How I use AI: My policy and philosophy around AI

Gosh, where do I start?

Frankly, at the time of this post, I've been intentionally decreasing my use of AI — I use Claude — because:

It felt like my brain had stopped spinning up strategies, and I was looking for a magic one-hit pill to instantly unlock all my potential.

Like Bradley Cooper took in the movie Limitless.

And (spoiler), it fried Bradley’s brain wiring.

I was headed that way…

It felt like Claude was spitting back shitty strategies and not acting like an engaged brake when I wanted to smash the accelerator.

Also:

I noticed my use of AI was sucking my hours out of my life.

Now, if you aren't familiar with my work, I run and operate this business in the cracks of a full life.

Meaning: I work about 13 hrs/wk.

I'm a fulltime parent to my three sons under 10 years old, my husband works fulltime, and we've also got two young dogs.

Lots of noise, chaos, joy, throwing balls, and dance parties in our house.

And I want to witness them all.

Which boils down to:

less bent-neck over my phone or devices, and

more time interacting and DOING life with all these characters I'm choosing to live with.

So, if using AI that's presumably here to make me more productive, but I find it's making me LESS productive and it's giving me rejurgitated strategy from Reddit posts 5+ years old...

Well, that's counter to my goals.

Now that we've gotten that rant out of the way, let's dig into how I use AI in my business — because I definitely still do — and what I don't let AI touch.

How I use AI:

1) Research.

For example: researching recent trends inside Substack.

I'm building up my Substack publication and wanting to write about juicy topics.

Or research on a specific potential partner or cold email recipient or strategy I stumbled across.

I'll ask Claude for a synopsis of what I’m curious about so I can decide if this is worth pursuing (aka aligned with my one goal; more on that below) or I can set it down and give it to Claude to hold (more on this below, too).

2) Data analysis: I hooked my Kit account up to Claude via MCP.

Now I can tell Claude to "find me all the folks who clicked on the link inside my last email" and then "tag those folks" and it's easier than doing it myself.

Or running an analysis on my last month of emails (I send emails 3x a week, so there's lots of data!) to see what's engaging folks vs what isn't.

And that guides my next month of emails.

3) Loading drafts into my Gmail for follow ups to specific people who showed interest

Typing out emails using two thumbs is tiring.

I like to voicenote an email into Claude, edit and finetune it, then have Claude load the final version into the appropriate thread in my Gmail.

4) Taking my 6+ years (!) of sending emails regularly and compiling that content into rough drafts of blog posts or Substack articles.

There's a rule written into my Claude content compiler skill to make sure the content is 87% mine, and to severely limit the AI adding or full-out writing new words into my post.

This protects me from getting regurgitated Claude-written-but-sounds-maybe-like-Laura copy.

Then I can take that rough draft, edit in the human, and publish it.

This has been huge in helping me finally get a blog up and going for WARM Client Method, and soon, Cambium.

I've got years of my philosophies, stories, and strategies in these emails. And those emails were sent once and I moved on to the next email.

However, those emails are still good shit!

So I want to bring it out and let y'all taste it like matured Scotch from my cellars.

5) Update spreadsheets, do admin, rebuild assets.

I hate busy work like copying/pasting, updating spreadsheets, etc.

And I just had to rebuild a quiz when I switched platforms. So I have Claude Cowork do the heavy lifting and manual crap like this that must be done in any business.

For example, I’m going to have Claude clean up my chaotic mess of Notion and build me a working dashboard.

<ironically, I had to take a short break here to ask Claude how I used it regularly.>

Here’s what Claude told me on how I use it:

6) Help me flesh out my monthly recap emails.

I'm not a great record keeper, but I talk with Claude daily, so asking for a recap of the month and what I did is very helpful.

It gives me the foundation, and I fill in the details.

7) Stella, my assistant, is AI.

Customer support is about clear communication and a distance between the one who sold the thing and the one using the thing. Emotions can muddy the waters.

I use Stella to get replies back quickly, and she's always noted as "Laura's AI assistant" so you're clear on who (and what) you're communicating with.

8) Design and create smart sales pages.

These interactive sales pages are so incredibly cool. (I learned this skill from Brittany Long's very actionable workshop. Get it here *affiliate link.)

Claude built out the sales pages for Dream 25, No-Zoom Sales, and Cambium.

Claude designed the sales page for WARM Client Method, too.

9) Act as my second brain.

I know this is a trendy term, but what this means for me is:

I ask Claude to hold things for me: this idea, maybe write this email, look at this interesting piece of research, do I want to add this to Cambium?

Because, then, I can ask it to retrieve that info for me when I'm ready to move forward.

