Since my last series of AI process posts, the much hyped Claude CoWork finally came to Windows. I was curious about it, so I upgraded to a Claude Pro account. I am enjoying the increased chatbot usage that comes with a Pro account. Claude, possibly, is not. Periodically it starts gently hinting that we are DONE with a particular subtopic and need to MOVE ON. As for CoWork itself, the Windows version is very glitchy. I’ve only gotten a modest amount of useful work out of it. Specifically, I wanted to go back and make clearer POV shift indicators for some of my past novels. I put the manuscripts in a subfolder of the only folder on the computer that CoWork has access to. Then I asked CoWork to review each manuscript, and write a report which quotes the first sentence after each POV shift. Here’s a couple observations:

-Opus 4.6 glitches out less often than Sonnet 4.6 does in Cowork mode, but lives up to its reputation as a usage hog. Reviewing one 50K-ish word manuscript and writing the report on it used up most of a Pro Plan usage session. I ended up running these tasks at times when my session limits had reset, and I did not expect to need Claude for a while. Late at night, early in the morning, times like that.

-Even Opus 4.6 glitches out in CoWork when asked to write fanfic. This might be a guardrail related to Anthropic’s predominantly corporate focus for CoWork. Or maybe I’m not giving it clear enough instructions. As you can see from the past “Fanficcing with Claude” posts, I tend to be pretty general in my fanficcing instructions.

-My initial feeling is that CoWork is going to be most useful for tasks too large for a chatbot and too specific to justify building a whole automation, but we’ll see how it goes.