Asana CPO on Why Every Employee Is Now an AI Eval and What That Means for How Enterprises Actually Capture AI ROI | Arnab Bose | E302

Cold Open Arnab Bose | Asana 00:00:00 If you take a look at all of the research that came out last year from Goldman Sachs and McKinsey about real productivity gains in the enterprise after AI spend had gone up, all those results came back at zero percent. Even if you start out with a shallow graph, if you can create this loop where it's proposing ideas or proposing specs or proposing PRs that are terrible, but you're, as a human being, giving it feedback and you're correcting it, you will get way, way, way faster every week, every month, every year.

We have made AI agents that are powered by the latest frontier models. They're not Carlos's agent or Arnab's agent. They're a teammate. They can automatically and proactively take action.

If you take a look at actual business performance of SaaS products, companies like Figma, Atlassian, Datadog, they've been having phenomenal quarters to date. So the actual end user utilization of these products hasn't tapered off in that dramatic way that investors have been thinking about. We brought PLG into product and made it a core job to be done with a general manager holding a revenue number inside product reporting to me. Because for us, PLG is an acquisition funnel, where even if you are a Fortune 500 company...

Introduction Carlos González de Villaumbrosia | Product School 00:01:00 Hey, this is Carlos, CEO at Product School and your host on The Product Podcast. Today's guest is Arnab Bose, Chief Product Officer at Asana. Asana is the work management platform built for human and AI collaboration, trusted by over one hundred and seventy thousand customers, including Accenture, Amazon, and Anthropic. The platform's Work Graph, a proprietary data model that maps goals to portfolios to projects to tasks, is the foundation for Asana's AI Teammates: collaborative agents that operate inside the graph, learn from human decisions, and compound their intelligence with every cycle.

Arnab joined Asana from Okta, where he served as Chief Product Officer and led Identity Governance from launch to over one hundred million dollars in ARR. Before Okta, he held product leadership roles at Salesforce and Microsoft. In our conversation, we cover why enterprise AI spend keeps returning zero productivity gains and what is structurally breaking the loop. Why every employee approval, correction, or rejection of AI output is training data that makes the system smarter over time.

How Asana wires its own processes through the Work Graph so that AI decisions write back automatically and compound rather than reset. Moving PLG, forward-deployed engineers, and AI agents all into the product org, each under a general manager who owns a revenue number and reports to the CPO. Why the future of AI at work belongs to whoever has the richest shared context, not whoever has the best model. Let's get into it.

Asana's Beloved Product in an Unloved Category Carlos González de Villaumbrosia | Product School 00:02:00 We just set up a podcast studio in about 30 seconds. I feel like we just need a plant or something to make this cozy. But let's just pretend it's you and me, Arnab. Well, first of all, welcome to The Product Podcast.

Arnab Bose | Asana 00:02:10 Thank you. Thank you for inviting me. It's so exciting to be here. Look at this full house.

Carlos González de Villaumbrosia | Product School 00:03:00 Pretty impressive to be right after lunch with a packed house of 1,500 product leaders here. I think part of the reason is that Asana is a very loved product in a very not-loved category. Tell me more about where you think that love and connection come from. Arnab Bose | Asana 00:03:15 Asana's been around for a long time.

It's a product that's been available for about 17-plus years, and the founders have this overarching mission and vision of eliminating the work about work, and doing so in a way where the end-user experience for the people interacting with the product is truly delightful. Things like performance, speed of updates, ease of use, easy to get started. That design ethos has carried forward to this day. I've only been at Asana for about seven months, and it's been phenomenal to be at a company where the product has so many end users: the project managers, people in IT and operations, people in marketing operations who use it on a day-to-day basis, and that's how their life runs.

Those businesses, those critical workflows run on Asana, and that performance, that speed, the delightful experience keeps people coming back. Carlos González de Villaumbrosia | Product School 00:04:00 As a fun fact, I actually used Asana for the first time back in 2010. The co-founder is one of the Facebook co-founders, right? Arnab Bose | Asana 00:04:05 That's right.

Carlos González de Villaumbrosia | Product School 00:04:06 It was just the beta product, and obviously it's incredible to see you're even a public company now. And you joined as CPO less than a year ago? Arnab Bose | Asana 00:04:12 Yeah, I joined in September of last year. Addressing the SaaS Downturn Carlos González de Villaumbrosia | Product School 00:04:15 I think we need to address the elephant in the room.

For people who work in SaaS, they're like, "Oh my God, the SaaSpocalypse." What is happening from your perspective? I see that Asana, even though it's a great product, is 50% down year to date. Arnab Bose | Asana 00:04:30 The investor sentiment in the entire SaaS market is fully risk-off, and that makes sense.

