Reading Time: 5 minutes

Hey Prompt Lover,

I need to tell you something and I need you to actually read it.

Since January, Anthropic has been shipping something new almost every single day.

Not small stuff either.

New models.

A desktop agent.

A tool that controls your computer.

Phone to desktop control.

Excel and PowerPoint integration.

A security scanner.

A plugin marketplace.

Scheduled tasks that run while you sleep.

And while all of that was happening, most people were still typing "write me an email" into ChatGPT.

I'm not saying that to be harsh. I'm saying it because there's a real gap opening up right now between people who know what Claude can actually do and people who think it's just a smarter Google.

That gap is going to matter.

So let me walk you through what actually shipped, what it does, and why you should care.

Ready to transform artificial intelligence from a buzzword into your personal revenue generator?

HubSpot’s groundbreaking guide "200+ AI-Powered Income Ideas" is your gateway to financial innovation in the digital age.

Inside you'll discover:

Download your guide today and unlock a future where artificial intelligence powers your success. Your next income stream is waiting.

Get Your Guide

Let's start with the models because everything else runs on top of them.

There are three now.

Opus 4.6 is the heavy one. It can hold roughly 750,000 words in its memory at once. That's your entire codebase. A year of emails. A full legal contract with every amendment. All in one go, no chunking, no workarounds. It's expensive at $25 per million output tokens, so you don't use it for everything. You use it when the work actually matters.

Sonnet 4.6 is what most of you should be using every day. Same massive memory window as Opus. 30 to 50 percent faster than the old version. And here's the part that surprised even me — in testing, people chose it over the previous flagship model 59 percent of the time. It's that good now.

Haiku 4.5 is the fast cheap one for when you're running things at volume. One thing to know: it has no protection against prompt injection. If you're automating things with untrusted input running through it, go read the docs before you let it loose.

Now here's the pricing change that nobody made enough noise about.

Before March 13, if you sent a request over 200,000 tokens, Anthropic charged you extra. A surcharge on top of the normal rate.

That's gone now.

A massive request costs the same per token as a small one. One company actually found their total costs went down after increasing their context size, because the model stopped wasting tokens re-reading old information to catch itself up.

That's the kind of thing that quietly changes how you work.

Now let me tell you about Cowork, because this is the one that made stock markets move.

When it launched in January, SaaS stocks dropped hundreds of billions in market cap within days.

That's not hype. That's investors doing math.

Here's what Cowork actually is. It's not a chat window. It's not a smarter assistant. It's a thing that sits on your computer, reads your files, writes your files, makes a plan, executes the plan, and puts finished work in a folder for you.

You describe what done looks like. You walk away. You come back to work that's done.

I know that sounds like marketing. It's not. Someone in the community asked it to go through 320 podcast transcripts and pull out the ten most important lessons for builders. Another person used it to organize 317 Disney World videos by GPS location.

Someone else came back to find fourteen job descriptions, a marketing strategy document, forty-seven emails, and website copy sitting in their folder.

The tool built itself in ten days. Anthropic used Claude Code to build Cowork. That part should make you stop for a second.

The reason most people struggle with it is they treat it like ChatGPT.

They write a long detailed prompt every time and wonder why the output is inconsistent.

The people getting great results spent one afternoon building three files.

A file about who they are and what they actually work on day to day. A file about how they write — their tone, their style, words they'd never use, a few paragraphs of their real writing as a sample. And a file telling Claude how to behave — ask before doing, never delete without permission, default formats, that kind of thing.

Those three files fix the cold start problem for good.

Every session after that, Claude already knows who you are. It already knows how you write. It already knows your standards. You stop re-explaining yourself every single time.

One person said after building those files they gave Claude a ten-word prompt and got a client-ready document on the first pass.

That's not a coincidence.

Since launch, they've kept adding to Cowork and the additions are good.

Connectors landed in February. Claude now plugs into Google Drive, Gmail, Slack, Google Calendar, and fifty-plus other tools. Not shallow integrations either. It can go into your Drive, find a document, read it, and use what it found to write something else. You set it up once and it's there in every session from then on.

Scheduled tasks came right after that. You tell Claude once to do something on a recurring basis and it does it without you being there. One person has it running every Monday morning — research three competitors, check relevant industry news, save a formatted brief to a specific folder. They wake up to a finished document. They didn't do anything.

Dispatch came in March. This one is the phone to desktop bridge. You're on a train. You message Claude from your phone. It executes on your computer. By the time you're back at your desk, the work is done. You didn't have to be sitting there.

And today — literally today as I'm writing this — Computer Use launched.

Claude can now use your mouse and keyboard. It opens apps. It fills in forms. It navigates websites. Anything you can do sitting at your computer, Claude can now do sitting at your computer.

That's not a small thing.

For the people who write code, Claude Code has become something different too.

It started as a command line tool. It's now pulling two and a half billion dollars in annual revenue and it's been running for just over a year.

The part most developers miss is that there's a whole layer underneath the prompts.

You can write instruction files that load before every session. You can build slash commands that inject real data — actual git diffs, real file lists — into the prompt before Claude sees it. You can create agents with their own tool access and model preferences. You can write hooks that fire automatically when Claude tries to do certain things, like blocking a commit if it contains a credential file.

That last one is deterministic. It doesn't rely on Claude getting it right. It just blocks it every time.

There's also a new feature called Channels where your Claude Code session connects to Telegram or Discord. You message your bot from your phone. Claude responds using your full codebase, your full dev environment, everything. Your terminal session just sits there, ready, any time of day.

One person told me they refactored an auth flow from their phone while grabbing coffee and came back to a pull request ready for review.

There's one more thing I want to flag because it's not getting enough attention.

Claude can now scan your entire codebase for security vulnerabilities.

Normal security scanners match patterns. They flag everything that looks suspicious and they're wrong between 30 and 60 percent of the time. Most of the time you're triaging noise.

Claude actually reads the code. It traces data flows across files. It looks for logic problems that cross file boundaries.

Anthropic says the false positive rate is under 5 percent.

In testing, their security team found over 500 vulnerabilities in well-reviewed open source projects. Bugs that had gone unnoticed for years.

If you have a codebase, pointing this at it is probably worth doing this week.

Quick note on the business side because it tells you something about where this is going.

Anthropic just raised 30 billion dollars at a 380 billion dollar valuation. Eight of the Fortune 10 are customers. Run-rate revenue is at 14 billion, growing tenfold annually for three years running.

And their own engineers now use Claude for 60 percent of their work. Up from 28 percent a year ago.

The tools are building the tools. That loop is what explains the pace. And the pace is not slowing down.

So here's what I'd actually do if I were you.

This week, build the three context files. It takes 45 minutes and the difference in output quality is immediate.

Connect at least one tool you already use. Google Drive, Gmail, Slack, whichever one would save you the most time.

Set up one scheduled task for something that eats 20 minutes of your morning.

That's it. Start there.

The gap between people who have done this and people who haven't is already real. Three months from now it'll be harder to close.

Reply and tell me which part of this you want to go deeper on.

Cowork setup.

Claude Code.

The security scanner.

The phone to desktop stuff.

Whatever you want, I'll cover it next.

— Prompt Guy