I have read summaries of emails and meetings that had the action items all sorts of wrong, sometimes completely inverted.

It seems to me that if an email or meeting is at all important, the stakes are too high to trust the summary, and if it is not important, neither is its summary.

Add on to that the fact that locally running LLMs are even more scatterbrained, I don't see how this fills even the limited need you're describing in any useful way.

So, they spent their limited available manpower on an unrequested feature, and to add insult to injury, the feature is unlikely to have effective practical uses. It might be capable of more limited scope text prediction like code auto-complete, but the field is already flooded with those. I think the Thunderbird users have far more use from improvements to Thunderbird than they do for other unrelated products.

Could you explain what email summaries have to do with this announcement? Thunderbird and Thunderbolt are separate applications. Thunderbolt doesn't include AI models, it is merely a frontend to an AI API of choice, similar to how Thunderbird is a frontend to an email server of choice.