QUILL: AI CHIEF OF STAFF WITH MICHAEL DAUGHERTY

E108 QUILL_ AI CHIEF OF STAFF WITH MICHAEL DAUGHERTY

Transcript

Dalton Anderson (00:00.632) Welcome to the podcast where we discuss entrepreneurship, industry trends, and the occasional book review. Joining me today is Michael Doherty, the CEO of Quill. Michael is an MIT grad and the former chief strategist officer at Angel's List where he had a front row seat on how one of the world's most successful companies operate and how he's now building the next generation of production tools, turning messy meetings into crisp, actionable workflows. We're diving into the future of AI agents, privacy, and

why your next hire should be AI Chief of Staff. Michael, it's great to have you on the show.

Michael Daugherty (Quill) (00:32.302) Thanks so much. Great to be here.

Dalton Anderson (00:39.556) So today's agenda is gonna be origin story of Quill, otherwise known as Quill or Quill meetings. And then talking about this chief of staff vision and how the product of position to deliver on that vision. And then the strategy of like competing against bigger organizations in the regard of like the stance of being fast, nimble and having.

I would say the utmost passion on solving the problem, whereas it might be shipped as a feature somewhere else, but at Quill, it's going to be something that is very meaningful to people working there. And then I think a little bit, not necessarily tech stack on the fact of like what tech you're using. It's more of, if we get into it at all, would be like, how do you think about like localized models versus a

Michael Daugherty (Quill) (01:09.582) Mm-hmm.

Dalton Anderson (01:34.4) model like a big LLM model in the cloud versus like, I the stance was to take like a localized model approach. then think lastly, if we have time diving in some of the insights that you've gotten from meetings and management at Angel's List.

Michael Daugherty (Quill) (01:50.572) Okay, all right. Yeah, so maybe I start with a quick intro to Quill and what Quill is. So Quill is what we call a chief of AI staff. And the position of AI in that sentence is actually important because you could imagine AI chief of staff as well. And the difference to me is an AI chief of staff ultimately manages people. That's what a chief of staff does. But a chief of AI staff is ultimately going to manage the way we work in the future.

Dalton Anderson (01:55.224) Yeah, perfect.

Michael Daugherty (Quill) (02:19.394) which is many different AI agents, multiple projects going on at one time. And so where we've started is we've started by trying to understand you, the user, through the context of your conversations. If we think about how work will happen in the future, we're going to have smaller teams, each person responsible for many more different projects, more execution. But the most important part of the company is actually the human judgment. So people are going to potentially have more conversations where

You're thinking high level, what's our strategy? What is most important for us to be doing next? What would be considered a good job there? And then we need to farm that job out to a lot of AI and or tools, et cetera, and coordinate with other people. And all that coordination overhead can go down very, very close to zero eventually because AI can actually do this execution. And so if you put up a Quill today, the very first thing it starts to

to do is you have your first meeting, try to have a meeting over 15 minutes, and it will give you a pretty crisp document afterward. It will analyze the type of meeting that you just had, recommend specific document formats, et cetera. But you can start to connect other tools to it. And so I've connected things like my linear, my air table, my obsidian, my notion, et cetera. And then as you have conversations, it knows that you also use these other tools and starts to learn who

who you interact with, who else is on your team, et cetera. And so it can come out afterward and either recommend to you, hey, it sounds like you just discussed a bug with a customer. Let's go make a linear ticket and let's assign it over to someone on your team. Eventually, it'll probably assign directly to an agent to start working on. Or it might say, looks like you just had a coffee chat with another founder. Here are three people from your Airtable that you might want to introduce them to.

And so, and do you want me to go ahead and write an email? So it can recommend things like that, but also at any point, of course, you can ask it and where it's going is it's becoming self-modifying. So it's going to start to learn over time, what are the types of things that you say yes to that you want to do? What's the feedback you give it? And how does it learn that and become better at being your assistant in the future?

Dalton Anderson (04:38.016) It's really cool. There's a lot to unpack there. I think one of the coolest things when I me on the brief demo was the, learns a lot about you and then can take actions on your half that reflect what you would priorly do, but also not only what you do prior, but make suggestions on what it thinks is the next action. And I thought it was really cool where I had the takeaways, the followups and the different nuances of stuff that you're working on when you showed me like a brief overview of like your quill.

Michael Daugherty (Quill) (05:01.72) Mm-hmm.

Dalton Anderson (05:07.02) And I thought that was really cool. But like the fact that you could see a future to where not only is it orchestrating your, say your day to day and your projects and helping you manage that, but also things that are in your personal life of meetings. Like that's another thing is like when you have a meeting with somebody and you can jot down some brief notes and then from there to, okay, like these are the other people that you've interacted before.

it might be good to do a warm intro to help them out. And like that kind of stuff is huge. And then another thing that you said that piqued my interest was eventually you'll get to the point where maybe a ticket gets submitted or you meet with product and product has some kind of idea gets approved. And then there's built out a plan. And then from that plan, it assigns to either engineers or like in the future, I guess we'll, they're going to be considered like AI engineers.

But like I think about it where I'm using a platform pretty early using it, a platform called Jules and Jules is by Google. don't know if you're familiar with that, but okay. Yeah. So Jules and I built some stuff with it and it seems quite interesting, but I could see a future to where the there's an AI agent that gets sent a task and then it makes a copy of the repository, does its thing, does its tests, builds it, brings it back. And then

Michael Daugherty (Quill) (06:13.387) Yes, yeah, I've tried that.

Dalton Anderson (06:33.72) there's some kind of approval process, either human or some kind of confidence threshold to where an agent approves it, like some kind of agent critiquer or reviewer, commit reviewer, and then that's committed back into the main branch. But you can think about that across a whole organization, just, yeah, it'd be crazy.

