00:09:40 · Dan Rogers, Guillaume Moubeche
Are there some rules of the road that we need to be following? Are there some guidelines that we need to be following? Is there some governance that we need to be following? And then what I would call the harness, which is really actually making that work happen, pushing that work out to the right people, taking action, involving those people, and then the right moments making sure that they can prove things and that it's fully governable, and in the end, ultimately a repeatable process. So this is where you need something like Asana, which is the human agent operating system, which is really that enterprise ledger, that strata for humans and agents to collaborate off of off of the same living plan. [00:10:24] S1: So your view is down the line to kind of have like agents perform like as many tasks as possible and get all the context from Asana while always like keeping a human in the loop to validate like that the work has been done properly, that the project is going forward, etcetera? [00:10:44] S0: Yeah. That's a very nice description. Maybe I'll make it even more real. So today, as an example, we have a thing called teammates. These are our agent characters. We have 30 pre built teammates. They perform a particular package set of actions that you have scoped and defined. With AgenTic Work Management, which is really where our platform is headed today, we launched it today actually. So it's very, you know, lucky to [00:11:08] S1: be Congrats. Conversation.
00:11:10 · Dan Rogers
[00:11:10] S0: Where with AgenTic Work Management, these teammates make themselves available to you as you're working in Asana. So it might say, hey. I noticed you're working on a product launch. I'm the product launch assistant. Would you like me to help? Hey. I've noticed you're working on some pricing thing. Do want me to help? I... I'm like a pricing analyst. Or, hey. I noticed you're doing some company competitive research. Well, I'm your competitive research assistant. So we have 30 of these people agents they make themselves known to. But what's beautiful about Asana is because we have this multiplayer coordination, any of those agents can interact with many team members. Many team members can improve them, correct them, and say, actually, no. No. These are our competitors. Please refer to this knowledge base. Now that competitive analyst that's been maybe informed by two team members can be used by 30 other team members, and it knows who the competitors are. And as it's coordinating actions, it's doing so again within Asana. It's literally taking tasks, performing them, writing it back to Asana, letting the humans know what's happened. So suddenly, this is fully coordinated and fully contexted. And it's live and a living plan because everything it does helps the very next run of that agent. So you'll end up with this tapestry of these agents working deeply within a workflow. It's super important. It's super different than how to think about some AI bolt on feature or even interacting with a chatbot. This is real enterprise productivity. It is kind of a game changer from where we've been. Where we've been in this evolution was single player interaction with models. Good, you know, helpful from a personal perspective. This is not how any work happens. The work is that myriad of handovers, the myriad of crossovers, the myriad of little things that need to happen to keep things moving. So for example, maybe you wanna make a change to some software and maybe you have like a billing system and you want it to do some new calculation. Well, it turns out you probably need to inform your customers. It turns out it's gonna affect the way you recognize your revenue, you're gonna involve your finance team. You've got some engineers that need to work on it.
00:13:12 · Dan Rogers, Guillaume Moubeche
You probably have some folks, you know, product management needs to describe this new thing to your customers in turn, etcetera. So let's say there's like five or six parties just for a single change in your product functionality. It's really the agents that we wanna do a lot of the coordination. It's the agents we want to figure out. Okay. Well, this is probably the 25 tasks that are gonna need to happen. And by the way, I'm gonna need to hand it over in this way. And realistically, we should say this is gonna take six weeks. It's one of the agents you want to kinda create that based on everything you already The humans meanwhile, you want to say, yeah, I wanna approve this. I wanna take a look at this. No. This is the way we like to do this. This is the way that things are done here. This is our taste that we want Nida to perform. And then at the end when it's all finished, it's the human that kinda says yes, this met our specification. Great. And now we can do it even better the very next time. So that's kinda human agent operating system in action. [00:14:10] S1: And for the models that you are using, are you leveraging like existing models from companies or have you like also built in your own model directly?
00:14:10 · Guillaume Moubeche, Dan Rogers
[00:14:10] S1: And for the models that you are using, are you leveraging like existing models from companies or have you like also built in your own model directly? [00:14:20] S0: Well, as I would say, from a model perspective, they become quite interchangeable. So we're very focused on being really the kind of the quality and performance of those agents for our customers, which actually does mean certain tasks get routed to certain models based on their efficacy of delivering them. And so under the hood, yes, we have interchangeability of many of the models. But we also work with the foundational models because some people don't actually want to work in the canvas of Asana. They want to work in the canvas of their chat screen. And so for things like, you know, codex, for things like Claude, for things like etcetera, we have this MCP integration where if you are an Asana customer and you're working with one of those foundation models, you can actually view tasks, hierarchy, project, and you can talk to Asana really through those chat discussions, and not even have to come into Asana. And so we have a very nice relationship as they kind of fold into that tapestry.
00:15:19 · Guillaume Moubeche
[00:15:19] S1: The reason I was asking is because I think the q three results were actually like really great, and I think AI teammates from what I saw went from 17% to like 25% of net new ARR, which is huge and means like the adoption is actually like really, really nice. Also, what I've seen is like, you're putting AI in most, if not every paid tier without changing prices, which means that you are also absorbing like a potentially a token cost. So can you like maybe walk me through, like, because for me it's the same, you know, like obviously I built a software company, margins were pretty much like, you know, like insane, and with AI token, obviously you need to make some sort of arbitrage, but it feels like you've already managed to still get like a really good margin on top of the input cost of token. So maybe you can walk us through like your strategy regarding tokens and how you manage it.
00:16:20 · Dan Rogers
[00:16:20] S0: Yeah. So firstly, let's talk about the success of AI teammates. As you said, now 25% of our net new ARR is coming from AI products, which is AI teammates and AI studio. In fact, for our largest customers as well, most of 25% of the new deals that we do with them are AI attached deals to them. So, yes, we are seeing that people have a new way of working. They want to bring AI into their core workflows, their business critical workflows. So in long way that trend continues very good for Asana. But then a little bit about, you know, how that works from a token usage and and a margin perspective. What we decided to do, I think is very bold and maybe a first in the industry, so I'm sure someone would tell me if not. But we decided that we think the right unit of charging or unit of measure for how effectively AI is performing in these workflows is the fulfillment of a request. So we decided to create a new meter that we're calling request. Why is that bold? Most other, you know, SaaS models are charging on credits or tokens, and they say, well, to perform this is 5,000 credits. To do this is 2,000 credits. You're like, I'm not sure if it's gonna be more like this. Okay. Well, in that case, it's 3,725 credits to perform. Very hard to keep a running ledger a priori. So before you do it, of what tokens you're gonna use, what credit scheme is. And we kinda said, don't worry. We'll obfuscate all of that to the customer. All you need to know is if you've asked this agent to do something, to perform some action, and it performs that action, ping, that's a request. And so we think it's a much more natural way of describing your interaction with the agents. And so, yeah, we think we're very bold in kind of coming to that kind of meter and obfuscating all of that complexity under the hood. The customer... End customer shouldn't have to think about which model's being used, how many tokens it can... It's just have I got the job done. But that's bold. The second part that's bold is we kinda said, we've had six months of experience now with with teammates and with Studio. And what we realize is that customers that get to a certain amount of usage become very sticky. But the hardest part is getting them to that amount of usage because of discovery.