How Asana uses Asana AI teammates to automate PMO

Hi, everyone. I'm Zoraida Diaz, and I'm a director of product marketing here at Asana. And I'm super excited to walk you through how we use AI teammates to automate PMO workflows.

So as we march towards building an agentic enterprise, we need to move beyond the hype around AI and really talk about what it means to use AI in practice. So we know the opportunity is real, but we are seeing a gap between experimenting with AI and putting AI to work in your team's day to day work.

So we know that ops and PMO teams are built for scale, but the path to scaling with AI means starting small.

First, you need to prove the value and then you expand. So we hear that a lot of organizations, 88% are actually already using AI regularly, but less than 10% have moved beyond pilots. And that's the gap that we'll be talking about today.

So when we talk to ops and PMO leaders, we hear three questions prop up very frequently. Folks wanna know, what's happening across my portfolio? What should we take on next? And why did we make the decisions we make? So behind these questions is what we call the coordination tax. And it's all the work that our teams are doing aligning around the work. So it's like chasing status updates, corralling information, figuring out why key decisions were made.

So it's not about the work itself. It's about the work about the work. So that's a lot of times our teams are spending corralling information that could actually be spent on strategic work. So today, I'm gonna walk you through how our own ops and PMO teams, tackled some of these taxes that we talked about.

And behind these three questions, there's three different flavors of the coordination tax. There's a reporting tax, the prioritization tax, and the memory tax. So I'm gonna walk you through how our PMO team tackled the reporting tax, the journey they took to get there, and then the five step framework we landed on for deploying AI at scale. So let's get into it. How did we start? We started with a pilot and a problem.

So our head of PMO last year was thinking through AI teammates and how to integrate them into their team's major workflows like goal management and resource management. He decided he wanted to start really small with a specific problem to allow the team to learn what was working, what didn't work, and figure out where AI can make the biggest difference for the PMO org. And from there, we actually landed on a five step framework that I'm gonna walk you through today. And the one thing that I want you to take away as I'm going through all these steps is the key behind all these steps is giving your agents, your AI teammates the right context at the right time. So let's get into it.

The first step to deploying AI at scale is really all around identifying the friction point. So think about where is work slowing down? Where are the manual steps? Where is your time... Your team spending a lot of time doing just repetitive work? And choose a specific pain point that you wanna address with these agents. And the clearer you are with the problem you wanna solve, the easier it'll be to design an AI teammate that solves that problem.

So for our PMO team, goal management was a top priority, and QBR season was right around the corner. We all know what that means if you're an opt in PMO. That means stitching together status updates across portfolios, countless hours of manual work. So we really honed in on this specific problem we wanted to solve for initially.

Which brings us to our second step to deploying AI at scale. It's all around scoping the job. So thinking about a specific job to be done or a task that you wanna have your AI teammates take on first. And the clearer you are with the output that you expect, the better results you're gonna get. So really, as you're scoping the job, think about the work you want that AI teammate to do, who the output is for, what the tone should be, how should it be formatted. And it's really critical to also link that AI teammate. Give it the right context whether that context lives in Asana, add it to a project, or whether that context is really living in files in Google Drive or SharePoint, making sure you're connecting your AI teammate to those data sources.

So here's our initial goals agent that we built. So you can see that we added some behavior guidance. Its job was all around synthesizing complex work progress across different portfolios. We gave it guidance around, hey, make sure you're leading with status in your output. Make sure you're tying tasks that you're looking through to the goals that we have as an organization.

And then we wanted to make sure that the AI teammate was also listing any assumptions that it was making as it was producing the output, and we gave it some guidance on the standards of the output that we expected, which takes us to the next step. So now you've built your AI teammate. You're ready to go. Now it's time to experiment and learn.

So really let the AI teammate run, assign it work. And as you're assigning at work, really keep an eye on the output that you're getting from that AI teammate. And that's typically where we had our first moment because that's where you really see the AI teammate doing the work. And as you're looking at the output, this is also a good opportunity to figure out if there are any gaps in the way your team is actually capturing information.

So look for missing information, maybe inconsistent status updates. So really take this experimentation phase as a way to improve your team's overall workflow, but also refine how that teammate is working for your organization.

So these are the initial outputs of our Gold Agent. It was drafting great exec level updates cleanly in a Google Doc. It was give us... Giving us insights and next steps. So this is really where we had our first moment around speed. We're like, wow. This goal agent can put together a status update that would otherwise take someone hours to put together.