Michael Daugherty (Quill) (06:52.75) So those sorts of agents exist. And where they are today for the engineers is you can get it going pretty far on task. You have to specify the task in a lot of detail upfront. You need to make sure it goes back and forth and understands what it's going to do. And what I've found works really well in our company is knowing what everyone's expertise is. So everyone has deep judgment and experience in some area. Our designer.

has great taste. Right? And so her North Star includes the line, like, say no to ugly. And a lot of stuff that AI is going to come up with is going to be ugly. And so she, but she knows, what's interesting is she knows the vocabulary and she's seen enough to like know what she likes. When she works with an AI on design, it's much better than when I work with an AI on design. And conversely, when I work with an AI on engineering and like system architecture,

it'll come out better than she does. And so I think this is kind of where the role of humans comes into play is everyone has their unique judgment, their area of expertise. They're also pretty broad and AI can help you become broad and understand like the goals of the company. But we're basically working with these agents and she's making sure that the design is coming out really well and giving them feedback and training them on how to do better design. I'm working with them on making sure the...

system architecture and the underlying data model is designed correctly, etc. And continuing to iterate on that. so but and in the future, like where Quill comes in is, is not that we have one AI manager that's managing the company and it's like monitoring her and monitoring me, but actually, in order for us to be as effective as possible, we each have an AI assistant, right? So we've got our little Chief of Staff and, and as

as we maybe the two humans, the most important people in the room actually meet and have a conversation. She says, so I think it should work like this. And I say, here's some other ideas about how something might work. We go back and forth, eventually kind of come to an agreement. And then we send it off to our agents. then I'll work with it on one side, she'll work with it on the other side. And ultimately what comes out is interesting work. But it also speeds up the number of things you're doing in parallel. And so.

Michael Daugherty (Quill) (09:15.63) These days, if I'm really in the zone of engineering, probably have five, six cloud code agents going at the same time. And I think a lot of work is going that way. So it's definitely multitasking. And so at this high level, you also need something to kind of keep track of everything that's going on and help you prep in the morning. Like, all right, I know what I'm going to do today. I'm pretty confident. I see my schedule. I've got little tips of what I need to remember at different times. And then at the end of the day, what do I need to compress and think about for the next day?

because life is just going faster and faster.

Dalton Anderson (09:47.613) I think that's perfect. I had the same problem where you just like, just, have something comes up where you had your day planned and you had some deep work and things come up. And then before you know it, it's 6 PM and you haven't planned for the next day. And then you've got some kind of networking thing or dinner, personal activity. And then you get to the next day and you're like, I, I don't have all my stuff squared away. And then it's kind of.

Michael Daugherty (Quill) (09:56.952) Mm-hmm.

Michael Daugherty (Quill) (10:04.931) Yeah.

Dalton Anderson (10:12.804) repeats and then you've got travel and then you have to take a couple of days to get back in the swing of things. then I feel like you're now things are moving so fast that you're just not, I feel like you're never in a good point of like, right, I, I know exactly how the day is going to be structured. At least like the idea of it, you have an idea, but it doesn't necessarily execute. Cause I think there's always a couple of things that come up like, like maybe this thing is dragging. I've got to follow up, do a meeting. Whereas like I think a lot of the

coordination of the planning can get taken off your plate and it's more of like, okay, now I've got the time to do the deep work. The, you call it the AI chief of staff or chief of staff AI?

Michael Daugherty (Quill) (10:53.71) Chief of AI staff is how I think about it. It's a little bit of an awkward phrase. And so maybe we won't push that so much externally, but it is how I think about it because I think work is going that way where you have a staff of AIs and you need to manage them.

Dalton Anderson (10:56.141) Chief of AI staff.

Dalton Anderson (11:01.12) Yeah.

Dalton Anderson (11:07.264) No, I agree. And then also helping you manage your schedule and then knowing what stuff to follow up. But then when you bring in this. Like horizontal scaling or like vertical scaling of using agents to to do different tasks and then reviewing the code and looking through it, then that's a whole nother ballgame to where if you have like so it's a five simultaneous things going on at once and then you've got meetings plus emails and people management, it's a lot.

and it doesn't necessarily scale with all that stuff because you're already operating at scale. So you need something to help put everything on a guide rail and or have the flexibility to build new roads to connect them back to where they need to go.

Michael Daugherty (Quill) (11:51.916) Yeah, so that's the vision with Quill. I will say where we started, maybe this gets into a little bit of how Quill started, was very much as a meeting note taker and a private one. But our initial thought on the meeting note taking side was when you come out of a meeting, what's frustrating about having a single bot in the call is that every person gets exactly the same summary. And so what that means is it can't actually be most useful to you because it has to be generic.

let's say you're meeting with a client, they're not going to have the same take away as you. There are some things that you guys agree on, but actually what they're going to do afterward is they're going to go back to their company and want to have detailed notes on the product you were pitching them and how it might help them and help them make their decision. You're going to come back and say, who are the decision makers? Like, what are the blockers this person brought up? What did they like? What did they dislike? I need to go talk to my co-founder about this. And so have different things that you actually want to do coming on call.

And that's why we had this thought that everyone should essentially have their own note taker. And what Quill does from the very beginning, like I said, is kind of categorize the meeting. It's got a lot of templates. It's got your job as part of the, you enter your job as part of the onboarding and can write as much or as little as you want. That goes into this categorization and understanding of what it should recommend for you actually to do next. And so even the very early version would say things like,

I just met an engineer. I should write an email intro to my co-founder because that was one of the things that I put into my set of templates and it would frequently recommend that. And I still use this. I interviewed an engineer yesterday and it popped up and said, you have an rubric that you typically use. You should just one click, run your rubric, and it gives me this nice document with evidence for and against various skills. And then I can make my decision.

and it's kind of organizing everything and it's very quick to operate that way. And that's very different than what I want coming out of an internal meeting. I want something totally different. And so it's all about like, what are you actually going to do? Who are you going to communicate with and what information do you need? And so that's kind of where Quill came from and where its core is today. And as we build out more of these features around like, now we go cross meeting.

Michael Daugherty (Quill) (14:15.47) We've got lots of ways you can search across your meetings and you can consolidate information. But now we're building this into your daily workflow. And so our homepage now is a view of your upcoming calls and it knows who you've met with before. And it'll put a little reminder, hey, last time you said you were going to follow up on XYZ, this sort of thing. And at the top it'll say, today it seems to be a focus engineering day because you've got demo day tomorrow. You've got all this stuff. I think there's a lot more we can do there to help you.

help you out, even as it is, it's so incredible to be able to go look at my contacts and know who I should talk to about XYZ and all this, because all that information has already been entered without me doing anything just through the conversation.

Dalton Anderson (14:46.564) Bug and beyond.

Dalton Anderson (15:00.836) Yeah, above and beyond what you can do now. And that's thing that frustrates me so much is like you have a team, you have the teams meeting. I'll name drop the company, like Teams, Microsoft is a big company. They've got plenty of talent. They're just, there's not enough focus on getting the meetings right. Like when you have meetings, like I think meeting recordings should be centralized and some, if it's like a department thing or you can list out a folder where the meetings go. And then if,

Michael Daugherty (Quill) (15:10.062) Mm. Mm.

