Looking Busy Is Not the Same Thing as Thinking

Every week there is a new best AI model but this week will represent a significant increase in the intensity of the new shiny thing effect.

Today is OpenAI DevDay - OpenAI will release its own always on agent as its answer to Grok/Muse style agents.

Plus, we anticipate multiplayer Codex and an Agents API, among other things.

The Hype Train

In the ether that surrounds the AI hype train, people move their workflows to the new best thing, declare the last one dead and then do it again three weeks later. I’m sure you have noticed if you’re playing this game. If we sift through the sand of the more recent releases to try to find the truly most useful, I think the gold that shakes out is Claude’s Opus 5.5.

It is clearly cheaper, faster and less irritating than previous Opus releases, plus it can keep up with Fable at a ridiculously awesome fraction of the cost.

Previous recent iterations of Opus were so annoying and sloppy that many cancelled their Claude plans to switch to Codex.

Between the expense involved with using Anthropic products, plus how much of a slog it was becoming to deal with Opus, this reaction makes sense. Plus, I think having loyalty to any of these companies is foolish.

Motion Without Quality of Thought

Codex felt like the answer to Claude because if you gave it a job, it just got moving (and it didn’t cost you half your week’s usage quota). When I took a closer look at what it was actually doing though, I realized that Codex was getting really good at looking very busy but didn’t seem to actually be reasoning throughout long running tasks. Sure, it did get things done which I appreciated, but it seemed to like to take very circuitous routes, a lot of which seemed unnecessary.

Sometimes I got a lot of motion without meaningful quality of thought, which ended up being harmful to the project.

As the model companies keep optimizing what users notice first: speed, tone, smoothness and the feeling that work is happening, it seems obvious that the name of the game is how to keep systems stable while the models churn underneath that.

There have been so many (justified) complaints about the endless hype train we seem to be on. I am personally profoundly thankful for this technology and also wondering how participating in that hype train is impacting our agency as this AI transition progresses.

Frictionless technology is not necessarily agency-preserving technology. And looking busy is not the same thing as thinking.