Become a Creator today!Start creating today - Share your story with the world!
Start for free
00:00:00
00:00:01
Inside Cursor (Part 2)
4 Plays9 days ago
Episode 2
Mohit gives us a peek inside Cursor and AI tools. What he uses, Where Cursor shines and where it still fails and some really interesting ways people are using Cursor.
The Root Cause - A podcast by RubixKube
Learn More
Recommended
Transcript
Unexpected User Interactions
00:00:01
Priyank Upadhyay
Also, Mohit, I wanted to segue towards what do you see in users when you're interacting with them? ask of i'll frame a question, and then I think you'll get the theme what I wanted to ask. It is, how does a user interact with a cursor that That would surprise the public. but What you're seeing, something different. Because so typically, you know it happens most often is you write a doc and there is certain way you you have built tools to be used.
00:00:41
Priyank Upadhyay
But you see people using it in a different way, and that grooms the product ultimately. But what's something that you saw recently in your conversations that surprised you and will surprise
Root Cause of User Issues
00:00:55
Priyank Upadhyay
my audience?
00:00:55
Mohit
Yeah. Yeah. I think like one great thing about being working with all these user queries and everything is like at the end, of course, you can blame the user for not checking the documentation. Or you can blame the user, hey, how did you not find this toggle button here? right But at the same time, I think every single query we receive, at least we try to like think of it as, like is it a documentation problem? Is it a problem of like product? Is it a problem of onboarding? Is it a problem of like just social media awareness? right like it it ideally will fall into one of these buckets, right? Like now you can't blame the user, right? Like it's not that user works in your company and has contact stories on Twitter all the time or just needs to know everything. So at the end, I think everything has to be like sort of intuitive in that sense in some way or the other. It could be an awareness problem or like just product onboarding problem or product problem, right?
Adapting to Platform Changes
00:01:45
Mohit
So once we like graduate from that thinking, I think, Personally, what I have noticed, like every time there is like a big change in the whole platform itself, it becomes like, oh, like recently we went from like just that IDE window to like an agent window, right? And initially it was a big change for users and a lot of like queries regarding that, that, oh, why did this happen? And then slowly users get acquired to it once you explain to them and enough that what are the advantages of this particular out interface, right?
00:02:13
Mohit
So I think for me, it's always interesting to see that gap in a way, which is like, And you know it comes from a sense of like, understand that, okay, again, I have this much context, not everyone has, right?
Misunderstanding Language Models
00:02:25
Mohit
So I think I have found like a lot of those things where between Tech Bros Circle and Indira Nagar, they are all probably talking about like, yeah, this is the automation you should be doing, Cloud Agent.
00:02:36
Mohit
But there's a guy out there who is like, hey, Agent is saying that it's sonnet when it's actually opus right then you have to make that user understand hey it's because like sometimes the lm's don't know their own identity they could be wrong right like they work in a different it's not deterministic they probably are not as self-aware as you would like to think so it might be still opus but it might say sonnet because it was trained on a certain data etc right like so i think that for me has been very like interesting to see like what the I wouldn't want to call it a gap. It's just like catching up, right? Like in a way that now they have to understand that this is how LLMs function. And we are still super early in how people understand these AI models, right? Like there's this side of people who might think that it knows everything about the world and it has answered to all key questions. But also there's a other side that LLMs don't know everything. They are...
00:03:27
Mohit
capable of like making mistakes, they are just as, you know, like it's not deterministic, like the right word to say. For me, like I think that has been interesting just to see that sort of like gap,
Intelligence Layer Surprises
00:03:38
Mohit
right? And when we talk about like a bit more advanced and experienced users, even then I have been seeing now like like People who have been working in the industry for a long time might be shipping a tool or like a feature which is incredibly complex and all of those.
00:03:55
Mohit
like It's just too technical here we and I could be probably reaching out to them saying, hey, but have you tried like a cloud agent? Have you tried like switching from an API to a cloud agent or like some sort of intelligence intelligence layers between API things which you are doing?
00:04:09
Mohit
And that somehow magically changes their output that, oh, wow, we did not know this was possible, right? Like to give small example, like recently I was interacting with someone and they had something to do with like, hey, our linear says this is gone from in progress to done.
00:04:24
Mohit
And then we, you know, let others know that this is done. But sometimes So linear can do that much and i'm pretty sure there's like linear AI tools out there not to like talk on behalf of them but at the same time now with cloud agents and cursor automations there is an intelligence layer you can put in between them that once a linear status changes then you put a prompt saying that see which PR got merged, check your code base, did it actually fix the issue it was like raised in the it was raising in the first place right like so That intelligence layer has come in now.
00:04:56
Mohit
For me personally, like I have been using it with my customers all the time. When I check linear tickets, I'm like, okay, did this actually solve this user's problem or i am I just letting them know this is fixed without it actually unblocking them, right? Or is there like an additional step I have to do?