Michael Daugherty (Quill) (15:18.67) Hmm.

Dalton Anderson (15:29.75) your co-lead, like your co-founder might not be able to attend every meeting, but wants to just be in the loop where there's somewhere like the meetings are stored and the trans the transcription is available and or the notes can be processed, but everybody gets the same transcription and or the AI generated result, which I find really frustrating because like a couple of years ago they would transcribe everything and you could set it up to transcribe automatically. But then

you could just just one click like they had the copy and paste icon and you could just copy and then paste into your AI trained chat or gym or whatever you want to call it template. And then you can make your notes the way you want them. But the issue with that is it doesn't constantly have all the context, like background context, which is really important, but it's good enough to where you can have a nice template and it's consistent. And it's important to what is for you in that.

area at least as good as it gets, but it's missing the background context of all the other meetings plus the other stuff you've got going on, which was what Quill will do. And the other issue I have is that everybody's got the same result now because the, made it difficult to get the transcription now. Like unless you're the owner of the meeting, you can't get the transcription, but you can get the AI notes. So I get the AI recap, but it's not in the format I want. I don't get the raw data. So I can't format it the way I want. And it's just like, like

Michael Daugherty (Quill) (16:53.773) Mm-hmm.

Dalton Anderson (16:56.324) This is, this is not optimal. Like one of the key things I don't like about the AI recap is that the takeaways are at the bottom. Like why are the takeaways at the bottom? Like that drives me insane. Why is it? No one's going to read the long, like multi-chapter thing. The first thing they want to know is what do you need from me? And then if they don't know what they're, what we're talking about, they could read the summary, but why are the, why is it like a 2000 word like summary, whatever it is, 800 to 2000 words.

Michael Daugherty (Quill) (17:05.812) You want them at the top?

Dalton Anderson (17:25.316) depending on how long the meeting is. And then the takeaways are at the bottom. No one is scrolling that far. Not everyone has Logitech MaxDirects keys, where the scroll is easy.

Michael Daugherty (Quill) (17:29.902) There are two...

There are two things you said that kind of come back to how I think about product. One is you were saying even two years ago, you were getting transcripts and you were going to chat GPT or whatever tool you were using and trying to get your own specific takeaways. The format you want, what you want to do with them. I like to look at the people that are far ahead. And back then that was considered quite far ahead.

of like the standard and it's a bit of a manual process, right? But people that are far ahead are getting a lot of value so they know what they're doing. And so we had some users that were doing that back then as well. And we kind of looked at them and said, all right, for these people, at least we can be faster. You can customize the templates. You can get one click what you were getting by copying everything, going over ChatTPT, saving the output back to your Obsidian folder and like this stuff. So it's at least a workflow thing. But what we can also do is we can

take that same workflow and by making it easy and making it a default, we can help another 80 % of people that are not thinking about this every single day. They're trying to get their work done and they're not as advanced as you, but we can take the things that the people who are the most advanced and getting the most value out and make them easy and automated to do and help the rest of the user base so that when you use it, actually makes you feel more on top of things, more productive.

The second thing you mentioned was like pulling data out and how it's gotten a little bit harder recently actually. And I think this is sometimes as large companies think about where the value of their data is. Sometimes they think, great, I've got all these transcripts. This is actually super valuable. can train my speech recognition model on this. Maybe there's something we want to store it in the future, et cetera. But I think this data belongs to you guys.

Michael Daugherty (Quill) (19:26.478) And I think that, and so our philosophy is like, let's make this data as available as possible. It is stored locally. We also have a local MCP server you can turn on, which other local apps can connect to. So we have a lot of users who connect to this through cloud code and pull in and use cloud code with their, know, with their, they're spending their $200 a month and getting Opus and all this stuff.

It can now search through your Quill meetings. You can pull them in. You can create documents on like, actually I did this myself the other day because I was thinking through all our different user personas that we have using the app and thinking about how do we continue to improve our product management direction. And so I had the personas. I started brainstorming them. I used Cloud Code to pull in all my recent conversations over the last three months with external parties and then see.

which of these parties map to the different personas, did I get the set of personas correct or not? And it kind of created some documents for me. And then I pulled in emails as well and was like, okay, well, how about support emails? Are they properly distributed? Are there personas that I didn't recognize? And this was really helpful. And it was because as you can pull your own data into different places, you can just do a lot with it. And so some of our more advanced users are

definitely out there. And they'll send us feedback on the MCP server. And they'll say, oh, I really want an MCP function where I can import contacts and add notes to them from my other systems, because I'm trying to build an automation here and centralize it in Quill. And then we'll say, great, actually, that's really interesting. Let's build that, first of all, for you guys. But then we can reuse that same function, put that into the agent we have in the product so that even people that aren't

as advanced. Again, they can do it just in the product now. can just ask. We call our assistant Quillium. It's like William plus Quill. He's a little hedgehog. But you can just ask him. You can say, hey, I've connected my PipeDrive, my CRM. I'm going to have a meeting with these people. I want you to find some of their background info from PipeDrive and annotate my contacts in Quill so that that information is available when I create notes. And it can go ahead and do that now.

Dalton Anderson (21:25.015) Yeah.

Michael Daugherty (Quill) (21:44.748) That's an example of something that came in from those advanced users because we let them do whatever they want with their data.

Dalton Anderson (21:52.342) No, all that's huge. And I think one of the things I've heard is trending is the people that are have on the show is like the people that have, I guess it's not an unknown secret, but I think it's more of like a do as I say more than as I do kind of thing where people that have a direct relationship with their customers and are constantly getting feedback and then integrate iterating on that feedback are the ones discovering new workflows and providing more value than

Michael Daugherty (Quill) (22:20.878) Mm-hmm.

Dalton Anderson (22:21.503) than otherwise. You know, I've never met with Microsoft teams and no one's asked me like how I would, how I'd want Microsoft teams to work. It's never going to happen, but like to have that piece is like one, they're your best marketers, they're your best customers and also they're your, your best product managers. Like they, they know what they want. And it's honestly a do all, say all type of situation to where if they say that this is what they want to do, that that's the direction that

Michael Daugherty (Quill) (22:26.126) You probably won't.

