What a Difference a Model Makes

Author

Jeff Sauer
October 17, 2026

As this lands in your inbox, I'm wine tasting in Napa.

I'm at Chateau Montelena for the 50th anniversary of the Judgment of Paris, celebrating my wife's birthday and some anniversaries with best friends and family.

I got my work done before I left, so I didn't have to open a laptop in a hotel room to write this note like I have done dozens of times before. That felt really good.

A month ago, I wasn't sure that was possible.

Here's my honest recap of this year with AI.

In January, I made myself spend a weekend learning Claude Code. It took about 30 minutes to see the light once I finally dug in.

Typing into the terminal on VS Code with a black background and white text? It felt like seeing the Matrix. You know the scene where the green code turns into the world Neo actually sees.

Of course, things have come a long way since then. Earlier this year, Claude's models had a 200,000-token context limit (or less), so you had to keep compacting your work in order to keep working on a project.

Then million-token context arrived, and it was genuinely great.

The models kept changing. Some were good; others made me question my life choices.

Opus 4.6 was great. Opus 4.7 was bad. Opus 4.8 was rough at first, then got good. In my experience, Opus 5.0 was the worst AI model ever invented.

August into early September was a dark stretch. After a week or two, I moved back to Opus 4.8 as my default model. But even that was still causing issues.

At home was chaos: my son hadn’t started school yet. My wife was dealing with family out of town.

I tried moving my workflows to ChatGPT but couldn't stick with it. I couldn't get an email written or a YouTube script in less than 4 hours.

These tasks normally take me 30-60 minutes to dictate and edit.

I was apologizing to my team every week for being the bottleneck.

Then Opus 5.5 arrived, and now I'm faster than ever. To the point where I would be fine if this model were the last model that Anthropic releases.

I hope they don’t pull the rug out.

The consensus on many podcasts I follow is that we may have turned a corner on bad model releases.

Maybe they're right?

With AI, it’s always two steps forward, one step back. I've written a few newsletters saying exactly that this year.

But I still believe in the dream of doing your work faster so you can spend the time on what actually matters: the people and the projects you'd otherwise miss.

Today, I'm living that while sipping wine and eating BBQ with some of my favorite people.

Until next Saturday,
Jeff Sauer

P.S. If you want help bringing AI into your organization, or getting better at it yourself, I do 1:1 coaching for exactly that. Just reply, and we can see if it’s a good fit.

New On YouTube