For all of my talk about how terrible LLMs (AI, colloquially) usually are, it might seem surprising to you that I shelled out $100 for a month of Claude MAX*, and will likely do it again.

But my beef with LLMs isn't that they exist. It's almost entirely the way people are using them, which is almost directly a result of the way AI advocates talk about them.

Case in point: Cal Newport's latest podcast episode, Is Claude Mythos Terrifying? | AI Reality Check, completely debunks what Anthropic's CEO said, proving that the "terrifying" risks existed not only in the previous version of Claude, but in much less capable LLMs.

As a reminder, these are the principles of my AI philosophy:

• Don't use LLMs for thinking work
• Don't refer to what LLMs do as "thinking"
• Don't personify LLMs
• LLMs/AI are computers and they are best for doing computer things
• Remember the cardinal rule of programming: computers do exactly what you tell them to do.

It's important to keep in mind that LLMs are word calculators. They're really advanced autocomplete. The most contextual version of search we've ever had.

When you remember those core principles, it becomes easier to determine what LLMs should and should not do for you.

So with all of that, what am I using Claude MAX for? I use it for Computer things.

Here's a short list of concrete solopreneur tasks I've used it for:

Number Crunching: Looking at analytics across my podcast, blog, YouTube channel, and newsletter, and seeing what topics are most popular. I used this info to also help with keyword research, as my general SEO strategy has been bad.

For example, my most popular blog post ever is a review I did on sleep headphones. It's so misaligned with what I do I can't even think of a good CTA, and there are no affiliate links in it (which is, I suspect, why it was so popular).

Processing Customer Research and Feedback: Uploading lots of coaching and customer research calls to find patterns in speech, how my customers talk, and any objections they have.

Aligning on My Goals: Using those crunched numbers and found patterns, I've come up with a general content strategy where the LLM creates "buckets" of content for me based on what works.

An extremely important distinction on the content strategy is that it's not suggesting content ideas. Just "themes" based on what's worked for me in the past. With the themes, I will get out a notebook and write 20-50 actual topics or ideas I have, then feed those into the Content Strategy skill, and have the LLM sort my ideas into the predetermined buckets and flag anything that doesn't fit into those buckets so I can either refine, nix, or do it anyway.

Content Packaging: Using all of the above, I have projects and/or skills to help take the content I've come up with and create packaging (titles, descriptions, thumbnails if needed). It also makes sure that I'm using all of my stated keywords in some way.

All of these things are possible because of two things:

1. A much bigger context window and Claude doing clever things to keep the context over long chats.
2. The ability for Claude Cowork to access folders (one of which I'm calling AI Memory) and connectors like vidIQ, Notion, Senja, Supernormal, Todoist, HubSpot, and more.

Giving an LLM the ability to access as full a context as possible makes its answers "smarter," but it can also crunch data better, which is really the crux of what I'm using it for.

A few other notable use-cases for solopreneurs:

1. Copy Editing: This is a long-standing use-case. It never touches my work; it only proofs and suggests changes. It's just a lot "smarter" now because it has much more context and data.
2. Deep Research: Mostly it surfaces a large number of primary sources for a piece I'm working on. Something worth mentioning is I never read the summary alone. I read the summary, then go to the sources that seem the best and pull all of my references from there.
3. Coding: I have feelings about this, but it's been really good at code. After all, it is a computer speaking the language of computers.
4. Call Summaries: Cowork has scheduled tasks. Every day at 6pm I have it check my email for Gemini Notes + Supernormal for other call notes, summarize everything, and add my action items to Todoist.
5. Calendar Management: My calendar app, Fantastical, has MCP, so it can access all of my calendars without giving the LLM direct access to all of my accounts.

One pretty cool thing I used Cowork for was putting all weekday afternoon Yankee games on my calendar. I usually don't realize the Yankees are playing an afternoon game during the work week until after I have a conflict. So I had Claude:

1. Go out and get the Yankees' full schedule.
2. Determine which games were being played between 11am and 6pm ET Monday - Friday.
3. Put those dates on my primary calendar so they're blocked in Cal.com / I know not to schedule meetings at those times.
4. Add the entire schedule to a separate calendar called "Yankee Games."

Once I fixed a sandboxing issue (that is, a security issue preventing Claude from going to any website), it worked flawlessly. This saved me a ton of time for so many reasons...one being it's not really a task I needed to do.

Now, it hasn't been perfect. Far from it; it still does that LLM thing of making stuff up, forgetting directives, or overzealously suggesting too much. It even made a change to Notion I didn't want it to.

One of my shortcomings is that I don't think I've read its output as carefully as I should or could have. One of my tasks for this week is to audit everything in the AI Memory folder and make adjustments — something I can do without an LLM because it's all plain text written in English.

On balance, it has saved me a ton of time and enabled me to do more targeted work than I would have...effectively and efficiently. Combing through all of my analytics, for example, isn't something I would have done well, if at all.

That said, I want to make this really clear: I trust my gut more than the LLM's output. Some of the rules I have are baked in to keep me accountable and remind me of decisions I've made. But it will often focus too much on a rule or concept (like everything pointing back to my Solopreneur Sweep).

Because ultimately, the cardinal rule of programming still applies: computers do exactly what you tell them to do.

*I know MAX is a plan, but the context of "more expensive plan" is important here.

My latest from around the web:

LAST WEEK ON STREAMLINED SOLOPRENEUR What is AI Brain? STREAMLINED SOLOPRENEUR: AUTOMATE YOUR BUSINESS, TAKE TIME OFF WORRY-FREE • S2:E What is AI Brain + Recommendations [Friday Wrap-Up] 16:45 MORE INFO I have a new series on the podcast called "Friday Wrap-Up." Here's the episode from this past friday, where I talk about "AI Brain" and make some recommendations. Listen at Streamlined Solopreneur

Other Stuff:

• ​Efficiency Isn’t Enough: What Every Solopreneur Gets Wrong About Productivity: In business, we often hear about what’s “efficient.” “Can I do this thing faster?” But efficient doesn’t always translate to effective…so is it really efficient?
• Quick programming note: Since I'm doing Monday articles and the new Friday Wrap-Up, I'm moving main episode of Streamlined Solopreneur to Tuesday (where they started back in the How I Built It days)! Look for a new episode on Discipline is Destiny tomorrow.

And let me know: are you using Claude MAX? What do you think of my rules? What's your favorite use case?

Reply and let me know!