It's July 8th, and I asked Claude to do something I'd normally do myself: pull together a rough plan for the week, sort my messy notes into it, and flag what was missing.

I expected the usual — a nice answer I'd then have to copy, paste, and clean up. That's the deal I've had with AI for two years. It talks; I do the work.

This time it just… went and did the whole thing. Step by step. Checked its own work along the way.

I actually sat back in my chair.

That's the moment the word "chatbot" stopped fitting.

Here's what nobody tells you: the tools you already use are quietly turning into something new, and most beginners are about to get blindsided by it inside apps they thought they understood. So before that happens to you, let me save you the confusion I had to work through myself.

There are really only three stages of AI, and they build on each other like rungs on a ladder:

A chatbot answers. An assistant helps you do a task. An agent goes and does the task for you — multiple steps, on its own, then reports back.

We spent 2025 learning to talk to chatbots. 2026 is the year they start doing our errands.

1. Chatbot — it answers your question. You ask, it responds. That's the whole relationship. "What's a good subject line?" → here's one. This is where almost everyone started, and it's still genuinely useful. Nothing wrong with living here.

2. Assistant — it helps you finish one task. Now it's doing part of the work with you: drafting the email, rewriting the paragraph, summarizing the document you pasted in. Still one step at a time, still you in the driver's seat — but it's hands-on, not just talking.

3. Agent — it does the whole job, start to finish. This is the new rung. You give it a goal, and it breaks that goal into steps, does each one, checks itself, and comes back with something finished — not advice about how to do it.

When I asked Claude to build and fill in my weekly plan, that was an agent. It didn't hand me instructions. It handed me the plan.

You don't have to go looking for agents. They're arriving inside tools you already have.

Same pattern everywhere: less "here's how you could do it," more "here, I did it."

And this isn't a niche thing coming someday. Industry analysts expect 40% of business apps to have a task-doing agent built in by the end of this year — up from under 5% a year ago. "Knowing how to work with AI" is quietly becoming the new "knowing how to use Microsoft Office."

I'll be honest — the word "agent" made me nervous at first. It sounds like handing over the keys.

It isn't. Here's the rule I use, and it's kept me comfortable:

Let it do the steps. You keep the judgment.

I let Claude run the multi-step busywork — sorting, drafting, first passes, "organize this mess." But anything with real stakes — what actually goes out the door, a decision that matters, money, anything I'd be embarrassed to get wrong — I still read every line myself before it's final.

An agent is an eager assistant, not a replacement for your own eyes. Give it the legwork. Keep the last look.

The 8-hour workday is becoming a 4-hour workday for people who know how to use AI.

Everyone else is still catching up.

This AI work playbook shows you exactly how to cut your work hours in half using AI.

Sign up for Superhuman AI and get:

Claim your AI playbook

If reading this made you itch to know what's actually happening under the hood — how the thing running your errands even works — that's a whole rabbit hole, and I didn't want to cram it into one email.

So I built a character to walk you through it, no jargon, no assumptions. Over on Tilly's AI Tidbits, the brand-new Behind the Counter track takes AI apart one plain-language piece at a time — what's really going on when it reads you, remembers you, or gets confidently wrong. No prior episodes required; you can start anywhere. Part 1 is live now if you want the very first building block — "It's Not Reading — It's Recognizing." Watch on YouTube or read the newsletter.

If all of this feels like a lot — you don't have to become an "agent power user" this week. You really don't.

The one thing worth doing: next time your AI offers to do something instead of just tell you about it, say yes to a small, low-stakes task and watch how it works. Ask Claude to sort a pile of notes into a rough outline. That's it. One little errand.

That single "oh, it just… did that" moment will teach you more than any explainer I could write.

You've got this — and you're earlier than you think, not later.

Spoiler: what I write next comes straight from your replies — so the question you're sitting on right now might be the one I answer in the next issue.

Have you had your own "wait, it just did that" moment yet? Hit reply and tell me what your AI surprised you with — I'll share a few, and let them steer where we go next.

Anna