Hey Tal. I'm going through the same process of exporting my transcripts. Hopefully it will be a straightforward process 😅 I was wondering, which alternative are you choosing to replace Granola? Thanks
Currently I started paying for Grain because it has basically the same features as Granola, but also has the bot that can join meetings and that allows me to have very specific speaker names when it's more than two people on a call and much more accurate metadata.
A few of my friends have been using it for a long time so that gave me a lot of confidence as well. (Once Granola started charging, there was nothing to lose by switching.)
Hi does this paywall going to impact the trick you covered recently regarding reading transcript from local drive instead of using MCP . Is there a difference?
Hi Omar - I can't speak for Granola's pricing policies, but it sounds like you're asking if they can paywall files that are already on my hard drive, and I doubt that is something feasible to paywall. Let me know if I'm understanding your question correctly.
Very interesting - Thanks for sharing - I will also try this myself!
Re strategies taken by the different agents, the ‘sub agent’ approach does seem to me to be a characteristically Claude-like way of doing things. Do you think this is less lossy than the compaction approach? I would have preferred it in general for this reason but admittedly I haven’t myself inspected the compaction outputs to make that judgement
I definitely did not go deep enough or inspect anything either! Speaking purely theoretically, I can see how both can have advantages - and neither is obviously better. (like you said though, it comes down to inspecting, and also doing this many times in each)
I see "compaction" as kind of like when I want to achieve something by retrying the same cursor thread over and over again, and I just go back to the last prompt and change it.
That's kind of the same thing here, and it's kind of the same thing as spawning sub-agents with everything that preceded that moment in time as the prompt. And like the Anthropic experiment of having Claude play Settlers of Catan for 75 minutes (link above), the compaction can be done very intelligently.
If one has both Claude code and cursor, should we use one vs another or are there opportunities to integrate both and use?
Hey Tal. I'm going through the same process of exporting my transcripts. Hopefully it will be a straightforward process 😅 I was wondering, which alternative are you choosing to replace Granola? Thanks
Currently I started paying for Grain because it has basically the same features as Granola, but also has the bot that can join meetings and that allows me to have very specific speaker names when it's more than two people on a call and much more accurate metadata.
A few of my friends have been using it for a long time so that gave me a lot of confidence as well. (Once Granola started charging, there was nothing to lose by switching.)
Hi does this paywall going to impact the trick you covered recently regarding reading transcript from local drive instead of using MCP . Is there a difference?
Hi Omar - I can't speak for Granola's pricing policies, but it sounds like you're asking if they can paywall files that are already on my hard drive, and I doubt that is something feasible to paywall. Let me know if I'm understanding your question correctly.
Very interesting - Thanks for sharing - I will also try this myself!
Re strategies taken by the different agents, the ‘sub agent’ approach does seem to me to be a characteristically Claude-like way of doing things. Do you think this is less lossy than the compaction approach? I would have preferred it in general for this reason but admittedly I haven’t myself inspected the compaction outputs to make that judgement
I definitely did not go deep enough or inspect anything either! Speaking purely theoretically, I can see how both can have advantages - and neither is obviously better. (like you said though, it comes down to inspecting, and also doing this many times in each)
I see "compaction" as kind of like when I want to achieve something by retrying the same cursor thread over and over again, and I just go back to the last prompt and change it.
That's kind of the same thing here, and it's kind of the same thing as spawning sub-agents with everything that preceded that moment in time as the prompt. And like the Anthropic experiment of having Claude play Settlers of Catan for 75 minutes (link above), the compaction can be done very intelligently.