In the first Chad vs Gooey piece, I wrote about being surprised that I preferred the chat/agent path even for tasks that seemed to be faster with human GUI.

Not two days later, I experienced a full-blown "holy crap agent chat is better" moment.

One of my side tasks at our firm is that I needed to publish a bilingual newsletter in MailChimp, on a monthly basis.

The old way: Once the content is ready, open MailChimp's drag-and-drop editor, click into each content block, type or paste text, use dropdown menus for formatting, drag elements around, preview, adjust. Takes about an hour to assemble, another 30-60 minutes to review. That means I typically need to set aside a good 2-3 hour block to do this.

This was never a problem per se, since I was used to it.

But once I tried the agentic way, I was instantly convinced that I'd never go back.

First, I sent past newsletters to Claude Code. It analyzed the previous newsletter's HTML structure (header, dark summary box, story blocks, footer) and we had a 2-minute conversation on what stays constant and what changes in future editions, and then Claude Code generated a complete new HTML file as template, with placeholders for the new content slotted in.

Once I had the content ready, I simply fed this to Claude Code. But I made a mistake: I tried to do this about 15 minutes before my daughter's school bus would arrive. I was sure I wouldn't be able to complete it. I was wrong.

Claude Code generated the html code for the newsletter instantly. I pasted the raw HTML into MailChimp's code editor, and it was done.

3 minutes.

(In fact more time was spent looking for where to drop the code in the GUI!)

The GUI wasn't slow because of bad design. MailChimp's editor is okay (though it does feel legacy). It was slow because it's a visual interface for assembling something that has an underlying representation — HTML — that an agent can write directly.

The GUI is a translation layer between you and the output format. If your agent can speak the output format natively, the GUI becomes unnecessary overhead.

This is going to happen to a lot of SaaS tools.

In the extreme end, any product whose GUI is essentially a form-filler or layout tool on top of a structured format is at risk:

The pattern: if the GUI exists because humans can't write the underlying format, and agents can write that format, the GUI is overhead.

I'm not convinced that everything dies. GUIs are still better in some cases.

One is spatial/visual feedback in real time — design tools, video editing. The GUI survives here because the fine-grain of the output matters and direct manipulation is faster than describing what you want.

Another is where discovery matters — you don't know what you want until you see the options, and GUI is still better for laying out options.

Also, if the underlying format is proprietary and undocumented, the agent has no target to write to.

But other than that? The GUI was built for humans who can't write code. Agents can.