Dalton Anderson (22:51.075) you got to take. That's not what I'm saying. But I am saying is like you do get a broad perspective. And that's one of the reasons why on this show I try to get people from all around different sectors. I don't specialize in like one thing. I just what I specialize in is interesting problems and interesting companies. Because everyone's got a different perspective, but there are trends across each each sector. And this is one of them definitely where like the most innovative ones are like, OK, like I'm really dialed in and focused, but then I'm also getting the customer feedback.

Michael Daugherty (Quill) (22:53.422) Mm-hmm.

Michael Daugherty (Quill) (23:07.096) Hmm.

Michael Daugherty (Quill) (23:19.918) I think you're probably also seeing a big trend with right now the world's moving a lot more towards everyone trying to, or like the advanced users trying to build their custom software and customize their workflows with OpenClaw and stuff. And that's something I think about a lot as well. And like, how can we help Quill become more customized to you? But I feel like I interrupted, so let me stop there and get your next question.

Dalton Anderson (23:41.637) No, yeah, I was gonna say two things one I was gonna say high high 20,000 foot view of what MCP is for like the model context protocol for people that are less technical that are listening to show typically the people listeners are quite technical or curious so I think they'd be interested and you might get them in a rabbit hole later on and then my next question was like hey I know we delayed this podcast episode because you had some stuff that was

Michael Daugherty (Quill) (23:51.96) Mm. Okay.

Dalton Anderson (24:09.445) in the works in February and I was wondering if you wanted to let the audience know about the news.

Michael Daugherty (Quill) (24:16.334) Sure. Yeah, so OK, so first thing, model context protocol. It's a complicated name. If you're familiar with APIs, it's really just an annotated API. It's a list of functions that can be called with instructions for an agent. And this can be combined with another markdown document, which is instructions, like instructions on how to use this thing. And that's called a skill.

The whole industry is kind of figuring out what the right packaging is for AI integrations, because what you need to do is give it capabilities and give it instructions on how to understand how to use those capabilities. One of more interesting things in Quill is that as you connect APIs to it, right now mostly supporting MCP protocol, Quill can, what Quill does is it proactively explores your MCP that you've connected or your integration that you've connected to understand how you use it.

And so it can almost build its own instructions itself rather than having to wait for the company to provide instructions on how your AI can use it. One of my favorite examples is this Airtable I mentioned because Airtable has an API, but everybody, if you use Airtable, you know, you can set it up any possible way you want. Same with Obsidian. You could have any sort of folder structure you might imagine in Obsidian.

And so might have an air table that is a personal CRM. You might also have an air table that's like a launch list or a to-do list for preparing for podcasts or one that just like tracks your old electronics and like which closet they're in. You could be using air table for all sorts of different things. And so you connect air table and typically you'd have to go tell your agent, I just bought this new electronic, add it to this air table specifically. Instead with Quill, because it pre-explores, it'll

write itself internal notes for its own memory and say, Michael uses this air table to track his electronics. If he mentions electronics, think about what could be updated. Think about what research you should do. Think about what you might add to it. And so then if I say, I threw away my old 2019 laptop finally, just got rid of it, it might come out afterward and say, the agent, after the meeting, like I said, it gets categorized and we think about what to do.

Michael Daugherty (Quill) (26:33.838) The agent might think, oh, this is a mention of electronics. I should check out the air table. There's one we might want to delete. Can I go recommend that as an action to Michael? And then that pops up. So I don't necessarily have to think about it or remember it. It just pops up and it's like, do you want to update your air table because you mentioned you were going to throw away your old laptop? And I just click yes. And it goes ahead. so high level, like MCP is just how to use a tool and what are the capabilities. And then the second question was kind of what are

Dalton Anderson (26:50.307) Mm-hmm.

Michael Daugherty (Quill) (27:02.966) What were we building towards in February? And we've been talking about it this whole time, really. Prior to February, we had hints of some of this capability, we were very focused specifically just on meetings. And so a couple of weeks ago, we released a completely rebuilt version of the app that rearranges a little bit of the focus and is trying to become, like I said, this chief of AI staff.

And starting from the fact that if you have conversations, we now have users who have put in millions of words over thousands of hours of conversations on their local computer that provide the context so the agent can actually be really smart. So now if you get a cool pro plan, you get your agent, you get these smart integration suggestions, and you get a mini personal CRM that as you talk with other people, it kind of annotates things you know about them and updates their profiles so you can

You can search them and we're just adding more features. The big thing I'm thinking about right now is extensions. So automations and extensions. Because like I said, the trend is people are trying to build their own software and software should be customizing to you. And so if we build a good extension framework for Quill, where you can teach it to do new things, teach it to every morning, do research on the people I'm about to interview or whatever it is that's important to you, you can tell it how to do this.

tell it once or twice, and then you can write an extension to build that in as a function. So if we get the extension right, humans can build their own extensions. then the great thing about AI is it can code. So then we can build a loop where we can get our agent to take that and just talk to you and create its own extensions. so this is kind of where Cool is going, is just more and more customized to what is your specific job and how do I help you get better at your job every day.

Dalton Anderson (28:42.743) Hmm.

Dalton Anderson (28:58.698) It's that sounds like product improvement recursion, like where it's like the product user builds their extension. And then if there's like value in that extension for other users, it's like, okay, it's recognized that many users have built a similar extension. Let's just add that into our core product. And then it just constantly just keeps iterating.

Michael Daugherty (Quill) (29:03.438) Did.

Michael Daugherty (Quill) (29:19.63) Yeah, well, yeah, I mean, I do think about like with AI in the world, how do I complete whole loops so that it can iterate? I don't know if saw Karpathy's recent open source project. He has this mini researcher now. I, Yeah, I'll tell you what it is. So basically, you know, he's an AI researcher and he,

Dalton Anderson (29:36.925) yes, I did. I haven't read it, but it's on my bookmark and people are like, this is crazy.

Michael Daugherty (Quill) (29:48.926) typically comes up with experiments on how to tune his models, how to adjust weights or architecture, whatever it might be. He's like, I've got this hypothesis. need to run it. I update the hypothesis. I run a training run. I evaluate the results. And then that goes into his mind. He thinks about what's the next hypothesis. What he did is he basically just completed the loop with AI. And as soon as you complete the loop, it can iterate on itself. And so he made the loop possible to go fast. His first thing was he said, all right.

you have five minutes, come up with a hypothesis, run an experiment, has to be done in five minutes, and then stop and evaluate. And then once you have, and so he had it broken down into steps, evaluate, come up with a hypothesis, et cetera. And then you can just run it overnight. And so if it's running every five minutes, you can get 100 iterations in a night. And even though it's not huge experiments, they're tiny little experiments.

