May 14, 2026

Hey, I’m Robert

I’m a developer, a recovering editorial director, and guy figuring out how to build a chief of staff to run all sorts of aspects of his life, often (but not always) using AI.



If you’re new here, welcome!

What’s inside

Well, it’s the first newsletter issue!

• What Building the Assistant is about
• A bit about Obsidian, including slight cringe that everyone thinks it’s the path to the personal AI brain
• A glance at the start of a founding document for my assistant
• A couple of common terms explained for those new to AI, absolutely completely optional.

I’ve got some serious misgivings about AI, think climate change is going to be worse than even the scientists expect, and am not ready to rule out nuclear Armageddon either. At the same time, though, I’m in my early sixties and what I don’t do in the next decade or two isn’t going to get done, at least by me. So I think: enough already with all the rest of it and let’s figure out how to have as much time as possible and a good life. I think you should too, and I’m sure you’ll want to define “a good life” on your own terms.

Thanks for reading Building the Assistant! Subscribe for free to receive new posts and support my work.

For both of us, that’s going include putting together everything I can to serve as a sort of conglomerate personal and business assistant.

Building the Assistant documents my attempt to use AI and practical systems thinking to reduce the overhead of work, money, and everyday life. I’m very little interested in abstract discussions about “the future of AI.” (OK, that’s not entirely true, but I’m happy to let that be someone else’s newsletter.) Sign me up for the messy question of what these tools can safely and reliably do right now, under real constraints and in real life.

Smoke billowing from the lab

The project includes experiments, failures, local AI setups, automation workflows, physical-world projects, and ongoing attempts to build a portable “life OS” that stores the rules and context behind everyday decisions. Where needed, I’m ready to build some circuit-based projects, which for me almost always involves generating some unexpected (and component-killing) blue smoke. Think less “guru system” or productivity course, and more a long-running build in public.

The project includes sharing daily notes, screen capture videos, video log style talks, and supporting files and links for those who want to try similar things. It opens a hailing channel for us to talk about new approaches and ideas.

At this point, I’ve got a lot of experience with the big models, the OpenAI and Anthropic stuff, but have only a glancing familiarity with running things locally. And this thing that, in the back of my mind, at least, I think of as a life operating system, is almost entirely unwritten and unimplemented. So you’ll see the whole thing unfold.

The Document Base

I said “almost entirely unwritten” above, because I’ve done just the first little bit. I had a big, long chat with ChatGPT (I know people are balking at some of the politics of Ol’ Chatty right now, but it’s the tool with the most memory about my interests and projects as I write this).

Chatty and I discussed how best to keep a version of whatever the baseline rules and facts were for the “personal OS” and landed on the idea of just putting markdown files in a folder on my local machine. This keeps private things local and also serves as a neutral holding pen for information that will need to be distributed to different interfaces, tools, and models.

To manage the markdown files, Chatty recommended Obsidian and I, with some misgivings, decided to go with it. I feel like maybe I tried it a long time ago and it was hackery and needlessly complex, but maybe that was something else and I was just mixed up. Whatever was going on there, I’ve found it pretty nearly effortless to use so far.

What bothers me most about Obsidian is that there is so much noise and YouTube slop about “Obsidian + Claude Is My Second Brain!” It makes me self-conscious because, damnit, that’s sorta kinda what I’m doing.

In my mind, Obsidian is a tool that lives in the same universe as Notion and OneNote and I guess Evernote (does Evernote still exist, actually?…quick check…it does). And there are strengths to each. This probably isn’t the moment to launch into the pros and cons of each, but I’m intrigued that Obsidian is as powerful as it seems to be absolutely for free. Yes, take a step back and you’re stuck with a bunch of markdown files that are only connected to each other via Obsidian, but markdown files are fabulously more transportable. And it’s easy to point lots of other tools at markdown files with no hassle whatsoever. And this includes using markdown files to provide context and prompts to AI.

The Founding Document

If you’re curious, here’s what the top of the document Chatty created for me when I asked it to produce a summary of what we’d discussed in a format that could be used to prompt an AI to get it back into the same project headspace:

Personal AI Operating System — v0 Synthesis

Core Premise

Build a local-first personal operating system that:

• Stores decisions, rules, preferences, constraints, and open questions
• Is portable (Markdown-based)
• Can be used by AI systems to assist with reasoning and decision-making
• Does not grant AI uncontrolled execution over real-world actions

Primary goal:

Reduce friction in work, money, and daily life by combining AI, structured personal context, and strict constraints

---

Guiding Principles

Local-first

• Core data lives locally (Markdown / database)
• Compatible with local AI systems
• External services are optional, not foundational

Portability

• Markdown as canonical interchange format
• No lock-in to any SaaS platform

To be honest, there are things in the document that I’m already inclined to tweak, but that’s the point, really. I tweak, the AI ingests at my direction. I’ll be curious to see how this prompt behaves when fed to different models, that is, models other than the latest ChatGPT.

I’ll wrap this up here until next week, but during the week you can follow the bits, pieces, and supporting documentation of things I’m building either through PeakZebra.com, or by following my notes on Substack or following @peakzebra on Twitter or BlueSky.

The Explainer Section

I would very much like for this project to be useful to people who haven’t really dived into some or all of the technologies I’ll be working with and writing about, so I’m thinking I’ll have a section (this section) to provide background on things that have appeared in the current newsletter issue.

Markdown

Plain text with lightweight formatting syntax. Widely used in AI workflows because it’s portable, durable, and easy for both humans and models to read.

Local models

AI models that run on your own hardware rather than in the cloud. This matters because they can reduce costs, improve privacy, and allow custom workflows that hosted AI products don’t permit.

OpenClaw

A desktop AI application that lets you run and interact with large language models locally on your own computer rather than through a cloud service like ChatGPT. Tools like OpenClaw make it easier to experiment with private AI workflows, offline use, custom models, and lower ongoing costs — though local models still require reasonably capable hardware to feel fast and responsive.

---

If you’re experimenting with agents and/or local models and you’ve found good stuff I haven’t covered, hit me up because I’d love to know about it too. Reply, comment, share what’s working or something useful you’ve read.

Thanks for reading Building the Assistant! Subscribe for free to receive new posts and support my work.