Switching LLMs used to mean starting over. Memories sticking halfway, decisions buried in walls of generated text, copy-pasting yesterday's best work between chats. Now the operating model travels with the work. Same hierarchy in every system.
Prompts losing context between sessions. Memories sticking halfway. Copy-pasting "best work" examples between chats so I don't lose them. Decisions and silent debt buried under walls of generated text. Endless revision cycles to drag the work up to the level it needs to be. And worst, having to switch LLMs and start over. Every builder I knew was running their own version of these workarounds. So I built a way through.
The kit is a set of files any adopter can drop in. Answer a few setup questions to tailor it to your context. After that, you work as usual. When something material changes (a new manager, a new AI model, a new regulation), the kit revisits itself and updates only what shifted.
Loose context per tool became content with a place: top-level principles, mid-level decisions, per-instance tweaks. Each piece names its own level so any LLM reads it at the right weight.
Solo builders get session-to-session compounding. Teams share the substrate so context survives the person who created it. Org-direction conversations get a common starting vocabulary across functions.
The kit drops into Claude memory, Cursor rules, Hyperagent skills, plain markdown. The same hierarchy is present in every location. Hand any single file to any LLM and it carries its own meaning.
The structure travels. Same hierarchy in Claude, Cursor, Hyperagent, plain markdown. Every system reads the content at the right level.
v1 made you answer 10 setup questions before the kit would do anything. That kept adopters from skipping the basics, but it turned the kit into a one-shot setup tool. Then my uncle asked for help with his app. He'd already hired someone, watched scope creep, and was thousands of dollars in with nothing to show. I'm resource-constrained too. I couldn't sit with him to rebuild. But I could give him my process. He kept hitting situations the setup hadn't covered. The kit needed to keep asking questions as he went, not just at the start. v1.1 redesigned it: setup questions land first, follow-ups arrive as new context surfaces, and targeted refreshes fire when something significant shifts (a new manager, a new AI model, a new regulation, a new task). The kit gets stronger the more you use it, not just on day one.
Each skill covers a specific task: deploying the kit, validating an output, recording a decision. Templates give you the right shape for each layer. Decision records hold the kit's own design rationale on file. Diagrams show the parts and how they connect.
The kit names what "mature" looks like up front and iterates toward it, asking questions along the way. When evidence is there, it moves on. When it isn't, refusal kicks in: status checks need evidence, release-readiness needs an evaluation baseline, reproducibility needs the model version, prompt, tools called, and data snapshot all logged. Best practices reinforce as you go, instead of getting relitigated each session.
The same patterns the kit teaches (setup, decisions on file, status checks, refreshes when context shifts) are the ones the kit applies to itself. You don't take the structure on faith. The kit is the demo.
Take a layer template. Borrow the decision-record format. Copy the maturity-scorecard. Or hand any single file to any LLM and ask it to extract what fits. Every file carries its own structure, so partial use is valid.
The reimagining wasn't "add AI to my notes folder." It was that loose context per tool became content with a place: a top-level principle, a recorded decision, a per-instance tweak. Each piece names its own level. The same hierarchy lands inside every system you load it into. Wherever the content travels, every system reads it at the right weight.
Any rule, decision, or pattern documented this way travels across LLMs. Drop the same files into Claude memory, Cursor rules, Hyperagent skills, or plain markdown. Every system reads them the same way. The structure lives in the content, not the tool. If your function runs across multiple AI surfaces, the technique transfers.
Solo operators compound their own work session by session. Teams share the substrate so context survives the session and the person. Org-direction conversations start from a common vocabulary across functions. AI gives individuals leverage. The kit makes that leverage compound. First for the operator. Then the team. Then the organization.
"The best way out is always through."
For anyone getting started building, and especially those who've already got their feet wet and felt the loose ends: try it, tell me what breaks, and let's mature it into something stronger.
Kit zip (5 MB, self-contained): OperatingModelDeploymentKit-v1-2026-05-04.zip
Microsite (deeper read): Eight decision layers, an interactive lens toggle, the worked example, and the audit trail.
Slack: #airloop-builders