I use three AI tools. Each one fits a different context. Here’s how they divide up my work — and what actually surprised me after six months of doing this for real.
I use three AI tools regularly, and they don’t overlap much.
At work, I use Copilot for most day-to-day PM tasks — drafting communications, summarizing meetings, working through stakeholder situations. I also have a Cursor license because our developers use it and I wanted to understand what they’re working with. Understanding your team’s tools isn’t optional if you’re going to manage the work intelligently.
At home, I use Claude Code. That’s what this post is about.
What “At Home” Actually Means
When I say “at home,” I mean a surprisingly large surface area.
This blog. A recipe app I’m building with my husband Scott. A personal journal app — offline-first, end-to-end encrypted, serious software. A garden journal. Health tracking. Business planning. Six repositories, each with its own context, constraints, and conventions.
Claude Code is where all of that lives. Not because it’s the only tool that could handle it, but because I’ve trained it on how I work — and at this point, it knows my projects better than any other tool I’ve tried.
The Thing That Actually Surprised Me
I expected AI to be useful. I didn’t expect it to sound like me.
After a few months of working with Claude Code on this blog, I noticed the output didn’t read like generic AI content. It sounds like how I write. Direct. No filler phrases. Practical before theoretical. The kind of writing where if you asked someone to guess whether a PM wrote it or a content marketer wrote it, they’d say PM.
That didn’t happen automatically. It happened because I shared my writing with it — including papers I wrote for my master’s degree, before ChatGPT existed. Academic work written under deadline, graded on clarity and argument, with no AI assist. That’s a clean sample of how I actually think and communicate.
Combined with the blog posts I’ve published since, Claude Code has enough of my voice that when it drafts, it’s working from something real. My writing style is in the project memory. When I start a new post, I’m not fighting the AI’s default tone. I’m working with my own.
That’s the part most people skip. They try AI, get generic output, and conclude it doesn’t work. The tool usually isn’t the problem. The training is.
And to be clear about what “AI-drafted” means on this blog: every post is my story, in my voice. I provide the information. Claude may ask follow-up questions, which I answer. It drafts from what I’ve given it. Then I review to make sure the story is correct and accurately reflects me — sometimes that’s minor edits, sometimes it’s significant. AI isn’t perfect. But it’s very good at drafting, and drafting is where a blank page becomes something to work with.
Guardrails: The Part Scott Helped Build
One of the most useful things about Claude Code is the ability to set rules the AI follows consistently. These live in CLAUDE.md files — and there are two levels.
A global CLAUDE.md applies to every project, even new ones. Mine has things like commit message format, how I want multi-step work planned before execution, and how I want responses structured. It follows me into every repo automatically.
Each project also has its own project-level CLAUDE.md with rules specific to that context. Mine aren’t suggestions. They’re constraints.
Content belongs in the right repo. Six related repositories means a lot of opportunity to let things drift — and I have been caught by this rule more times than I’d like to admit. If I’m working on the sustainability site and start heading toward recipe app features, the AI stops me and tells me which repo that belongs in. Without the guardrail, I’d keep going and create a mess. With it, I get redirected before the damage is done.
Financial data requires validation. I write about the costs and savings from our zero-emissions home — solar, geothermal, EVs. Before anything goes live, the rule is to validate figures against actual invoices and date every number. The AI enforces this in review. It doesn’t let me wave at a number and move on.
Git discipline. Every repo has a branch-before-commit rule. I don’t commit directly to main. The instructions are specific enough that the AI will ask, redirect, or just refuse depending on how the rule is written.
Content integrity. Vendor recommendations get disclosed. Paid relationships get disclosed. This isn’t just an ethical preference — it’s written into the project instructions so it’s never an afterthought.
Scott helped architect the guardrails for the app development side. Building a journal app with server-side encryption and offline sync isn’t simple. Having rules that enforce TypeScript, encryption requirements, and API-first design means I’m not accidentally making choices that undermine those constraints halfway through a build.
The apps aren’t published yet. But the foundation is right — and the guardrails are part of why.
What I Use It For, Practically
On any given week, Claude Code helps me:
- Draft and refine blog posts in a voice that sounds like mine
- Manage the blog schedule — tracking status, scheduling posts, finding gaps
- Work on the recipe app and journal app alongside Scott
- Keep all six repos organized without context bleeding between them
It also remembers things across sessions. How I like the schedule displayed. Which posts are waiting for external approval. What’s in progress and what’s next. That continuity is what separates it from a chatbot. A chatbot answers questions. Claude Code is a working context that picks up where we left off.
What This Post Isn’t
This is not a deep dive into prompting strategy, memory management, or advanced Claude workflows. Those are coming — there’s a whole series planned on leveling up with each tool, including a dedicated post on Claude specifically.
This is just the honest answer to a question I get occasionally: what do you actually use AI for?
Three tools. Clear contexts. One trained on my voice, my projects, and my rules.
That’s the setup. The tactics are in the Level Up posts.
Part of the AI + PMO series. Coming soon: Leveling Up Claude — the habits, workflows, and memory setup that make the difference.
Related posts:
- AI + PMO: Manual Work to Eliminate — what to automate first
- A PM’s Guide to Choosing an AI Tool — when to use which tool
- AI Ethics for PMs — the guardrails that matter at work