I kept losing my thinking. Not my notes, not my files. The reasoning. The connections between ideas that only existed inside a conversation, and then the conversation would end and all of it was gone.

I was using ChatGPT for complex structural work: worldbuilding for a novel, a side project called ScopeForge, career planning, brand development. Months of interconnected thinking where one decision fed into the next. And every time the session closed, the next one started from zero. I could paste summaries, re-upload documents, try to reconstruct. But the thread that connected all the pieces was gone every time the context window closed. When I came back to something months later, I told the tool: I feel like I did so much ideation that got lost or buried.

That loss is where Savepoint Protocol came from. I started typing “give me a savepoint” into conversations as a survival mechanism, and the name came from saving video games as a kid. Same instinct: you’re about to lose your progress, so you save your state before the session ends. Just dump everything into a copyable block before the context closes. It showed up in cooking conversations, fitness conversations, gaming, web dev. It was a reflex before it was a protocol.

1,643 conversations

I want to be clear about what ChatGPT was to me, because this isn’t a story about a bad tool. Between January 2023 and early 2026, I had 1,643 conversations with it. I used it for everything: cooking technique, career planning, teaching approaches for my kids, enterprise UI strategy, brand development. The first few months were simple lookups, and then in September 2024 something cracked open and I went from a few conversations a month to dozens a week. I was trying to figure out what I actually was, professionally, after twenty years of building things across so many domains that no single title fit. ChatGPT became the place I worked that out.

The tool helped me think. It just couldn’t hold what I was thinking. ScopeForge was the project that made it undeniable. I’d been trying to build a tool for mapping project context and dependencies, and every time I sat down to implement the next piece I had to re-explain the repo structure, the decisions I’d already made, what connected to what. Three messages later it would contradict the architecture we’d just agreed on. I said to it at one point: the issue I’m having is, ChatGPT is failing to help me in engineering tasks. It’s just not capable enough.

A friend mentioned Claude.

What changed

Claude Code is a command-line tool. It runs in your terminal, inside your project directory, with access to your files. I subscribed on August 25 and started using it immediately, though at first I kept it quarantined to GitHub Codespaces because I wasn’t ready to let it run locally. But even in that sandbox, the difference was obvious. It could read my repository, the directory structure, the config files, the stylesheets, the data, the content. I didn’t need to paste context because the context was the project itself.

And then there’s this file called CLAUDE.md. It sits in the project root, and it contains everything a new session needs to know: what the project is, how it’s structured, what decisions have been made, what conventions to follow. Every session reads it on startup. So decisions accumulate there instead of evaporating when the conversation ends. That’s the thing I’d been building workarounds for since I first started losing my thinking across sessions. A file in a repository, and the tool reads it, and the continuity problem that had been killing my momentum for two years was just solved.

Decisions accumulate there instead of evaporating when the conversation ends.

What I built with it

In February 2026 I started building this site in Claude Code sessions. Not as a portfolio to showcase my history — as a laboratory to develop and test the methodology. The layouts, the design system, the navigation, the SEO infrastructure, all of it happening inside the repo where the tool could see everything and I didn’t have to re-explain anything between sessions. The character of the work was completely different from the ChatGPT era. Those conversations were exploratory, me trying to figure out who I am professionally and working through naming and positioning and existential questions about my career. The Claude Code sessions were building. I needed a real project to develop the governance system against, and the site became that project.

I think both tools were the right tool for the phase I was in. ChatGPT was the right tool for working through the questions. Claude Code was the right tool for building the answers. I didn’t switch because one is better than the other in some abstract way. I switched because the work changed, and I needed a tool that could work inside my file system and remember what we’d decided.

Why it matters to me

The goal was never to become an AI person. I’m a systems architect who has been building things for twenty-five years. I tried a tool, used it heavily, found its limits, tried another tool, and it stuck. The reason it stuck is specific: it works where my projects already live, and it reads a file that accumulates decisions instead of forgetting them.

The real goal is to get good enough at this to bring it to my team at work so we can all level up. That’s what this is about. I needed a better tool and I found one.

What I built with that tool over the following months turned into something I couldn’t find anywhere else: a governance system for AI-assisted creative work. Evaluative lenses extracted from practitioners selected for specific diagnostic needs. Voice governance derived from how I actually talk, not how I write for publication. A coordinator architecture that runs these together and surfaces where they agree and where they disagree. The rest of these posts explain how each piece works.