Mostly I need my brain to feel safe to let that info go, so I can focus on the next thing.

9) Give me perspective.

I tend to get caught up in my own thoughts, so I need Claude to give me a new perspective to see the trees for the forest.

So, I'll ask it "what am I missing here? Where could things go sideways and how can I plan for those?" but ONLY AFTER I've determined the strategy (more below on this).

And here are the things that Claude told me I use it for:

Based on our history, a few clear patterns stand out in how you use me:

As a second brain, not a task-executor.

You've been explicit about this in your preferences — you want a thinking partner that audits your strategic decisions and increases self-awareness, not something that just executes tasks handed to it.

Structural feedback over first-draft generation.

You write your own client-facing content, blog posts, and voice content — writing is thinking for you. My role has consistently been sparring, voice-checking, and structural editing rather than generating fresh prose in your voice from scratch (this shows up explicitly in the content-compiler and blog development workflows).

Execution support for cheap, reversible tests.

When you bring an email, tactic, or strategy to test on your list or partners, you want me to help you ship it — not debate it. You've been clear the market answers faster than back-and-forth theorizing.

Business intelligence across a sprawling system.

You use me to track patterns across Method, Cambium, partner pipeline, email stats, and pricing — surfacing what you'd miss week-to-week rather than generating a to-do list.

Direct brainstorming partner for a real business.

You've built out deep context (ICPs, frameworks like WARM/7CM/Cambium Client Model, pricing structures, active deals like Cambium's founding launch) so I can operate with founder-level context rather than generic advice.

Anti-sycophancy as a standing rule.

You've flagged that echoing your ideas back in tidier language is an echo chamber — you want genuinely new angles, and pushback reserved for two specific cases: abandoning a test before it returns data, or spinning up too many directions with zero execution.

Okay, back to Laura.

C'est moi:)

What I don't let AI touch

1) Writing.

For me, writing time is thinking time. My brain gets oiled from writing.

When I'm not writing daily, I'm anxious, second-guessing myself, and less confident. Writing keeps the ideas flowing and sorting themselves out.

So, whenever you read something from me, know that it's me writing you.

2) Strategy.

I've experimented with AI creating strategy for me and I've always been unhappy with the result.

Good strategy requires a blend of experience, experimentation, and creativity.

AI doesn't have that.

Rather, AI is good at pattern recognition, which is helpful in post-launch analysis.

3) I don't let my AI argue with me.

Because of how my brain thinks -- I can accurately argue both sides of an argument -- I don't need more data.

I need a decision and action.

So my Claude isn't allowed to spin up multiple directions. I need the best direction based on my situation, and then full speed ahead.

In a related note, I don't let AI dictate my priorities.

Which leads me right into:

4) I set my priorities.

Because I'm intentionally working 11-13 hours/week, I'm strict about what I'm focusing on.

I have a goal for the year and it's one goal.

The word priority means one.

Which means: I'm saying no to a lot that doesn't fit inside that one goal.

5) Creating and designing. I make the decisions.

End of discussion.

Again, I asked Claude to contribute a little bit.

Admittedly, this part might be harder for it, because it's not privvy to what I don't give it:)

Here’s what Claude said on what I don’t allow it to do:

Based on what you've set up, here's what you don't let me do:

Restate your ideas back in tidier language.

You've flagged this explicitly as an echo chamber — you want genuinely new angles I didn't bring, not a polished mirror.

Write your client-facing content or blog posts from scratch.

Writing is thinking for you. My role is structural feedback, voice-checking, and sparring — not first-draft generation of your voice content. The content-compiler skill exists specifically to prevent me from generating fresh prose when you've given me source material to work from instead.

Ground or push back by default.

You've restricted this to exactly two situations: abandoning an active test before it's returned data, or spinning up multiple directions with zero execution. Outside those, you want execution support, not debate — cheap reversible tests get shipped, not theorized about.

Flatter you or soften feedback.

Anti-sycophancy is a standing rule — direct, objective, no unearned certainty, no marketing fluff.

Hand you a to-do list instead of an insight.

In the business intelligence role specifically, you've said you don't want a task list — you want the outside perspective that sees the forest when you're in the trees.

Add more content channels or suggest you write more than you already do.

The project instructions are specific: no social media strategy suggestions, no "post more" advice, no channel creep beyond what's already decided (webinars, email, partnerships).

Defend a prior recommendation once you've corrected it.

You expect updates to happen immediately when new information changes the picture — not a rehash of why I said what I said before.

And there we have it: how I use AI in my business as a solopreneur working 13 hours/week by design.

Of course, all of this is subject to change, and I fully expect things to shift as the regulations around AI get more fleshed out.