Their target is not coders, it is the professional world who do 90% of their work in Office applications, like me. A $200/m model absolutely does not fly when rolled out to entire corporations. It needs to be a $20/user/month product.
But I agree, it sucks. It is the only AI we are able to use at work and for tasks that it should be good at (compare comment sheets against a deliverable register and assign to specific packages) and it just can’t do it. It can read the spreadsheet and understand them just fine but outputs are garbled nonsense.
[dead]
Copilot is actually significantly more reliable at technical tasks with SQL or C# than others, i've found. Do we have different use cases?
Copilot seems to hit the technical level I'm asking about much more reliably. It keeps a more grounded general semantic model.