He said he wakes up the next morning and it's found 20 out of 100 that improved things. And overall his performance has improved by 10 % overnight. And you didn't have to do anything. And that's because he completed the loop. If you have only a part of it, all you can do is kick it off. It can run an experiment, then you have to evaluate it. Now you can't run it overnight. Now you can just do one at a time and you've got a human there. As soon as you complete a loop with AI, it can start to self-improve, but you need to really guide it in terms of what is.

what's a quality output and you know, when does it break and stop and ask humans for guidance?

Dalton Anderson (31:17.144) Yep, that and the architecture. You gotta have something that makes sure that the architecture of what you're doing is tight, where you're not just going off and making whatever classes or methods, reusing, not reusing methods and making it like hard coded, like you don't want to, like you could get some garbage code pretty quickly.

Michael Daugherty (Quill) (31:19.479) Yeah.

Michael Daugherty (Quill) (31:26.997) Exactly.

Michael Daugherty (Quill) (31:33.526) Yeah. AI can go off the rails because it goes so fast. And so if you complete the loop, but you don't have enough control over it, know, electrical engineering has always had these like control problems with feedback cycles. It's like when you have a microphone next to a speaker and it feeds back the wrong way, it like hurts your ears. But with AI, it'll either blow up your code base and it won't be maintainable. Or if you have it connected to like your email, it'll

delete all your historical emails and email a thousand people in one night because it wanted an answer and no one's responding fast enough. It goes crazy. So you need some controls over there.

Dalton Anderson (32:12.5) Yeah. Do you see those crazy stories where like, cloud code deleted, deleted my production database and all my snapshots. And so now it's gone. Like the last two years, my data is gone. Or I think I read something like AWS was out and they had an outage because they were using Claude or some kind of agent to coder. And it just to optimize the code base. Like it was like, well I haven't verified the story, but it was like more of like a tech feed.

Michael Daugherty (Quill) (32:34.99) Mm.

Yeah, I saw a headline today as well about something like that.

Dalton Anderson (32:40.524) Yeah, where it was like they deleted a piece of the code. They're like, we need to refactor the code. And then like, yeah, was like, the best way to do that is delete it and then rewrite it. And that caused issues.

Michael Daugherty (Quill) (32:49.922) Yeah. Yeah. Yeah. it's something as you, as you said, with AWS, every company is figuring out how to, how much freedom to give AI in different domains. and it's further along in some domains than others, but it still doesn't have that human judgment about like, what's the right thing to do? What's the right way to do it? it can, it can execute if, if you give it guidelines, but it's very hard to say, I'm going to, I'm going to turn over my judgment to an AI.

because it doesn't really have responsibility. can't hold it responsible. can't withdraw its pay or something. Yeah. It'll say, I apologize. You are so right about that. Let me try again.

Dalton Anderson (33:26.276) I can't believe you did that, Quilliam. That's horrible. I'm sorry. But like you can't blame your messed up slides or you missing a media on some AI agent. Like that's not gonna fly in any world where there is responsibility. So no, I think those are great points. And I'm excited to read that paper.

Michael Daugherty (Quill) (33:43.628) No.

Michael Daugherty (Quill) (33:47.331) Yeah.

Dalton Anderson (33:54.393) the whole closing the loop and they made, you made like that research, like the premise of the outcome was like now, like research is kind of weird, becoming a much weirder space because like, if you have these closed loops, like closed loops iterations between the AI just iterating, iterating, you can make like insane research papers pretty quickly, because you're.

Michael Daugherty (Quill) (34:16.622) Mm-hmm.

Dalton Anderson (34:18.306) you're not necessarily having to like go back, reevaluate, tinker, think about the problem. It's like, okay, if you run an experiment every five minutes, all day, every day, you can really turn out a lot of stuff. And I had like a prior professor, his name's Dr. Noel. and that's, that's how I refer to him. And he, he talked about like a couple of years ago, he was like, we need to move away from the

doctrine of you need to have like a research paper that you submit that gets approved for a PhD. Like his push was like, let's, let's do like advanced research for a PhD and the evaluation of that as like a project or some kind of like crazy thing that they build versus, that, that, that can be different in different domains, but it's really more relatable and like the technical domain versus like a paper that you worked on. I mean, it could be a project as well, but like,

I think he wants to get away from the notion of a research paper that you submit versus, and I'm dumbing it down. Like I get that it's, there's a lot of work that goes into, to getting approval to become a PhD. But he was like, let's move from that to a project because of, because of the nature of AI and how it's going to be moving always next couple of years. And that,

project that you're talked about is like the exact reason it was like why he felt that way.

Michael Daugherty (Quill) (35:47.438) Yeah, I think there's also something around.

A couple of thoughts. One is, of course, the reproducibility crisis with papers. And one of the nice things about if you ship some software, someone can go run it on their computer and prove that it does the same thing. And so it's a little harder to hide behind obfugating data or something. But actually, the project idea is very interesting. I think people have to practice with new tools and new skills.

I saw a founder yesterday who was starting to think about how to build a school for people in AI First World. And she was also talking a lot about project-based learning, like give people a challenge and get them to complete it, but they have to use different AI tools. One of the great things about AI to me is that you can actually have AI teach you. It's like our first tool that can talk back to us. And so...

You do have to learn what it's good at, what it's not good at, how to control it, as we talked about. But you can actually ask it to teach you and help you come up with things to learn. And you don't have to be afraid of asking too many questions. So all of us should think about five-year-olds, and they just keep saying, why, why, why? If you don't understand, just ask another layer. And make sure you're constantly judging if it's internally consistent or whatever. But if you build that sense, can use it lot.

Internally at Quill, I've built inside of Cloud Code, I built an agent called Learn Quill. And that on boards our team members. Because what it will do is it will first stop and try to assess your level of experience as an engineer, your level of knowledge of how the Quill code base, et cetera, learns. And then it creates a curriculum and tracks this curriculum over time for you. So you can go back to it any point and it will teach you the next lesson.

Michael Daugherty (Quill) (37:50.126) and it will do it interactively, find real examples in the code base or whatever. And so our designer is using that because she's now integrating more with cloud code so that rather than her iterating a lot in Figma and...

Dalton Anderson (38:05.22) is now there's an MCP connection with Figma and Claude code, right? Are they're working on it or?

