And it can do your whole job now.

My X timeline over the holidays was filled with two things: Claude Code and Claude Co Work. If you haven’t seen them, Claude Code is like having a personal 10-year veteran engineer who can write code, build websites, and turn your ideas into working products. In the past week, I’ve built a crypto wallet for Tempo, a stablecoin dashboard, and a custom AI model trained on Brainfood (it's still rough, but watch this space).

The problem with Claude Code is the problem with most LLMs.

You have to type to it. It’s not a graphical interface.

Enter Claude Co Work.

It can read your files, move them around your computer, use applications, and do entire tasks. Like: take the summary notes from this meeting, send a thank you note and follow-ups to the participants, then offer some slots for the follow up.

You now have a personal engineer and a personal assistant.

Claude is cooking.

For most people, using Microsoft Co-Pilot or ChatGPT is occasionally useful and often frustrating.

You’ve probably experienced the excruciating pain of trying to get ChatGPT to do that thing you want, but it just doesn’t get it and runs off in a completely weird direction. Some tasks, for some people, some of the time, are handled at a world-class level. Better than any doctor, software engineer, or lawyer. But most of the time, for most people, it isn’t.

That’s the jagged frontier. The boundary between “AI can do this brilliantly” and “AI will waste your afternoon” is unpredictable and constantly shifting.

But software engineering just crossed that frontier entirely.

100% of Nvidia software engineers use AI for coding, and they’re shipping on average 3x faster with higher quality.

Why did software cross first? Two reasons:

My timeline is filled with engineers confessing their undying love to Claude Code. They’re not just generating bits of code. The agent can read your setup, identify config issues, and fix them. Let’s say you just got a new Mac and forgot to install Homebrew. The agent sees an error, and asks “It looks like you’re missing Homebrew. Would you like me to install it?”, then executes the installation and configures the path variables.

Claude Code doesn’t stop until it works. You can set it to auto-install missing dependencies. Like trusting a junior engineer to handle that stuff. (Claude — dangerously-skip-permissions is the new YOLO).

This is the shift from “just works” to “just does.”

The difference is autonomy.

 The AI watches it, writes the summary, emails the attendees, and schedules the follow-up.

2025 was the year “just works” became normal for software engineering. 2026 is when “just does” reaches everyone else.

Claude Co Work is the moment the agent leaves the terminal and enters your Documents folder.

What would you do if you could just delegate it?

This is what moves AI from a tool you use to a colleague you trust.

If your team primarily uses ChatGPT to draft basic emails, summarize documents, or write boilerplate marketing copy, they’re operating at a fraction of what’s now possible.

The shift is from “how do I use AI to help with this text?” to “how do I use AI to do entire tasks?”

Your team’s new core competency isn’t using the tool. It’s orchestrating agents. The most valuable employees will be those who can conceptualize a complex, multi-step task (like “launch a new card rewards program”) and delegate execution to a small army of autonomous AI agents.

This is a management and strategy skill, not an execution skill.

If execution becomes cheap and fast, the constraint on your company’s growth shifts. It’s no longer engineering resources or operational headcount. It’s the quality of the concepts you feed your agents.

You have a team of digital colleagues now. They don’t sleep, they don’t complain, and they scale infinitely.

But they only scale your vision.

Claude Co Work isn’t generally available yet. But if you want to get ready for it, the best thing you can do is use its sister, Claude Code to do things.

I’m willing to bet a decent percentage of you haven’t vibecoded anything yet. And that’s ok. CLIs are confusing, IDEs are confusing, and config is a nightmare. Just setting it up can take ages. But Claude Code is available right there in the desktop app.

Here’s how to start:

Then ask yourself: What’s the biggest, most audacious thing I would ask my company to do if execution were instantaneous and free?

That’s your new job.

Now go write the prompt.

ST.