invite with a video link. “Just curious” is a fine reason to book.
Prefer email? info@barnicleproductions.com