Michael Daugherty (Quill) (38:09.26) There is. Sometimes it's better to be in the tool that actually edits the code. I want us to both have the same source of truth that we work on. But it's a bit of a mind shift for a designer. And so she's using it. It has a curriculum for her. I've written a decent amount of our code base. And I have it going, because we've made some changes. And other engineers have made changes to parts of the code base that they own.

And so I have mine going too, and I'll do a lesson once a day and it'll be like, Hey, do you understand this part of the code or do you understand this advanced usage of, of cloud code? I'll say, actually I'm not a hundred percent clear. So let's do a lesson. and so that's also something we're starting to think about how to build into Quill is to make the agent more proactive. So you don't get a standard onboarding experience of everybody has to click the same buttons and go through an onboarding experience, but it'll, but over time, how does it pop up at the right moment in time and say, Hey, I see.

you're a product manager, I see that you've been connecting to linear a lot. Here are a couple of new things you could try or let me ask you a few questions and kind of guide people through using Quill. That's not quite in there, but that's something we're definitely experimenting with because we found the internal Learn Quill agent is really good. This idea that software can teach you how to use itself and make everybody an expert user.

Dalton Anderson (39:31.749) That's also really cool. I don't really have a great ex, like I don't really have a great response other than that's, that's sick. That's, that's my immediate response is like, that's sick. I, there's a couple of reasons why I think that's really cool. Like one, you, as somebody who's like, you're running the company, like you'd want, you'd want people to feel at home faster and one because of their people too. So you don't want people to feel uncertain about what they're working on or,

Michael Daugherty (Quill) (39:37.162) It's good.

Michael Daugherty (Quill) (40:00.27) Mm-hmm.

Dalton Anderson (40:00.389) You want them to feel settled at their new position. Also, it's good to gel people quickly because then if they're on, if you reduce the onboarding time, then that increases the productivity. And if people are out for some reason, they can have some kind of workflow to review the changes to code base. If you guys are moving fast, like there's a whole bunch of reasons why it's super cool and useful, like from a human side, from a productivity side, then to bring that into the, and then I think the last point.

was the reduction of tribal knowledge. think that's probably one of the biggest issues like in a code base where like things start growing, they're growing fast. And Michael's the only one who knows how to do the, the learn code, Claude AI thing that was built. And there's a couple other parts of the, like 30 % of the code base that like no one knows anything about. And that's a problem because then when say something happens to that engineer, they move on, they go start their own thing. You're happy for them.

Michael Daugherty (Quill) (40:31.608) Hmm.

Dalton Anderson (40:54.584) There's just such a massive gap and it takes like a couple of months for people to really wrap their head around the code base in that area or that project or whatever technical project it is. And it's such a waste of time. And I know there's there's AI helps with documentation, but documentation is different than understanding. It's like listening versus hearing. Like if you're listening, it's intentional. You understand you're stopping. You're thinking about it. Hearing is like one ear out the other.

Michael Daugherty (Quill) (41:15.907) Yeah.

Dalton Anderson (41:23.62) And so like this learn, this learn quill agent that you built is more of a, a comprehension thing, not a documentation thing, which is way different. creating the curriculum is really cool as well.

Michael Daugherty (Quill) (41:39.384) Yeah, I mean, I do think actually for kids these days, it's tough, but it's tough for all of us because if things are easy, humans naturally want to just go to the easy version. The easy version of vibe coding is saying yes to everything. can edit everything. But as you mentioned before, it can get your code base out of like, can go a bit crazy and then you don't understand it. And suddenly now you're in somebody else's code base. I mean, it's really yours, but it's the AI that built it.

Dalton Anderson (42:08.42) Mm.

Michael Daugherty (Quill) (42:08.63) and you're trying to understand what it did and fix it. And if you've ever started in a new company where you didn't build any of the code or you don't know how they do things, it's much harder to operate in that environment. And so naturally, our inclination is going to tempt us to outsource decision making to AI as much as possible. And so this is where internally on our team, I'm trying to think about how do I make sure

we all build that muscle of saying, yeah, it's harder sometimes to stop and try to learn some part of the code base that we don't really understand and like recognize we don't understand it. And that's a signal that it's okay to stop and try and dig in. How do we versus accepting it and letting go, because that's where you get like compounding issues. As a young person starting a career, I think it's very

because you have to know how to ramp yourself up to having good judgment and good end knowledge very quickly. But also if you're getting a bunch of tasks assigned to you, it's very tempting to just like outsource everything to the AI.

Dalton Anderson (43:15.576) Yeah, yeah. Whatever you're saying. And then I your other point about using AI to help you learn. think there's either a couple of mindsets, like there's a growth mindset or the doom mindset. You're like, AI's gonna take over the world. I'm gonna have a job. There's no point. I'll just quit my job and go sailing or whatever it may be or live in the forest. And then there's the other mindset or the other side of the aisle is like, all right, well, now...

Michael Daugherty (Quill) (43:26.51) Mm.

Dalton Anderson (43:43.001) we have all this knowledge in a centralized area and there's issues with it sometimes, yes, but it gets it right enough at the time where you can ask questions and learn. And so if you're curious about anything, it's really just an aggregation of the internet and different things. I mean, there's more to it, but it's aggregating a lot of the internet knowledge, code-based books, research papers, internet blogs, whatever. And then it brings it to you. That's awesome. So you can ask whatever.

Michael Daugherty (Quill) (44:01.07) Mhm.

Dalton Anderson (44:13.13) and get information consistently. Whereas before you might have to search for it, take a lot more time, but now you can ask questions and really dive in and understand. if reading is not your thing, throw that into a notebook LM, make a podcast. you don't like podcasts? Okay, well now you can make videos. You can make like 10 minute videos of like what you're trying to learn. like it's just insane, like how quickly you can learn and ask questions. Like I'd recently done like the New York compliance exam for like my brokerage exam.

Michael Daugherty (Quill) (44:27.842) Mm-hmm.

Dalton Anderson (44:42.564) for the New York state, it's like a 450 question and exam. It's a lot of information, blah, blah. None of it all makes sense. But I threw all the legislation into notebook L.M., made some videos, made a podcast, a couple. It's interactive podcasting, ask questions and then made a and made a question set of like a couple hundred questions that I think would have been covered and then went through and did did multiple exams a day. And that's how I studied and I was able to learn.

learn much faster than you would typically because typically you would need some kind of guidance. And so like you can make your own learning tracks pretty quickly, like minimal effort, which if I were to do that myself, that would take me forever to make that big question set. Like just so you make sense to even take on that challenge. So like just the way that you apply the technology is important and the way you think about it and where it could help you is, is I think the difference maker and either like swimming

Michael Daugherty (Quill) (45:26.275) Mm-hmm.

