Vol 2, Issue 1 | JAN 7 2026

Dear Reader,

Something shifted over the holidays.

Since Claude Code Opus 4.5 launched in November, a pattern has emerged among knowledge workers using Claude Code with their Obsidian vaults (or any plain-text knowledge system). They're discovering that plain text files combined with the right AI tooling create compounding productivity gains — the kind where the system gets smarter the more you use it.

There's a real transition happening from "How do I prompt better?" to "What gaps in my life can I program with a skill or an agent?"

The first question is tactical. Useful, but limited. It assumes AI is a thing you talk to when you need something.

The second question is strategic. It treats AI as infrastructure — something that can be systematically extended to handle the recurring friction in your life. We're starting to see the promise of computers as thinking partners, rather than just question and answer machines.

This mental model shift is what matters, more than any specific tool or technique. While I'm not going to get into the debate over what constitutes real agentic AI or if we're going to see AGI in 2026, I will say that we need to stop thinking as much about prompting as we do about helping to identify and plug gaps in our own systems.

In Practice

Here's a real example from my own setup — not work productivity, but life admin.

With multiple schedules to manage (personal, work, kids), staying on top of everything is constant work. Over the holiday break, I worked through this with Claude. I explained where my gaps were: knowing day-to-day what the kids' lunch and homework schedule is, whether my health insurance covers something, when bills are due.

I built an agent called home-life-ceo. Its job is to be in charge of our household logistics. It knows when our insurance renewals happen, when school holidays fall, what the home maintenance schedule looks like, when the car needs servicing.

The key detail: it searches my vault — plain text markdown files stored locally — not the web. Every time I add a document about a new insurance policy or a school calendar, the agent gets smarter.

More than that, I’m treating it like a manager that delegates to other specialised agents I’ve built:

• financial-strategist: Business finance, pricing, revenue analysis, detailed spending
• workflow-coordinator: Business/work task orchestration
• vault-organizer: Filing and organizing vault content
• learning-coach: Personal development and skill building
• librarian: Book library research for health, finance, life management advice
• perplexity-researcher: Deep web research on any personal life topic
• travel-planner: Trip planning, flights, accommodation, family travel logistics
• home-maintenance: Spanish home maintenance, vendors, warranties, seasonal tasks
• education-researcher: School decisions, kids activities, educational resources

Each of those agents connects to different parts of my notes and apps. The financial-strategist, for example, reads directly from my finance software’s database. I still open the app, manage accounts, balance transactions — but now I have a partner that can analyze spending patterns, spot budget gaps, and help me plan for bigger expenses.

It’s not about replacing manual work. It’s about having context-aware help when I need it.

If you want to see how this works in practice, I’m documenting and updating my entire setup at cerebro.jimchristian.net. Not as a template to copy, but as proof that this approach works at scale.

The Compounding Context Principle

Here’s what makes this different from most AI productivity approaches:

More documentation = smarter system. Every note I add to my Obsidian vault becomes context the agent (or agents I program) can reference. The system compounds over use and time.

Plain text files = vendor independence. Markdown files work with Claude Code today and any LLM tomorrow. I’m not locked into a proprietary format.

Local-first = private by default. My family’s insurance details and school schedules stay on my machine, not in someone else’s cloud.

This is fundamentally different from consumptive AI tools — the ones where you type a question, get an answer, and the interaction vanishes. Those tools don’t get smarter over time. They reset to zero with every conversation. That’s likely going to change as time and AI progresses, but this approach gives you a head start and builds institutional memory into your life.

---

When the System Starts Solving Its Own Problems

Here’s where it gets helpfully recursive.

Eventually, I’d built enough tools that I needed a tool to track my tools. So I built one — a skill called Xita's Sheets that catalogs every skill, agent, and automation in my system.

Honestly, having dozens of custom skills creates its own management overhead, but because it demonstrates the compounding pattern in action. Every solved gap reveals another gap. The more friction you eliminate, the more visible the remaining friction becomes.

This feels endless at first — then it starts feeling like progress.

What Makes This Possible

Everything I’ve described runs on a system I call Minervia — a framework for vendor-independent knowledge work built on plain text, local-first principles. A ‘Co-Operating System’, if you will.

The philosophy: your knowledge system should outlive any particular AI tool. Plain text files have survived every technology shift of the last fifty years whereas proprietary formats have not.

​Minervia provides the conceptual foundation; the specific implementation is up to you.

Open Source By Default

I’ve open-sourced my most useful skills, agents, and MCP plugins at signalovernoise.at/open-source.

But here’s my honest advice: don’t just copy the setup.

The real value isn’t in the specific automations I’ve built — it’s in learning to think in gaps. Look at your day and ask: what do I do repeatedly that could be systematised? What do I look up over and over that could live in context?

Those questions matter more than any particular tool.

Your Homework

Here’s what I want you to sit with this week:

What do you do repeatedly? The task that shows up every week or month that you handle manually each time.

What do you re-lookup constantly? The information you know exists somewhere in your files but can never find when you need it.

What would you automate if it was easy? The friction you’ve learned to live with because the alternative seemed too complicated.

2026 is the year to stop living with those gaps. The tools are mature enough now that building your own automation is accessible to anyone comfortable with text files and basic configuration.

If you want guided support making this transition, I’m offering implementation coaching for the new year at signalovernoise.at/coaching. But honestly, you can start on your own. The open-source materials are there, and the thinking is what matters most.

Until next week,

Jim