00:05:10
Mohit
So for me, I think that's interesting that there have been just this new sort of a perspective to how you can solve a problem, right? Like by adding that intelligence layer to your existing API calls, existing dashboards, and kind of just making it so much better. so i think For the advanced users, that's what I've been seeing. They get incredibly surprised when we say something like, hey, we get that you have a dashboard and then you do a manual review. Have you tried also just putting an LLM triage into it? like like Give them one more layer of like intelligence and then you make a manual call. right like
00:05:41
Mohit
Again, human in the loop is always in important and like there are some very critical pieces where it should be the default too. But I think it just has enabled like one more layer of intelligence for a lot of these people who were like, okay, I'll make the automation and then I will still be the one doing the intelligence part of it.
00:05:58
Mohit
But I think at this point, we can also like sort of outsource some level of intelligence to these LLM tools, which has been super interesting to me.
Coping with Fast-Paced Releases
00:06:06
Priyank Upadhyay
One of the things to blame on Cursor is they are shipping so fast that it becomes difficult for developers or engineers to keep up with that. So every week if you are releasing a new feature, we need time to adapt to some things.
00:06:25
Mohit
oh, we take a week to release a feature? That's, so I think we are doing it much faster than that. I'm pretty sure there are like at least three features being pushed out in a week. But I get i get that, right? Like definitely it is tough to keep up and it's tough to catch up.
00:06:40
Mohit
bit of a segue, but also like what I like to relate to back in the time around 2019, before this whole AI thing happened, I read a book by Yuval Noah Harari, which was 21 problems for the 21st century. And in that, one of the key themes this author kept talking about was like how in our generation, we need to keep on adapting ourselves again and again and again. right like there was a few There was a past where my dad can work in the same bank for 20 years in the same role and still be happy and everything is good. right
00:07:14
Mohit
Unfortunately, I don't think we are in that world anymore. right like It's set at the end of the day. It is on us to like kind of adapt ourselves and Just keep up with all of these. And again, I think there needs it's like news, right?
00:07:26
Mohit
like There's like 500 news articles every day, but are you going to concern yourself with 500 news articles every day? No, you are probably going to look at like the 10, which actually matters to you about your local municipality, about your stuff, right?
00:07:36
Priyank Upadhyay
Out of those 500, 400 could be AI generated. So filtering the signals from noise also is getting difficult.
00:07:42
Mohit
Yeah.
00:07:46
Mohit
It is getting difficult, but I think that is the important skill. right like Now, those are the kind of skills we need to adapt in this career. that okay how If I see 20 new fundraising announcements every day from a different AI company, am I going to test all of these tools out? Probably not. right like Still take us take a step back. like Even in the context of Cursor, right as you said, we probably ship like multiple features in a week, but I think it's always important, like okay does this actually something which is directly impacting me now?
00:08:14
Mohit
do Should I really fully immerse myself in it? And if the answer is a straight up no, then it's fine. Like you don't have to bother yourself that, right? Like update, go go ahead with your regular workflows, but just good to keep in mind.
00:08:24
Priyank Upadhyay
and of cool I often correlate it with one of the like things we used to do earlier.
00:08:25
Mohit
Yeah.
00:08:34
Priyank Upadhyay
Let's say I had to download a game from a sketchy website. Finding that download button just based on my judgment was something we used to tackle.
00:08:45
Priyank Upadhyay
I think this is what will come in handy when we have to identify a signal from a lot of noise.
00:08:50
Mohit
Yeah, absolutely. Like I, I think I relate there, like you find seven different download links and it's your gut, which tells, okay, this is less shinier. This is probably the inherent website one. Right. So yeah, I relate that. Like it has to be that, that you have to kind of learn how to separate that noise from the signal, as you rightly said.
00:09:08
Priyank Upadhyay
Makes sense.
Creative Uses of Cursor
00:09:09
Priyank Upadhyay
What's the dumbest or weirdest thing you saw people create with Cursor?
00:09:15
Mohit
yeah I don't think anything is dumb. but I think everyone is just finding this as a way to like explore their sort of creativity, right? Like it's like an inherent sort of bias we have that we want value out of every software we want like a lot of net positive but i think personally i find it okay to like if someone has created a website which has you know you type prompt and it will generate a flower right like and you can maintain that garden i have seen that project and i find it interesting that okay probably this does not solve anything in real life but it's like if you give me a canvas and a paint and if i paint something it can either be a van go or nothing right like so for me i think
00:09:53
Mohit
I would refrain from calling anything dumb because i I think it's just a sort of like medium of self-expression at this point, right? However, like there have been like some very funny ones.
00:10:03
Mohit
Like, you won't believe I was talking to a user and all they wanted was like every time Cursor complete its chat to make a FAAA noise, right?