So we continued to use the teammate, but we hit some snags. So we saw the teammate was struggling a bit, and we were kinda scratching our heads. We were like, it's probably because we built the teammate wrong. So we did more testing and iteration, and we found out that the teammate actually didn't fail.

Oops. It exposed the gap into how we were capturing information in Asana. So our team wasn't actually putting all the goal information in Asana. So the team was starting to make assumptions, and that led to inconsistent quality output.

So at this point, we've realized that before expanding the use of AI teammates across our ops and PMO teams, we actually had to build the foundation, which brings us to our next step around setting the guardrails. So this is a really important moment before deploying AI at scale because you really need to ask yourself, where should there be a human in the loop? Where should there be a reviewer?

So think about the coordination heavy work that you want your AI teammate to focus on, and then think about where you want people in the loop. Where should folks be making the calls, reviewing the output, validating, making those judgment calls? And it's really important along the way that your team is also showing your AI teammate what good quality output is like, so giving it guidance on how the output can in... Improve. And also, this is a great opportunity to let your teammate, your agent know what to never do on their own without a human in the loop. So following these guardrails, setting these guardrails puts your team in control but also maintains the integrity of their work.

So for our ops and PMO teams, they typically get together and they look through all our projects and our initiatives and they figure out, like, what's feasible to take on the next quarter? What should we prioritize?

So as the team was thinking through how to best leverage AI teammates, they drew a clear line. So the teammates are gonna focus really on the information gathering, synthesizing information, really driving the output. So looking through the context across portfolios, handling that multithreading. While the humans are really gonna focus on the strategy, the planning, the decision making, figuring out what was most valuable for the team, making that final call. So they decided that, hey, our AI teammates are gonna focus on the context gathering, and we're gonna focus on the strategic work, the bigger picture, how to move to the next, path forward.

Which leads us to our next step, which is all around iterating and expanding. So you've set the guardrails and you are building trust around AI teammates across your teams. The next thing is to really figure out how you're gonna measure success for your pilot. So think about metrics that are important for you and your team. Is it around refining the output of teammates? Is it around really saving time?

Is it making sure stakeholders across your organization are engaging with your teammates? So really choosing a specific metric you wanna keep an eye on. And the goal of this is to figure out when your pilot is successful. And once you've hit that point where you're like, okay, I see these AI teammates are working for us, that's the time you can look across your workflows and figure out how to repeat that pattern, that success story that you have with that specific workflow into other workflows. So as our team expanded the use of our goals agent, we are looking into additional workflows that we could implement AI teammates for.

So our goals agent actually evolved to this MPR drafts teammate, which is used across all of our enterprise apps portfolio. So what this teammate does, it looks across all the cross functional portfolios and drafts summaries of the latest information on where work stands. But it also drafts talking points that our cross functional partners can use when driving decisions and planning ahead for the next half, the next quarter. So this teammate runs on a weekly basis. It drafts these summaries, and then we have a robust tracking system that tags who the right human is to bring in the loop and actually validate the talking points that this teammate has put together. So, again, this teammate is focused on the information gathering, that heavy lifting, and then everyone on the team can focus on, like, decision making and figuring out the strategy.

And since we've created this teammate, our team has looked across different workflows, and now the ops and PMO team uses AI teammates for all their decision making across, conversations with partners.

So I walked through the five steps we took for deploying AI teammates at scale. So we started small. We started with the goal management use case. We learned. We iterated. And then we looked at other workflows that we could expand this success story with. So I really encourage you to look at these five steps, figure out how you can implement them in your organization to solve for the coordination tax.

And earlier, I mentioned that there's three flavors of the coordination tax that we see cropped up frequently. So that reporting tax, prioritization tax, the memory tax. So I'm gonna hand it off to my colleague Jonathan who's gonna walk us through how we can use our pre built teammates for ops teams to actually tackle all these three, taxes.

Hi, everyone. I'm Jonathan Kim. I'm on our solutions tier... Team here. And like Zareta mentioned, we have these three taxes. We have our reporting tax, our prioritization tax, and memory.

And unlike what Zereda had just shown you just a moment ago where our PMO team put together these custom bespoke teammates where they gave it custom instructions and custom skills and all those things, What I want to show you today is how you can take those same concepts but apply them with out of the box teammates today. And so I want to focus first on the reporting tax, what it... Which is what is happening across my portfolio.

