Comment by nitwit005
8 hours ago
> Project context signals (repo/PRs? issues? tasks? calendar? a “project doc”?)
There's a cost to recording what you're working on, so usually the only people who track it in a fine grained way are those that need exact numbers for billing. It's not worth the time otherwise.
There are hints to what people are working on. Connecting to a database means SQL may happen, but maybe not.
It's a big issue with personal assistant ideas in general. It's very difficult to get any real context on things. Even data that seems firm like calendar appointments, isn't in practice. Look at people's calendars, and you'll see them triple booked.
You're totally right — fine-grained “what I’m working on” tracking is usually not worth the overhead unless it’s tied to billing or some hard accountability.
One direction I’m exploring is to stop treating context as a single ground-truth stream and instead use cheap, probabilistic hints: a small “active projects” list (explicit but constrained), plus weak signals (recent files, open tabs, issue activity), and then ask for confirmation only when confidence is low. Calendars are a great example of why “seems firm” isn’t actually firm.
If you were designing this, which would you rather tolerate: (a) fewer suggestions but higher precision, or (b) more coverage with lightweight confirmation prompts?
And yes — this is exactly the problem I’m outlining in my HN bio/profile if you want the longer version.