Blog

Listening to a review

Late last night I asked an AI to review Latte the way a tech reviewer would: use it, test the meeting features, and score it. It gave Latte 7.5 out of 10 as an early build. Most of today went into the things it pointed out.

Where it was right

The best catch was about the AI in Latte. When it wrote up meeting notes, it sometimes turned a plain fact into a follow-up the notes never mentioned. Briefings could also assume last week’s action items had been done. Now the write-up lists only follow-ups your notes actually state, and a briefing asks whether earlier action items happened instead of assuming.

Briefings also show where each item came from: “From your notes”, “From a file”, “From the web” or “Latte’s guess”. If something is wrong, you can click “Wrong?”, and later briefings with the same people won’t repeat it.

It also said three hours of meeting listening a month was too little for people with a lot of meetings. Pro now includes ten hours, and there’s a 7-day free trial for anyone trying Pro for the first time.

A calmer start

The reviewer called the fireworks and page flip charming the first time and noise by the tenth. That’s fair. New users now start with them off, and anyone can turn them on in Settings. The dog stays. On the first day, the tour is the only thing that pops up, and after that there’s at most one offer a day.

A few smaller fixes came from the same review. The task bar fits on one line, all-day items in Month view are thin bands so they don’t look like meetings, and the Add Task buttons are now brown so they’re easy to tell apart from Add Event.

Meeting notes that are easier to read

In a meeting’s details, the briefing and your notes now sit in two cards with different colors, labeled “Before the meeting” and “During and after the meeting”. Once a meeting is over, your notes come first and the briefing folds down to one line.

When Latte listens to a call, it can tell your microphone from the other side, but it can’t tell voices apart. So now you can tap “Them” next to a line and pick who was speaking.

What it didn’t like that I can’t fix today

Some of what it asked for only comes with time: proof that the meeting features hold up on lots of real calls, and a record of not losing anyone’s data. Listening to a call in another tab still needs Chrome or Edge on a computer. I’d rather name them than pretend they’re done.

So far, as of October 27: about 65 hours of hands-on time over 29 days, and about $109 spent, not counting my Claude subscription ($200 per month), which I also use for other things.

Claude and Claude Code are made by Anthropic. Latte isn’t affiliated with or endorsed by Anthropic.

Comments

Loading comments…