So what I want to first start off with is talking a little bit more about what I mean by out of the box teammates. So if you've ever created a teammate, you've probably seen this screen here before, and it's the place where you can go to find a teammate to tackle some of the work that you're trying to do. So it's organized based off of things like department and function. So we saw marketing, IT. I know here we're specifically talking more about strategy and operations.

So here I wanted to actually focus in on these four or these couple teammates here that we have, again, out of the box.

And so for our first tax that we talked about, the reporting tax, I actually wanna call the help of our Portfolio Advisor. And the reason why I'm calling out the Portfolio Advisor is because it does this. It monitors Portfolio Health, it flags missing alignments, and all these other things that are gonna help me understand how my portfolios and projects are going. So let me show you this in action.

So this happens to be our FY twenty seven strategic initiatives, and this happens to be a portfolio inside Asana. And since we're here, I think we all generally know what a portfolio is, and there's a lot of great benefits of having portfolios. You're putting and gathering all these projects together in one place so that we can have a lot of information such as we have dashboards at our fingertips that tells us the status of our projects. We also have things like timelines to see where our milestones are starting to line up.

With this, we also, even before diving into any given project, I already know kind of how things are generally going based off of my statuses and milestone progress. So, for example, I have a couple here that I want to call out.

We have, I think here, one called EU Data and Security Readiness. This project happens to be at risk. Our EMEA expansion project is on track. And lastly, our finance systems consolidation, that project is currently on hold.

So, again, this is all Corusanna stuff that you all should know about. We have plenty of data at our fingertips. The problem here though is that this is a lot of projects to have to actually go into and make sense of so we can actually have a meaningful discussion about these projects and how we should strategize going forward. So what our PMO leadership team decided to do is we decided to put together pre reads.

And so that's what this project here at the bottom is calling out.

We have these pre reads that we put together on a monthly basis. And one thing that you'll notice here is that this actually used to be done by hand. You'll see here that poor Martin here at the bottom... Oops. Let's see.

Martin Jones here at the bottom, he was doing it first, and then he handed it off to me, and I said, no way. I'm not doing this by hand again. Let me go ahead and get a portfolio adviser to do that for me. And so that's exactly what I did.

So I want to show you an example of this. It's October. I want to show you our October pre read. And so here, let me dive into what we actually did and what I prompted my teammate to do on a monthly cadence. Basically, I've told it, hey, go look at that FY twenty seven strategic portfolio. Take a look at every single project that's in there. Go ahead and sort of understand what's going on and bring out the three most important things that we as a leadership team should be talking about.

In our leadership team, we like to use Google Docs. Therefore, hey, put it into a Google Doc so I can send it out to everyone. And so that's what we have here.

And so as you can see here, my Portfolio Advisor did exactly that. It put together that Google Doc for me. Let me zoom in a little bit. I'm not gonna read this whole thing for you and I know even with the big screen, it's probably pretty small. But I did wanna call out a couple things here specifically.

The first thing here is that you might have remembered earlier, I called out a couple specific projects. One of them was our EU project that we said that was actually at risk, and then there was another similar project, the EMEA expansion project, and that one we said was on track.

The funny thing here is that our teammate did a double click, kind of went in deeper and looked one step further and said, hey, actually, this EU project that's at risk has some dependencies alongside the EMEA project. And so even though our project manager or even though the status update said that our EMEA project was on track, if we actually really think about it, these dependencies mean that the EU project being at risk also means that our EMEA project is at risk. Again, digging a level deeper. The other thing... The other example that I want to call out here is actually towards the bottom.

I'm going to skip the second one here. I'm going to show you this particular insight that this document is showing us is that it's not actually about one specific project, but actually across a number of our projects, we're finding that our security team is being tapped all the time and that there's actually a number of milestones and deliverables that are all starting to line up across the same time. It's calling out a few time frames here, in the in the notes. And so what this is ultimately calling out is that even though the status updates of those individual projects might say, hey, these projects look like they're on track, we know that if we don't address this sooner, we actually might start getting squeezed by the time the security team needs to actually execute against all of these different things.

So again, this is something that I was doing, Martin was doing beforehand, by hand. It was taking us hours. But now I'm actually just prompting our AI agent to do this for me on our monthly cadence and it's just ready for me to send out to our leadership team.

