6 Claude Lessons a Retired Pharmacist Learned Fast

I’ll be honest. Every time a new Claude model drops, I forget half of what I knew about using the old ones. I bet you do too. That’s why this post stopped me mid-scroll. It’s a reminder that you won’t get good at Claude unless you keep going back to the basics, again and again.

The author recently taught Claude to their mom, a retired pharmacist. They boiled the whole thing down to six lessons. I love this approach. If a lesson works for someone who’s never touched AI, it’ll make the rest of us sharper too.

Here are the six lessons from that session. I’ve added a few notes on how you can put each one to work.

1. There’s not one Claude. There are four.

This is the first thing the expert wants you to get. Claude isn’t one tool. It’s a family of models, and picking the wrong one costs you quality or tokens.

The original poster doesn’t hold back here. They call Haiku 4.5 and Sonnet 5.5 the two weakest options and say you shouldn’t rely on them, even when they’re free. They’d rather use free ChatGPT. Their one exception is quick spelling checks.

For anything complex, the advice is to start with Fable 5.2 for your first two prompts. Use those prompts to work out what the final result should look like. Then hand the actual work to Opus 5.5.

2. Open every complex chat with a planning prompt

This is my favorite part. The author shared a simple starter prompt to paste into any new chat on Fable 5.2. Here it is, exactly as written:

“Goal: a step-by-step plan for [your goal]. Don’t build it yet.
Context: [what you have, what you want at the end, your limits: budget, deadline, what’s off-limits]
If you need more context, ask me questions.”

See the “Don’t build it yet” line? That’s what makes it work. You get a plan first, so you’re not stuck fixing a half-built answer later.

The next rule matters just as much: don’t switch models in the same chat. Once the plan is ready, ask Fable for a handoff brief. Then start a fresh chat on Opus 5.5 (High) and paste the brief in. Here’s the handoff prompt the creator uses:

“Write a brief so a new chat on Claude Opus 5.5 can take over this work. It won’t see this chat or your thinking. Write it for a smart reader who saw none of this. Plain words. Copy the exact names, numbers, dates, links and file names from this chat. Only the final version of anything we changed. Add nothing we didn’t say.”

I think this is really clever. The new chat gets a clean, compressed version of everything that matters, and none of the clutter. It works like handing a project to a new coworker with a well-written memo.

3. You can’t fully escape AI writing patterns

The LinkedIn creator is upfront about this one. AI text has tells, and you won’t scrub them all out. You can make them a lot less obvious, though, by steering Claude away from the most common giveaways.

A practical way to start: keep a running list of the phrases and habits that make text sound robotic to you. Then give that list to Claude as things to avoid. Lesson 6 has a related trick that makes this even stronger.

4. Five Claude features you’ve probably never tried

This section is a gold mine. The expert listed five features that most people skip:

  • Research mode: Claude plans its search, reads sources, and gives you something close to an analyst report.
  • Claude Artifacts: You can build small working apps and share them publicly with a link.
  • HTML images: Claude isn’t built to make images. But ask for an image in HTML format and it’ll never get the text wrong.
  • Incognito mode: Find it in the top right. Use it when you don’t want memory to color a chat, especially when you want a genuinely fresh idea.
  • Interactive charts: Upload a CSV and ask for an interactive chart. The author says they use this all the time.

The HTML image trick surprised me the most. Text in AI-generated images is often garbled. Because HTML renders text as real text, that problem just disappears. It’s perfect for simple cheat sheets, quote cards, or labeled diagrams.

5. Don’t make your Claude dumber

If you only take one habit from this post, take this one. The original poster says to start a new chat after 40 prompts in the old one.

Long chats pile up context. Past a certain point, Claude starts to hallucinate and drifts away from what you asked for. A fresh chat, maybe with a handoff brief from Lesson 2, resets the whole thing. It’s a cheap fix for a lot of frustration.

6. Five behaviors that are a hard no

The post ends with a list of habits to drop right away:

  • Don’t use Fable 5 for jobs Opus 5 can handle. You’ll waste a lot of tokens.
  • Don’t switch models mid-chat. Claude burns your tokens re-reading the whole conversation.
  • DO NOT connect your bank accounts. It should be obvious.
  • Don’t prompt with adjectives. Paste an example of bad writing and say: never write like this.
  • Don’t chase every new launch. If something works, stick with it.

The adjective tip is underrated. Words like “punchy” or “professional” mean something different to everyone, Claude included. A real example of what you don’t want gives Claude a much clearer target.

The last point might be the most important one for keeping your sanity. New AI features show up every week. Constantly switching tools feels productive, but a solid workflow you know inside out usually beats it.

Why this works so well

None of these lessons needs technical skill. They’re habits: pick the right model, plan before you build, keep chats short, and hand off context cleanly. A retired pharmacist could pick them up in one sitting, which tells you they’re the kind of fundamentals worth relearning every time the tools change.

My suggestion: try the two prompts from Lesson 2 on your next real project and see how much cleaner the results get. Then head over to the full LinkedIn post to see the original lessons and join the conversation.

Scroll to Top