Dalton Anderson (45:41.061) and surfing the tide or like bearing with it and just keep swimming, I guess.

Michael Daugherty (Quill) (45:48.268) Yeah, I do stuff like that too, because AI is pretty good at shifting domains. So for some of more complex specs that I'm building these days, I have had it write blog posts with animated versions of animated prototypes that will slowly step through the algorithm and say, hey, this is changing, and then this changes, and then this gets called. And it does animations. that's actually, I've only recently started doing that. It's pretty cool. Yeah.

Dalton Anderson (46:16.382) I'll write that down. That's interesting. I've never done that before. And that's the first I've heard about that.

Michael Daugherty (Quill) (46:18.734) Yeah, because you can ask for it to do it. You can ask it to do anything. So however, if you've ever seen something that really caught your attention and is and like you couldn't stop reading it or you couldn't stop listening to it, you felt you were learning lot, like try that format with the AI. the other point I was going to bring up is, is I think, you know, you mentioned AI is kind of an amalgamation of the knowledge on the internet, right? It is always a little bit out of date.

Dalton Anderson (46:46.308) Mm-hmm.

Michael Daugherty (Quill) (46:46.892) And so, and then its strongest knowledge is what is most commonly known. And if we're talking about knowledge, process wise, it's different, but knowledge base, it will know the things that people write about the most, the best. What's interesting is, I think always the frontier of knowledge is the most interesting. And that's kind of where startups always operate. And so you want to try to get to the frontier in something in your career.

It can take a long time. That's what people in PhDs do, right? They keep studying, keep studying and researching, and then eventually they get to the frontier and they move the frontier slightly. AI should help you get to the frontier faster, but you do have to realize it's not going to help you. It's not going to move the frontier by itself. You can't create a physics paper by just asking CHPT to write physics paper, but it can help. But if you're knowledgeable about like where you sense it's getting to the end, it can help you get there much faster, get the basics.

Dalton Anderson (47:29.688) Yes.

Michael Daugherty (Quill) (47:41.432) train you on the basics, and then suddenly you can write your own physics paper because you've learned all the formulas that people know about today. And you also can probe and understand, well, where is it not clear? That's probably actually an interesting area to do some research on my own.

Dalton Anderson (47:59.759) Michael, that's a great point, using AI to get you to the frontier faster. Like that's the goal in your career, trying to get to the frontier, frontier of anything you want in your life. Using that to be a catalyst for growth, knowledge growth, skill growth, to get to the frontier and push the frontier slowly, be an innovator. Whatever you're trying to do, try to innovate, which is awesome. Speaking of innovation is the, I guess,

Michael Daugherty (Quill) (48:09.262) Mm-hmm. Yeah.

Dalton Anderson (48:29.314) The natural progression from note taking to integration of like the agentic command center, like however we want to phrase it, I think you're still coming up with a moniker that you like to display externally, but like this agentic command center. And then the next part about these extensions you talked about, which isn't public and it's like a work in progress and it's not ready for production or anything like that. It's more of a thought process that we talked about in the show.

Michael Daugherty (Quill) (48:40.898) Yeah.

Dalton Anderson (48:58.424) this like natural progression of being the command center. How does this play in the mind of everyone at the company when you're competing against like Microsoft teams or Google meet? Whereas I know that there's different approaches that we talked about prior where there isn't a like ghost person on the meeting there. The models are local. They're not in the cloud. It's your data and your

providing a different approach and different option, but just wondering the thought process and approach from the competition. Like when I say competition, really like, they're not necessarily in the same sphere of competition, but they have a product similar, not necessarily similar, but they have a meeting, notice the product. Yeah.

Michael Daugherty (Quill) (49:36.876) Yeah, no, I understand. Yeah.

Yeah, right. It's okay. get it. Yeah. So, so meeting notes, think can be to your question, meeting notes can be a feature of a larger product and can be shipped by a company. I think there are a few things that you and I talked about. One is just going in depth in this, in a specific area is itself a differentiator. If you just ship generic meeting notes as a feature of your company, you're going to ship it. You're going to move on to the

you'll do your quarterly review, say, I shipped this thing and move on to the next product. And that's where you get basic meeting notes, get sent to everybody the same way. Whereas if you are thinking about that, this is our core data model. This is truly brand new data that people couldn't access even five years ago because you couldn't transcribe all this stuff and make it legible to a computer. What is unlocked by that? That's where you go down in depth and you say,

And you really think after meeting, what do I do? And it may or may not be viral, as viral as like always send everybody an email and like get into every meeting on their calendar. But you think about things like, how do I use this outside of Zoom or Google meets? Well, for that, you should probably be local. You should be recording so you can have in-person conversations, because those are very important too. So that you can have your calls on WhatsApp, because sometimes that's where the brilliant ideas happen.

And so to do that, you have to have a local app. can't be a feature of Teams where the goal of the product manager is to get more people to use Teams because you have to kind of meet people where they are. You think about should we have the meeting note taker? What level should it sit at? Is it the company level or is it each person has their own takeaways, their own actions are going to do afterward? And then kind of keep.

Michael Daugherty (Quill) (51:37.486) keep going down that direction. I do think a lot of companies are going towards this agentic space of like, how do we help you do work? We're all coming out of it from different directions. I think the meeting note taker direction is very interesting because the volume of knowledge that you get about a person by passively understanding what they're talking about is both so large and so comprehensive across

personal conversations, things you wouldn't ever necessarily write in email because you start the conversation, you like, are the kids and all this stuff. But that actually informs your worldview. And so it should actually know you better and can infer a lot of the other information and help structure that. It also makes it super sensitive. So that also informs the decision of, we think people, if they are really going to have conversations with all sorts of different people in their lives,

Do we want to centralize that and make it a risk that it gets hacked or whatever? And do we want to build our value on like storing people's private information? Probably not. People will actually want to use it more if they know that they have control over the data and the product gets better if they use it more. So therefore we should give them control over it and we should keep it on their computer. And then they can put more information into it and it should understand them better over time.

So I think when thinking about like large companies, they're often thinking much more incrementally and the startup needs to be thinking, how is life going to be lived five years from now? And what products are going to be important in that space? And like, how do we build them now? so it's a little bit different and also means you can focus in there and just go deeper and deeper than large company would.