So that was our first tax. Now let me dive into our prioritization tax. What should we do next? So with the prioritization tax, I'm going to actually enlist the help of a different teammate, and this one's called the Program Ops Manager.

And so I'm using this one because as you can see in the description, it's good at managing PMO intake, end to end, routing, demand capacity, so on and so forth. So for this one, let me go ahead and show you this again in action.

So what I wanted to show you here... Oops, sorry. Went too fast. I want to show you this one in action here.

And back to the FY '26 strategic initiatives, we actually have this other project. It's the PMO Intake and Planning project. And it's probably very similar to many projects that you all have. An intake project where you receive new requests, then you start doing things like scoring. Hey, what's the strategic impact? What's the effort level? Which teams may need to be involved?

And then from there, we actually then can have a discussion around which of these things should we actually work on. And so very similar to what I had shown you before, this takes a lot of work for us to even identify and understand what these requests are even about, then after that identifying which of these we should actually work on. And so this is where... Actually, let me even call out a couple of these by example.

So first off, at the very top here under needs review, we have a request coming in from our CSM saying that they want an early notification, and this is a high strategic impact and low effort. Great. We love those kinds of requests. The second one here is that our executives want a new dashboard.

Medium impact, very large effort, so we'll think about that. And this last one here is that our sales team is asking for more support with their security questions from you all. So that's high strategic impact in a very large effort. And so these are things that we need to start weighing against each other.

So, again, I... What I really want to do is have our teammate be able to make that first pass judgment, first pass assessment on what we should actually work on and actually give us this recommendation. So similar to the previous teammate, I've actually tasked our program ops manager to give us that first iteration, that first draft on what should we prioritize.

So here, let me show you how that conversation went. I told the ops manager, hey, give us that analysis. And then it came back around and gave me a very thorough, lengthy response in the way AI agents tend to do. So as you can see here, it gave me a snapshot of all the work that's currently going on, all the recommendations that it wants to put forward, some other conditional discovery, things I should look into, other things to defer.

Honestly, this is way too long for me to read. I'm not gonna read all this information. And so just like like you might do with ChatGPT or anything else, I said, hey, this is helpful, but can you actually just make it a little bit simpler? And so that's exactly what I told our teammate to do in chat, just the same way that you might have a conversation with one of your coworkers or teammates.

And so like a good neighbor, this particular teammate went and actually did exactly that. So let me go ahead and show you the new response, and all it did was give us suggestions on what to move forward with, what I should possibly defer, and why it came up with those decisions. And of these things here, I actually want to call out one in particular, and it was the one about the sales team needing support on answering security questionnaires. And so if you remember from our previous conversation, the security team is already starting to get pretty tapped for resources.

And so from a teammate's perspective, this...

Deferring this project till later on makes perfect sense. Hey, security's already at capacity. We don't wanna add more to their workload by bringing in...

Bringing them in on a project that will require their resources. But I actually have some additional context. One of the things and one of the reasons for why our security team is so strapped for resources is because our sales team is constantly pinging them, asking them for these security questionnaires and these reviews.

And so if we can actually get them to be more focused on the strategic initiatives and actually take this off of their plate, they would actually have much more capacity and bandwidth to take on the strategic work. And so that's the context that I bring. The agent brought in a first pass sort of assessment. And so now the conversation continues.

I said, hey, I disagree. I think the security questionnaires probably should come up first for all the reasons that I just talked about. And of course, like like we mentioned, this conversation continued and ultimately, the the AI agent came back with a new set of recommendations that included the security questionnaires for our next pass. So that's what you're seeing here at the very bottom with that conversation.

So that's what I wanted to talk about about this particular text. And let's jump into the last one here, the memory text. And I think this one's pretty interesting because we all make decisions all the time.

Six months later, we know what the decision was, but we don't always remember why we made the decision, and hence we have the memory tax. So what I want to show you here is yet another agent that I want to enlist for support. So in this case, I'm gonna use the executive reporter. And I'm using this one here just because of that very last piece here at the description that says it helps with things like decision logs. And so let me show you actually a decision log that our PMO team is using today and how we're using a teammate to move this forward.

So here we go. The very last project in my portfolio. It's our PMO decision and assumption register. Basically, this is the amalgamation of all the strategic decisions that we've made across our different projects in our FY twenty seven strategic initiatives. And so as you can see here, we have not only decisions, we also have assumptions, we have open questions that we may want to come back and revisit.

