"Building Product at Reflect" with Alex MacCaw, founder of Reflect (1/2)
How A.I. can augment our lives as our deeply personal thought partner.
You can watch this interview on YouTube, listen to the audio, or read the transcript below.
Disclaimer: The interview has been lightly edited for length and readability.
What did you learn from Stripe, Twitter and Clearbit that you bring to Reflect?
Jacob: You have a history with Twitter, Stripe and then founded Clearbit. What are some of the early lessons you've learned from your experience working for these other companies as well as founding your own?
Alex: Stripe gave me a lot of lessons about what to do and maybe Twitter about not what not to do. I would say I came out of Stripe with a real appreciation for good design. Stripe has a world-class design team, and when I joined Stripe, I was working on some of their designs, implementing them. I was a software engineer. And I wouldn't make them exactly perfect. I would think, you know what, it's 99% of the way there. It's good enough. And then they would ask me to do it again and again.
And that was really good learning, honestly, just getting that level of perfection, it's really important. Especially when you're running your own company. Cause the only person who's gonna say that bar is you. So that helped, that was a great thing to learn.
Jacob: We're gonna circle back around to taste in a little bit, because I really appreciated your article on that, we'll put a pin in that for sure.
Alex: Oh, you have done your homework. Yeah, I'm a big believer in taste companies.
And then, when it came to Twitter, I was there maybe when there were a thousand people and it just really showed me how bloated orgs can become. This was a long time ago and they have only gotten worse since then.
I had no idea what half the people there were doing. I knew they weren't doing much work because I wasn't doing much work. And it's amazing what has happened recently. Whether you agree or disagree with some of the changes being made at Twitter, you can't deny that they are shipping faster and they're shipping faster with 20% of the company.
What does that teach you? That teaches you that shipping speed is entirely a cultural thing? They already had the engineers, they weren't shipping stuff. It is to do with culture and it's to do with internal process and now they're just shipping like crazy.
Jacob: And looking back on their history, they've cycled through four or five different Chief Product Officers in the last five years. And despite each of them being world class in their own right, the shipping cadence didn't really change much during those five years, which is pretty wild.
Alex: I think it is up to the CEO. You have to come in with a sledge hammer and change the culture completely, and that's what happened. One Head of Product unfortunately, is not enough to move the tanker.
What did you learn from world-class CEO coach, Matt Mochary?
Jacob: Speaking of CEOs, you have been, blessed to be coached by Matt Mochary who is a world-class CEO coach. What are some of the takeaways that you've learned from him?
Alex: I got him right at the front-side of his career. Otherwise he wouldn't work with a company my size. But I actually ended up co-writing a book with him, or I say co-write. I lightly edited. He wrote the book called The Great CEO Within, which has got a lot of the things he taught me in that. The great news is you can go and just read that right now to get an idea of what his coaching is.
But I would say it's very process oriented. He comes into your company and he sits down with you and he's in all your meetings and he just sets up these processes so that you can essentially do all of your one-on-ones back to back on one day. And then you have the rest of the week freed up. As CEO, that's amazing because then you have the rest of the week to be creative. But that's one of the main things, man.
There's so much in that. He taught me to really listen, taught me to repeat back, to make people feel heard. He's just the master at systems, like he has a Google Doc for everything. So I really recommend, if people are interested, go and read that book.
Jacob: I believe he also open sourced all of his word docs that you mentioned as well. So anybody can go and check those out.
Why did you choose to build a note-taking tool?
Jacob: I wanna move into the meat of this interview, which is Reflect. I've gotta start with why did you decide to choose to build a tool for thought? The way that you've done it has been so beautiful and I'm just such a fan. Bring us into the thinking around why you chose that.
Alex: Let's just say there are no taking apps and then there are note taking apps. A lot of developers do try and scratch their own itch, but typically they leave it at the kind of the markdown phase, which is just the very start. Building a notes app is unbelievably complicated.
Like a proper, full featured one, one that has a WYSIWYG and offline sync and end-to-end encryption. these are really difficult problems. And now are you adding A.I. into the mix as well? This is like the cutting edge of technology to build these notes apps, which is honestly what makes your question even more valid, given that it's so complex.