Dalton Anderson (53:35.717) Well said, I think one of the points that you talked about was there might be two theories of thought is one is to iterate and make sure that the user uses the platform. And there's two ways to get that there's two levers. There's one to make it hard to get the data out. And there's the other process of, okay, we'll make it really easy. You choose however you want.

which makes people, it's like the whole concept of Obsidian, like why Obsidian is so cool. Like if you wanted to get rid of Obsidian and you wanted to go somewhere else, you can export all of your docs and then drop them somewhere else. Like they're just markdown files. You can put them anywhere. And so people are really open to using Obsidian and putting all their stuff in there and their workflows and getting super complex because they can just, they have full flexibility of it doesn't matter. They can use YAML, they can use whatever file format they want.

Michael Daugherty (Quill) (54:13.987) the

Michael Daugherty (Quill) (54:30.539) huh.

Dalton Anderson (54:31.428) And it's not a big decision and they like the company and like the thought process and the brand, but also, you know, things change over time. it's people like that flexibility to where it's not, okay, well, if I do this, I'm probably never gonna take my stuff out because I've got to export it individually, like one by one. And I'm not gonna do that for thousands of notes. So like, this is my life's work or whatever I'm working on. And it's stuck.

Michael Daugherty (Quill) (54:52.771) Zzz

Dalton Anderson (54:57.88) which freaks me out when I have those type of situations, which is more common than, when it's more common than not. And when I have to make those decisions, I'm like, I'd rather just not use them if I have the choice.

Michael Daugherty (Quill) (55:09.998) Interesting. That makes sense.

Dalton Anderson (55:14.788) but that makes me want to use the product more because then I'm like, okay, like now I can, I can go all in and if something happens, I don't know, obsidian goes away, which probably not happening, but I'm just using obsidian as an example, then it's fine. I could just drag drop, it, put it elsewhere. Or if I need some kind of capability, that's not going to be around. There's extensions that people build that are open source. So like there's this whole thing that ecosystem that allows you to utilize the product and then iterate on it either by a user or individual contributor.

I like.

Michael Daugherty (Quill) (55:44.736) It's definitely very different than building a pure SaaS company. In some ways, some ways harder. You have to work on many different computers and configurations and everything. But it does actually feel nice that something goes wrong with our servers. The app can still continue to work. People can point it to their local LMS. They can use it offline, whatever it might be.

If they need to, they can get the data out of the local file system and extract it and put it somewhere else.

Dalton Anderson (56:18.798) Do you have plans, and it might be a little question, but do you have plans to offer some kind of like cloud storage for your customers if they want it, like as like a free integration? Not free, but like an integration that you can pay for instead of storing it locally.

Michael Daugherty (Quill) (56:27.822) So, yes, yeah, Yeah, so actually we have this. What we do is we do end-to-end encrypted syncing. So if you have two computers, your data on one can go to the other. We're actually rebuilding this because the UI is difficult.

not going name names because it has happened to a lot of people, but people will come to us and they'll say, I love that it's super local, that you guys don't have any access. And then a week later, I threw away the old computer. Can I get all my meetings back? And we say, it was local. We didn't have access to it. And so, so we did build, build this, the syncing, system, but people have to remember to take their private key and take it to the next computer. We are in the middle of a big rebuild on this to make it,

Dalton Anderson (57:09.465) Hahaha

Michael Daugherty (Quill) (57:24.91) much more robust, it supports more data types and also a bit easier to use. So it's easier to recover for individual users and add additional devices, as well as share within teams. People do want to collaborate on stuff, but they just want control over how they collaborate on it. And so for the companies, for the enterprises, we are planning also to offer fully self-hosted versions of this. So even with the encrypted data, like if you want to keep it yourself.

Dalton Anderson (57:53.711) Like it's airtight versus airtight like in the cloud somewhere where you don't have access, you don't know what's in there. You're just handing the keys to the user and the user authenticates. Yeah, no, really cool. I like the aspect of being able to say, hey, some stuff I don't care if it's in the cloud, some stuff I don't want in the cloud. Whereas I can make the choice like, but then there's some positions in some companies that you're working at or say your easy one would be like intelligence or defense like.

Michael Daugherty (Quill) (57:55.062) Yeah,

This is

Michael Daugherty (Quill) (58:03.448) Yeah

Michael Daugherty (Quill) (58:22.83) Mm-hmm.

Dalton Anderson (58:23.2) Absolutely that stuff can not be in the cloud, that's local. But having the flexibility is cool.

Michael Daugherty (Quill) (58:26.51) And some of the companies need to absolutely know which AI models are interpreting their data, because they have got particular regulatory requirements on their industry. so being able to give them control over which LLM they use, where the transcription happens, even to the point of running everything totally locally, lets them feel confident that they

that they have control over it. That word is actually in some ways more important than local, it's just control. Like you get to decide.

Dalton Anderson (59:05.22) No, a hundred percent. I think this is like a wonderful episode, Michael. I thought it was quite intriguing and all the stuff that you guys are working on and going really deep into probably one of the biggest time sinks that people feel consistently. Like everyone has the complaint is like, ah man, like the meetings, the meeting notes, like coordinating follow-up, like it's just such a pain in the butt. I wish that there was a better way of doing this. And there's been attempts, the attempts are typically lackluster. So I think this is really exciting. And then,

Michael Daugherty (Quill) (59:21.816) Yeah, yeah.

Dalton Anderson (59:34.468) wondering if people feel that same excitement, how would they get in contact with you or Quill?

Michael Daugherty (Quill) (59:38.894) So quillmeetings.com, go there. You can also email at quillmeetings.com as me. it. Quillmeetings.com will probably get you a faster response. yeah, can email us. You can go to quillmeetings.com and download and try the app. And I'd be happy if you do. I'm very excited to hear about how people use it every day.

Dalton Anderson (01:00:06.885) I'll do that. I'll probably put your support emails like maybe the, maybe your LinkedIn or something instead of your email. Yeah. Yeah. But yeah, man, it's a great episode. Really appreciate it. It was quite interesting as I said before, and it, is the closing of the show. It is wherever you are in this world. Good evening. Good afternoon. Good morning. Thank you for listening and thanks. Listen in next week. Goodbye.

Michael Daugherty (Quill) (01:00:10.186) Okay. Yeah, LinkedIn is good too.

Michael Daugherty (Quill) (01:00:30.766) All right, thank you very much.