Hey, it’s Hasen

Three months ago I upgraded to Claude Pro.

I've used it every day since.

Okay, the "productivity glow-up" is a bit of a myth.

Most days that first month, I got less done than before I started.

The problem wasn't Claude, it was me.

I didn't know Claude isn't one tool. It's a few different tools stitched under one name. Each one touches the internet in its own unique way.

Once that clicked, the whole frustrating month made sense.

Two Failures, Same Tool

Two moments stuck with me.

The first: I asked Claude Chat to list every upcoming crafts market in the Bay of Plenty area, now through the end of September.

The list looked solid.

Until I checked it. Half the markets had already happened. Some, years ago.

Chat wasn't making anything up. It just couldn't tell the difference between information that exists and information that's still true.

The second: I wanted a summary pricing table for all bracelets from a members-only supplier site. I'm already signed in. I asked Chat to grab it.

It couldn't touch it.

Same tool. Two different failures.

And that's when I started wondering if Claude was broken.

Here's what I didn't understand.

Claude Chat and Claude Cowork don't have the same access to the internet.

That gap explains both failures.

Claude Chat can look things up. It can't open anything.

When Chat searches the web, it pulls back snippets. Small chunks of text from a handful of pages, based on what looks relevant to your question.

Give it a direct link, and it can go a step further. That's called a "web fetch." It grabs the full text of that page.

Better than a snippet, but still just a copy.

Chat cannot click, scroll, or open a second page to double-check something.

Both of these fall under a term you'll see everywhere in AI writing: RAG, short for retrieval-augmented generation.

All it means: Claude looks something up first, then answers. Instead of relying only on what it already knew.

When I asked Claude to list upcoming crafts markets, it pulled listing pages through search. But it had no way to click into a listing. No way to check an archive page for a post date.

Chat didn't get the date wrong.

It never had a way to go looking for one.

Claude Cowork can actually operate a browser. Through a separate tool called Claude in Chrome.

This is the part I didn't know existed.

Cowork's browser ability doesn't come built in. It runs through a browser extension, Claude in Chrome, which you install and connect before Cowork can touch a live webpage.

Once it's connected, Claude can click buttons. Scroll a page. Move between pages. The same actions you'd take yourself.

One important note: Claude in Chrome doesn't log into anything for you.

It won't type a password. It won't handle a login screen.

What it does is work inside sessions you're already logged into. Signed into a site in your browser? Claude can act on that site through the extension. 

That's why Cowork was able to pull the pricing data and turn it into a summary table and Chat couldn't. It worked inside a session that was already open. Scrolling through the pricing page the way I would have done manually.

The practical split:

Quick fact lookup — Chat: good fit. Cowork: excessive.
e.g. courier delivery zones

Data behind a login — Chat: can't access. Cowork: works inside your session.
e.g. supplier portal

Multi-step navigation — Chat: can't click through. Cowork: can navigate page to page.
e.g. competitor price research

Filling out a form — Chat: no. Cowork: yes.
e.g. stockist application

Known article summary — Chat: good fit. Cowork: excessive.
e.g. report you already have the link to

Once I understood the gap, I stopped expecting Chat to do Cowork's job.

The takeaway:

Chat reads the web. Cowork operates it.

Need an answer? Use Chat.

Need Claude to click, scroll, or work inside a site you're already logged into? That's Cowork's job.

Next time you're about to ask Claude to dig something up, pause and think.

Do I need an answer, or do I need Claude to go do something on that page?

That single question saves you the kind of month I had.

One more thing, before you go.

You don't need Cowork for this one.

When the publish date actually matters, don't just ask Chat to search. Ask it to check.

A plain search usually skips the date entirely. Push Chat a step further. Have it fetch the full page for each result and scan the text for a publish or last-updated date.

Prompt template:

Search the web for [your topic]. For each result, fetch the full page and check for a publish date or last-updated date. List each result with: the date, the link, and a one-line summary. If you can't find a date, say so explicitly instead of leaving it blank.

Two honest caveats. This only works if the date is sitting in the page's visible text, Claude still can't dig through an archive to find a hidden one. And it's slower than a normal search, since Claude is fetching multiple full pages.

Save it for when the date is actually load-bearing.

Talk soon,

Hasen 

Infinite Output — Practical AI for solo founders.

Coming Up:

The mistake that cost me even more time. I had the same instructions pasted into three different places in Claude, thinking that made it more consistent. It didn't. Global settings, Project instructions, Cowork instructions do three different jobs. 

Keep Reading