The problem here that you can probably already imagine is this looks great, but who the heck is doing this? Who's actually gonna put in all the effort to put this together and actually maintain this? And that's actually where this is great as a teammate who can do this for us. So let me show you actually two examples of how we do this today.

The first one's actually going to be a little bit more explicit where I'm going to tell the teammate to go do this work. So to do that, let me show you a project that we have that's currently in flight. It's called the Enterprise AI Enablement and Governments Project. It's probably a similar project in your repository right now.

And what I wanted to call out here is that one of the things that we wanted to do or needed to decide is that we needed to finalize the scope of this AI pilot that we wanted to launch with. And so that's a decision that we need to make. And so that's in this particular project as a task. And so let me show you ultimately how we made this decision.

You can see here the task is assigned to Martin. And as I scroll down here, you can see that there's a conversation happening between Martin and Neha. Neha...

Martin thinks that, hey, I actually want to use customer customer use cases in our pilot. But then Neha actually wants to be a little bit more conservative, and she says, hey, maybe we should just stick with internal use cases only. The important part here is that they go back and forth. They discuss the pros and cons.

And ultimately, at the very end, they come to an agreement on what they want to do. And so Martin says, great. Agreed. Let's capture this decision.

And so he tags the executive reporter, says, hey, capture this decision. And so, of course, it's going to go ahead and do that. So you can see here the response that our executive reporter said, hey, I've captured this decision. This is the decision that we've made.

There's actually still an open question that we should probably review in about a month or so.

And it also does two things. This is inside that particular governance project. But of course, we're going to leverage my favorite feature in Asana, which is multi homing. And so we can see that this decision is also captured inside our assumption register here.

So as you can see, what we've now captured coming from the previous project now into the register is we know what the decision is, when was that decision made, who made those decisions, and other contexts that you want to capture.

So that's great.

But the problem here again is that this is called the memory tax. I had to remember to tag my teammate to go do that. And again, because it's the memory tax, I'm probably not going to remember. So what I want to show you is the other way that we can actually proactively look for these kinds of things as well by assigning this to my teammate.

So I want to call your attention here now to the very bottom. Hopefully you can see this. It's our section for decisions that are recommended for review.

These are ultimately... I have a decision and assumption here waiting for me. And basically, what I want to show you is our teammate is looking through our entire portfolio, that entire FY twenty seven strategic initiatives. It's taking a look at all of that work, the project, the tasks, the conversations that are happening there, and it's looking for things that look and smell like decisions that are made across the team. So here, ultimately, what I wanted to call out, I don't need to necessarily get too in detail about what this... I keep on pointing the wrong direction, sorry.

What this decision actually is about. But what I want to show you here is it's giving me a recommendation, why it matters, but most importantly to me is I want to understand where did this decision come from? So as you can see, this is coming from a multihone project. It's somewhere else in our repository...

Or sorry, in our portfolio. And as you can see, very similarly to the previous example, here's a conversation happening between Devin and Charlie, two coworkers, and they're having a discussion about how to make this decision, and ultimately, they do come to a decision here. What you'll notice is different here though is that Devin never tags the executive reporter. He never says, hey, capture this piece of information and log it some...

For me somewhere. Instead, they just made this decision, they had this conversation, and then they left. And so what I wanna show you again is that this ultimately showed back up, if I go back, in my decision register as something that I, as our PMO head or someone else on the team, should review as, hey, is this something that we actually do want to capture and log as a decision that's been made for our strategic initiatives? And so, like we've talked about, AI is doing the work of the heavy lifting of digging through materials, finding insights, and presenting that information to me.

But ultimately, it is still my responsibility as the human to make that decision. So what we have here at the very end is that this is an approval task. I can say approved. Yes.

This is ready. This is something that I've approved and identified as a decision that we want to keep track of and move with going forward. So with that, really just to wrap up is that I've shown you the three taxes and I've shown you three out of the box teammates. These are not custom bespoke teammates that I created custom, skills and custom instructions to do.

This is all completely out of the box. All I did was pick and choose the right teammate for me. And so ultimately, that's what I want to encourage you all to do is to go go back to your instance, go in, grab a teammate, feel free to use one and just see how it works because you're going to be very surprised at the high quality of the work that it's able to produce. And with that, if you have any other questions or you want to come talk to one of our amazing solutions or product people, come visit us in the demo zone in the Expo Hall.