Anthropic's own course taught me things I didn't know. One of them no longer existed.

A few weeks into using Claude Code daily, I did something I probably should have done on day one. I took Anthropic's own Claude Code in Action course.

Not because I was struggling. Because I had a suspicion I was missing things. Turns out I was right.

The course is excellent. Practical, honest, well-paced. It doesn't feel like marketing. It feels like someone who actually uses the tool every day sat down and wrote what they wished they'd known earlier.

One of the first things it covers is the # memory shortcut. Type # at the start of a message, Claude asks whether to save it to project memory or global memory, done. Natural language in, persistent context out. Elegant.

I opened Claude Code and tried it. Claude Code had no idea what I was talking about.

The # shortcut had been deprecated. The official Anthropic course, published by the people who built the product, was already teaching something the product no longer did.

I laughed. Not at Anthropic. Genuinely with them.

Because this is the moment we are all living in right now. Anthropic is shipping Claude Code faster than they can update their own courses. Enterprises are rolling out AI policies for tools that will look completely different in six months. Developers are writing documentation for APIs that change under them mid-sentence. Practitioners everywhere are building muscle memory for workflows that might not exist next quarter.

Nobody is exempt. Not the companies building the tools. Not the teams adopting them. Not the people writing the courses about them. We are all, every single one of us, learning in motion. The map is being redrawn while we navigate with it.

That's not a failure. That's just what this era actually feels like from the inside.

The replacement for #, by the way, is just editing CLAUDE.md directly. It's better for anything structured. Your content model, your conventions, your team patterns. Stable, version-controlled, loads every session. The shortcut was convenient. CLAUDE.md is more reliable.

The course also taught me two things I had genuinely missed in weeks of daily use.

Planning Mode, activated with Shift + Tab

Claude switches to read-only. It explores your codebase, asks clarifying questions, builds a full plan. Nothing gets changed until you approve. Think of it as asking someone to think out loud before they start moving furniture.

One rule of thumb that stuck with me: if you can describe the change in one sentence, skip it. If you can't, use it.

Thinking Mode, activated by just using words

This one surprised me the most. Include a specific phrase anywhere in your prompt and Claude reasons more deeply before responding. The phrases form a scale, lightest to heaviest: "think", then "think more", then "think a lot", then "think longer", and finally "ultrathink" at the top.

Example: "This is a tough task, so ultrathink about the best way to implement it."

ultrathink sounds like something a ten-year-old made up. It is a real input that changes model behaviour. The terminal even syntax-highlights it in colour, which feels like the product winking at you.

Don't reach for it on everything though. More thinking means more tokens and more time. Save it for the genuinely hard problems - complex bugs, architectural decisions, the legacy code that makes no sense at 4 pm.

You can combine both modes. Planning Mode plus a thinking phrase gives you structured analysis and deep reasoning, with zero files touched until you say go.

The course asked me to fill out a survey at the end. I flagged the deprecated shortcut. I hope someone on the team sees it, because the rest of the material is worth keeping sharp.

We filled out the feedback form. They'll update the course. The product will ship three more features by then. And we'll all keep going, learning in motion, doing our best with the map we have.

That's not a problem to solve. That's just the era we chose to work in.

Wouldn't trade it.