00:10:13
Mohit
Like the Instagram FAAA noise.
00:10:14
Priyank Upadhyay
Yeah.
00:10:15
Mohit
And I was like working with this user like, OK, how can we enable this for you? right like It's interesting. It's like, I don't know how this adds a net positive value to your workflow, but it's definitely interesting.
00:10:26
Mohit
And now we have a custom chime sound whenever cur cursor completes its chat. right So you can probably install like a far audio file to it and be like, every time your code gets like pushed, you can be like, far, far. right like so It's just interesting. Again, I don't think it's dumb. It's just like people are trying to now use this tool to make their life fun. How can I make it less boring? I know someone who's had a sort of like this tool where every time our meeting is gonna come up, like in five minutes or two minutes, they will play the breaking news.
00:10:57
Mohit
voice of Archstuck, right? Like Archstuck noise they would do and it's like prepping them for the meeting. So I think it's still like incredibly cool that, okay, people are finding like joy in small things now that they can do coding without actually knowing how to code.
00:11:09
Mohit
So yeah.
00:11:10
Priyank Upadhyay
I'll try that out today.
00:11:12
Mohit
yeah
00:11:12
Priyank Upadhyay
I love that far sound after my code is done.
00:11:16
Mohit
Oh, I tried it at 11. I was shipping something and there was a FAAA in my bedroom. Trust me, it did not go well. It was like, damn, what is this? So yeah.
00:11:25
Priyank Upadhyay
Nice.
Emergence of New User Categories
00:11:28
Priyank Upadhyay
Any new category of users you are seeing emerging for Cursor which you did not expect at all? Like developers, engineers are fine, but something that's totally strange for you.
00:11:43
Mohit
I would actually place myself in that bucket, right? Like even while joining Cursor, I was like thinking that, okay, this is great, incredible. Now I'll work with the users, I'll answer their queries and I'll move on, right? But like now when I'm here, like it's just become like, oh, okay, I see like an internal tool we can create here to make this process easier.
00:12:03
Mohit
And then I end up just shipping it, right? And I'm sure like I'm not the only one in this, right? Like there's enough and more people like me across different companies, industries probably doing the same that, okay, this used to be on our Google Sheets and we used to manually go update it but can we now like connect it to our Slack channel and like put some automation around it and like auto-update itself, right? Like, so I think for me, like this whole category of like, again, as we...
00:12:27
Mohit
said right like non-degree qualified sort of people who were not computer science graduates or don't have the experience are now able to ship like something which is Might not be, again, from a conventional engineering point of view, the best work, like the code is optimized. Might not be that. But is it getting the job done?
00:12:45
Mohit
If the answer is yes, I think they are still good, right? like So I think that definitely is a category of people i find myself falling into and a lot of my teammates and like people outside who we work with.
00:12:57
Mohit
So I think that has been interesting. other than that i also think like this has enabled sort of people who and i think this has been spoken about quite a bit on twitter and across different companies but like i think it has been very interesting to see how people who sort of thought they graduated from this ic level to like managers senior managers head of like you know our department But now you see and they are also coding and like shipping something. right like It's no longer like, hey, let me text my you know like the person I was managing, let me text them and ask them to like work on this for the next week.
00:13:29
Mohit
They are themselves on curses now and like be like, okay, what dashboard will be the best for me to manage my team? So I think for me that shift also is something which has been pretty interesting that now people are going back to the whiteboard themselves instead of thinking of themselves like okay we are only here to manage processes and people and we are not here to really do anything in a way right like but now they are actually doing and it's actually making the organization more efficient as a whole so i think that is also super interesting
00:13:57
Priyank Upadhyay
Do you have some stats, numbers about the category of users or any kind of personas that you identified who is using cursor most often versus most least?
00:14:12
Mohit
Oh, I might not have like the exact, like any sort of like figures as such, but I think I can confidently say like Cursive used to be like this tool, which only hard code engineers were using. And like, they needed to know a lot of like these, how does the, you know, like a particular Python package works and then I'll be able to like install it and like all of that. But I think now it has been shifting. I have been seeing like that, uh,
00:14:38
Mohit
bar go a bit down that it's no longer like just for like the hard code engineer. Like as I said, right in the beginning of our chat, when I started using cursor a year ago, I found it a bit intimidating, right? But now in the advancements and everything, I have been, like I have 4,000 transcripts on cursor, right? Like I have a 12.8 GB transcript file on cursor, right? So that just goes on to say that if I am someone who was non-technical and have chatted with cursor 4,000 times, and of course I work here, that's partially it, but also just,
00:15:07
Mohit
so shows that there's so much I can gain from it, right? So for me, I think like definitely there's a more inclination of engineers like people who are software-driven in that sense using Cursor
00:15:18
Mohit
But at the same time, I think the gap has been coming down. like Increasingly, a lot of non-technical people are getting in on it too.