I’ve been interested in AI for a while, but keeping up with it is its own full-time job — models change, tools change, whole paradigms shift in the span of a few months. At the start of the year I signed up for the AI Daily Brief New Year’s Resolution mostly because it gave me something concrete to work toward. Ten weekends, a structured path, real skills to come out the other end. In a space that moves this fast, an opinionated curriculum is worth something.
Week 1 got done. Week 2 got started but never finished. I found I was outgrowing it.
Prompting Is on the List. Other Things Jumped the Queue.
Prompt engineering is a real skill and one I intend to develop. The resolution was a reasonable way to do that — learning to craft better inputs, compare models, understand their strengths. Useful stuff, especially when most of the work happens in a chat window.
What I found was that once I left the chat window, other skills became more pressing. Working with tools like Claude Code, the actual leverage isn’t how cleverly a request is phrased — it’s in how well the context is organized and how clearly tools, like MCP servers, are integrated. A well-structured project beats a well-crafted prompt, pretty much every time. That felt more urgent to get right.
The Moment the Problem Changes Shape
I’m a software engineer at WP Engine. Most of what I do day to day is what software engineers have always done: shipping features, fixing bugs, improving existing systems. What’s changed is the assistance layer — tools like Claude Code mean I’m rarely doing that work alone, and the collaboration is different from anything I’ve used before.
But even there, what I’ve noticed is that the prompting skills the resolution was building aren’t the bottleneck. The bottleneck is context — making sure the agent understands the codebase, the constraints, the goals well enough to be a real collaborator rather than a very fast autocomplete. That’s a different skill entirely. And it’s pointing me toward something I want to build next: custom agent harnesses that can do meaningful automated work on projects like Local. Not just AI-assisted development, but AI-driven workflows.
That distinction — assistance vs. automation — is where things get interesting.
Two Tracks, One Shift
I’ve started thinking about my AI work as two tracks that are developing in parallel.
- Work track: Right now this is growing my skills around agentic tools and AI-assisted development. The near-term direction is building and maintaining custom agent harnesses that can tackle larger, more complex work. The AI isn’t the product; it’s the infrastructure that makes the engineering better.
- Personal track: Building an actual assistant. Not a chatbot. Not a suite of tools. A named, persistent entity that knows my goals, my context, and my preferences — and works on my behalf without me having to think about it much.
Both tracks are moving away from the chat window. Both care a lot more about context architecture than prompting craft. And the personal track in particular led me somewhere that made the resolution feel like a different conversation entirely.
The Assistance Paradigm
Daniel Miessler — who created the Fabric toolkit and has been writing about personal AI infrastructure since 2016 — puts it well in his Personal AI Maturity Model. The model has three tiers: Chatbots, then Agents, then Assistants. Most people are still deep in the agent tier. The interesting work is starting to happen at the assistant tier.
| Tier | Stages | Where we are |
|---|---|---|
| Chatbots | CH1–CH3 | Basic chat → advanced tooling and context (2022–2024) |
| Agents | AG1–AG3 | Standalone frameworks → continuous background operation (late 2024–2027) |
| Assistants | AS1–AS3 | Named companions → proactive, trusted advocates (2027–2030+) |
The difference isn’t just complexity. It’s orientation. Chatbots respond to requests. Agents execute assigned tasks. Assistants advocate — understanding goals, monitoring the gap between current state and where things should be, closing it without being asked.
That framing clicked something for me. The AI Daily Brief’s New Year’s Resolution was optimizing me for the chatbot tier. My work track is pushing me into the agent tier. And personally, I’m drawn to the assistant tier — which requires a completely different kind of thinking.
What I’ve Been Actually Building
The tool I’ve started exploring on the personal track is PAI (Personal AI Infrastructure), also by Miessler. It’s open source, it’s opinionated, and it’s built around the idea of a single named assistant — in Daniel’s case, his assistant is called Kai — that acts as a persistent interface to everything else.
The setup involves defining goals, context, and preferences — what Miessler calls an “ideal state” — and giving that to the assistant as the thing it’s constantly working toward. The agents, the skills, the workflows all sit behind that interface. There’s no direct interaction with the infrastructure. There’s just my interaction with my assistant.
I’m early in exploring it. But the mental model it requires is so different from “write better prompts” that I couldn’t keep treating both as the same kind of work. His Personal AI Maturity Model is worth a read.
What the Resolution Actually Gave Me
I didn’t finish the 10 weekends. I also don’t think the resolution failed — it gave me exactly what I needed, just not what I expected.
Week 1 forced me to build something and ship it. That’s always useful. Week 2 pushed me to think systematically about which models are good at what — which turns out to matter when building agent systems, where picking the right model for the right task in a pipeline is a real decision.
But the bigger thing was the clarity. When I stopped trying to complete the resolution and started asking why it didn’t feel urgent anymore, the answer told me a lot about where my head actually was. I wasn’t thinking about chat windows. I was thinking about context, about systems, about what it would look like to have a persistent AI collaborator rather than a tool I pick up and put down.
That’s a different kind of skill to develop. And I think it’s the one worth developing.