And yet there are some really good solutions out there. Why did you do this? Why does the world need another notes app? I don't have a good reason for you other than it's a very special passion project for me. I came out of Clearbit, which is a data business that I built to a couple of hundred people, and I hadn't been coding for a few years. I really wanted to get back to basics. I wanted to get back to shipping every day. To writing code. That's what I loved doing. I wanted to try my hand at consumer. I had been in B2B, the whole world. I wanted to see, if I could do consumer, it's a lot harder. I had no idea when I started how much harder it is. But I wanted to try that.
I wanted to build a tool that I use every day that, that, again, that's probably not really possible in B2B. It's at least, it's a lot more difficult. I also wanted to build a small company. Again, B2B is difficult. You have to have a fairly sizable sales and marketing team, and I wanted to avoid all of that. And today, Reflect is just four of us and we serve roughly 2,200 customers, which is amazing that you can do that these days. This kind of company has only been possible in probably the last decade when you have all these tools and frameworks that let you automate things like SysOps, and even when it comes to your accounting, it's all fairly automated at this point. So I can just concentrate on doing it what I do best, which is build product.
Jacob: Is there a particular reason why you chose to go with the outliner note-taking style?
Alex: I was just looking at my dock and thinking, “okay, I wanna build something I use every day. Which of these apps can I replace?” And then I was like, you know what? Apple Notes, I'll build a note taking tool. Yeah. Not a very sophisticated market analysis I grant you, but the whole point of this project was, its a very personal passion and, if there were other people out there who liked what I was doing and had a similar way of thinkings in mind, then great. If not, no worries. Luckily at this point, it's not really a money motivation for me.
And so that was kinda the genesis of it. There are lots of different types of note-taking tools. There are some which are just like strict hierarchy, like maybe Workflowy. There are some that are completely unrestricted, like Apple Notes. You have Roam, which is what I classify as an outliner, where they are just restricted to bullet points.
I wanted something on the fairly unrestricted side. Just because that's how I like writing notes. Everyone's a little bit different. I sometimes like writing with bullet points, sometimes I love writing long essays and I wanted just a beautiful interface that would support both.
How is Reflect thinking about A.I.?
Jacob: Outliners have nodes which are infinitely nest-able. And so you don't get stuck in this hierarchy of folders. And what's interesting with the outliners and back links is we've effectively set ourselves up for success when it comes to A.I. and that's the next part I wanna dive into.
With nodes, having parent nodes and children nodes that, like I said, are infinitely nest-able, there's a a good chunk of context that's built into a single node. And so when you run a query against it, there's a lot that can be done just based on the context of its parent or its children.
Can you talk more about that and how Reflect is thinking about implementing A.I.?
Alex: Absolutely. So if you think about how the brain works and remembers things, it typically works with association. So if you think back to the last thing that you forgot, maybe your keys, what have you, the way you remember is you're like, okay, what do I remember that was associated with that item? And it might be the last room you were in.
With Reflect, you know that person works at a specific company and you may have forgotten their name, but you remember the company they work for. Because all these notes are connected through associations, you can trace this path. And because it's in digital form, it doesn't erode like our brains erode over time, and so we can use this as a backup to our memory.
What's really fascinating is seeing some of these A.I. developments and how they work because, was it like 10 years ago, there was a paper released called, "Word to Vector." I don't know if you're familiar with this, that changed everything. Someone figured out that they could essentially put a word into vector space, and it's like a graph, except, a graph might be 2D. But this is very similar where a word is associated with the other words around it.
And people developed this technology and they started using the meaning of sentences rather than just words. But this is exactly how GPT-3, GPT-4 work. It is incredible technology and all they're doing is looking at this vector space and trying to predict the next word based on the previous words.
So that gets us to Reflect. Like how could A.I. impact this? Like you said, we already have a lot of these associations being tagged through backlinks. We can use A.I. to do this automatically. We can use semantic search to essentially put all of our notes in and the words in them into vector space, and then move through the nodes via vector space rather than having to manually backlink and mark up these nodes.
And the potential is massive. The privacy implications are also massive as well. Your notes are your most sensitive items. I think probably more sensitive than anything else, than anything. You can change your credit card number if it gets lost, but you might not even want your spouse to look at your notes.
These are essentially your innermost thoughts. That is why, in Reflect, we end-to-end encrypted the notes. This obviously has some drawbacks. When it comes to A.I., for example, we can't just send all of your notes to OpenAI to vectorize them and ask questions about them. That's the line in the sand that we've drawn.
There are companies out there like Mem for example, that will just send everything over there and I just feel uncomfortable with that.
Running models client-side
So what we are doing is essentially waiting until we can run these models on the client side and it's gonna happen very soon. We have actually been sponsoring a bunch of GitHub projects to make this happen even faster.
Chrome just released WebGPU which is the next thing we need to implement this. So essentially we'll be able to run this models client-side. So what does that let us do?
It lets us do semantic search so you can search via meaning, not just exact words.
It lets us do auto backlink population.
It lets us do predicting the next word.
It lets us do spelling and grammar detection much better than is natively possible on the OS.
Reflect’s existing A.I. tools
And then we've already done some of the things that we could do already.
So we have a GPT-4 integration and we have a Whisper integration. And GPT-4 is essentially a tool you can use to transform text. So a text prompt goes in one end, different text comes out the other end. What happens in the middle is up to you. You can get very creative about it. We have shipped a GPT-4 integration with a set of system prompts.
And these prompts will let you do things like,
"Hey, take this big block of text and pull out the to-do items, the tasks, the action items,”
or, “summarize this text”
or, “turn this text into a blog post”
or, one prompt I have loved is taking my favorite writers and saying, “turn this text into a paragraph as if written by this writer.”
And that has been working really well with GPT-4. And we let you save these prompts and reuse them.
And then, we have integrated Whisper, which offers nearly human level audio transcription, and this I use every single day. In fact, it has become my primary interface to the computer.
I use a little app called superwhisper.com, which is still under development, but I use that instead of typing now. And inside of Reflect, I use our native Whisper integration to journal. And I've actually found that when I say my journal, rather than write it, it changes in a way that I like. It becomes more filled with emotion. And I cover more things and then I also summarize it as well.
So that's, that is how A.I. can influence Reflect. When I started Reflect more than two years ago, I didn't anticipate A.I. to be at this level already. It's just a happy chance that Reflect came along at exactly the same time as this A.I. is coming along and now it's integrated in the product and it's awesome.
Does A.I. eliminate the need for organizational structures like folders, tags or backlinks?
Jacob: If you have something that exists in vector space, let's say you type a note about how you were feeling yesterday. That semantic meaning is captured by its placement in the vector space. It is relationally and geographically near other notes with similar semantic meanings, right?
Which to me - and I could be ignorant here - eliminates the need for things like back links or folders or any other structural organizational system that notes have typically adopted in the past.
Am I thinking about this wrong or in the future, will we be able to throw notes at a blank screen and it be semantically associated with your other notes and we don't really even have to worry about back-linking.
Alex: Yeah. So and when you say “throw it,” I suspect it'll be transcribed, and you'll be able to hover over any sentence or click it and see a list of all the similar notes, right? It's gonna be amazing. It's very close. I suspect all of this will happen this year.
How will we interface with A.I.?
Jacob: Absolutely. It makes me naturally think about the movie Her. Have you seen that movie?
Alex: Yeah. I think the future is honestly gonna look a lot like that with a lot less user interface. It'll just be you communicating with this A.I. It'll be very personal. Its prompts will be built up over time, so that they're very accustomed to you. And I think it'll be people's best friend, honestly. It's scary to think about, but it's also very exciting.
Imagine having a genius thought partner to work with, right? Now, if you want a genius thought partner to work with, you have to pay them a lot of money. And most people just can't afford that. You essentially have to start a large business, attract this kind of talent, pay them a huge amount of money, and even then they will do other stuff during their day. They're not ready at your every whim. Whereas this is gonna be like a very personal thought partner that never gets tired and is the smartest person.
Jacob: What's interesting to me is the idea of taking that genius thought partner, combining it with your notes, and effectively you're having a conversation with yourself, with your own thoughts, which to me is mind shattering because I've naturally had to, over the years, go search for something that I may have wrote about at one point.
So I'm navigating through my back-links, trying to find that one thought that I knew I had and I couldn't quite remember. And now with speech-to-text, with Whisper, and then eventually, with text-to-speech all being so fast and local on your device that you could feed it in the context of your notes, you could literally have a conversation with your mind. I'm so excited.
Alex: We're so close, all we need right now is to have slightly bigger context.
What is the impact of being able to have conversations with our notes… our past-selves?
Jacob: One of the interesting things that Sam, your Head of Growth, wrote about in an article was being able to ask yourself questions.
For example, which city should I live in? It could analyze the thoughts that you've had over the years and say, “you really seem to like all of your vacations to Colorado. You love skiing. You love hiking in the mountains. It's only a five hour drive from your parents and the cost of living is appropriate with your salary.”
Stuff like that could be analyzed and extracted out in a way that maybe, could suggest some cities you've never even thought of before.
Alex: Yeah, that's quite true. Some of those big questions like, where should I live? Who shall I marry? It seems feasible.
You know what I'm looking forward to that’s gonna happen in 10 years time, is the singularity. Essentially, these A.I.’s start improving themselves, and a lot of people don't like talking about this, even though all these A.I. researchers fundamentally believe in it and are working towards it.
And we don't like talking about it cause it sounds absolutely crazy. It sounds like we've joined a cult and that we're waiting for the end times, and I always thought, you know what? Yeah, it does. It's got a lot of religion cult vibes to it. But if I was being pitched to by someone at my doorstep and they were like, “look, this thing is happening in the next 10 years.” And I'm like, sure, fine. And then they were like, okay, let me prove this to you. I'm gonna show you like a little miracle right here. I'm gonna do a miracle in front of you. And then I saw this miracle happen. I be like, “wow, maybe they're right. Maybe they're onto something.” And that's what I think is happening right now.
What does the future of Reflect look like?
Jacob: Every founder has a vision for their company that goes beyond what the company is currently. What is that vision for you, for Reflect? Let's call it the 10 year vision of what Reflect could be in its most - I hesitate to use this word, but - perfect form.
Alex: I'll give you the 1 year vision and the 10 year, cause I think it's quite important not to lose focus because all these shiny things can come along and you can get excited and work on the next thing.
Speed is the priority
Ultimately, if your notes app is not fast or is buggy, then it doesn't matter if you've got the shiniest features integrated. We have been very stable for a long time, but we are not fast enough. Our app doesn't open faster than Apple Notes on mobile. And we also have memory issues with big graphs. So that is like the number one thing we're working on right now, we've got to improve the speed of it. It's gotta open as fast as Apple Notes, otherwise you're not gonna use it instead of Apple Notes.
Task integration in Reflect
I want to build out a really good tasks integration. I think it is gonna be very difficult to do this because I don't want to build a tasks app, and I think you could go down that rabbit hole and I could totally see people still using Things, for example, alongside a Reflect, which is my favorite task app. But to-dos and notes are very closely related. It is our number one feature request. We need to build like a really nice UI for tasks.
Alright, that said, let's talk about 10 years time. Ten years time, maybe Reflect doesn't have an interface at all. Maybe 10 years time, it's basically just the movie, Her.
I do think ultimately the best kind of UI is no UI. That's why I love using Whisper and audio transcription. I think that could be the future of Reflect and because of the open source movement, it's possible. This is one thing I did not predict. I've been looking forward to AGI for a long time, but I did not predict this open source movement. It's been amazing. And now like over the weekend there were like four open source models released. It's nuts. And it's so cool that we're democratizing this and everyone will be able to run these on their own machine.
So yeah, I'll leave you with that. But thank you very much for, your time today.
I hope you guys have enjoyed this interview as much as I did. If you enjoyed it, please share it so others can learn from world-class builders like Alex.
Subscribe to get more content like this in your inbox each Tuesday!
❤️ Smash that heart!
If you enjoyed this article, smash that heart icon to show some love! 🙏