If you're an investor and that's what you do for a living, you're seeing all of the innovation that's happening with the foundational models and the amount of productivity you can get out of Claude Code or Cowork or Codex, and you have to be judicious about how you invest money. So that's one reality: it's a risk-off reality from an investor perspective. But then if you take a look at actual business performance of SaaS products, we were talking in the green room about companies like Figma, Atlassian, Datadog.

These are products that deliver real value, that ensure 99.99% uptime reliability, that ensure enterprise-grade governance, security, and trust. They've been having phenomenal quarters to date. So the actual end-user utilization of these products hasn't tapered off in that dramatic way that investors have been thinking about, because there's legitimate value that they're bringing to real-world business workflows.

The third thing I would say: if you think about what is the structure of AI winners versus losers, today all the investors are focused on the AI infrastructure layer because they're kind of guaranteed some amount of returns if you invest in picks and shovels. But then if you look at what value can be provided on top of those picks and shovels, the value and the job to be done hasn't dramatically changed. The human beings in businesses, brick-and-mortar businesses, online businesses, they still have the same business workflows.

So the question is, how fast can companies evolve to take advantage of the AI tailwinds to deliver better, more meaningful experiences to customers? And I think that is the challenge for companies like us, like Asana, going forward. Where Is Asana's Moat as LLMs Proliferate Carlos González de Villaumbrosia | Product School 00:07:00 I think part of the narrative for the market to not always be bullish is the fact that there are a bunch of LLMs, and they're promising that they can, in some cases, replace the functionality.

Curious to know your perspective on where your moat is now as you also start integrating with some of those LLMs. Arnab Bose | Asana 00:07:15 I spoke at the Code with Cloud event in San Francisco a couple of weeks ago, and Anthropic themselves had a detailed guide for what they want app builders and developers to focus on versus what those developers should not focus on. Their guidance was: anything that brings the unique business context of the job to be done that you're focused on.

Asana is focused on ensuring that human beings and AI agents can work harmoniously together to supercharge business-critical workflows across marketing, IT, and operations. So anything that brings that enterprise context, that enterprise brain, to the front door of the AI agent, and does it in a way where authentication, authorization, trust, uptime reliability, and the end-user human interaction experience are figured out. They themselves are saying that that is a compounding benefit.

That is an investment that will ensure that your app constantly gets better over time. Things that are more focused on the inner workings of the model itself, like routing rules or fine-tuning, those are the things that are compensating for gaps in the current model. That is their messaging already to the industry, and I actually agree with it. The interesting thing that got me into Asana in the first place was that even in September of last year, working at Okta previously and seeing a phenomenal uptake of non-human identities in the enterprise, it was clear to me that LLMs and AI products would upend the system.

The interesting thing about Asana, and you've seen this yourself having been a user starting from 2010, is it's designed to be this enterprise context graph where you can go from a mission or goal to a portfolio to projects to many tasks. Filling out that context graph helps you define any sort of end-to-end business process. And that, to me, is an amazing way to bring the right data of who does what by when and how to the front door of agents. When you bring that agent into the system, you're going to supercharge that entire compounding process of moving things faster, learning things faster, completing things faster.

Building for Enterprise and SMB Without Losing Either Carlos González de Villaumbrosia | Product School 00:10:00 I hear you talk about enterprise, and super interesting because when I first used your product, I wasn't at an enterprise company. I can imagine that you still segment your product across different types of businesses. So as a product leader, how do you go about actually building in ways that a large enterprise feels like you are catering to them while an SMB still feels the value?

Arnab Bose | Asana 00:10:15 When you think about the kind of work that we do, which is ensuring teams can be highly productive, a team within a marketing organization in an SMB is working in roughly the same way as an enterprise. The differences are probably around integrations with third-party sources. Where does your data live could be different: it could be more SharePoint or Office 365-oriented in the enterprise versus Notion or Google Drive-oriented in SMBs, plus compliance certifications, things like that.

What we are trying to do is ensure that the end-user experience is as delightful as possible and takes that ethos of what it would mean to make an SMB successful. And then as we explore up-market opportunities, we consciously choose what is the right ROI for investing in those incremental integrations or certifications to unlock serviceable market. PLG Inside Product with a Revenue Number Carlos González de Villaumbrosia | Product School 00:11:00 PLG is still king, right? People get to value relatively fast, and then from there they don't need that much interaction with the sales team.

But if you go enterprise, at some point you also have to do some sort of custom integration, and forward-deployed engineer is a hot term, still kind of unclear what that really means. How do you apply that type of go-to-market motion for an enterprise so you can help them get to value faster? Arnab Bose | Asana 00:11:30 I'll answer this in two different ways. Within the product organization at Asana, we've adopted a slightly different organizational structure and areas of responsibility than historically.

The entire PLG team has moved into product, including pricing and packaging, the engineering work required to do the experimentation, and the product management resources. We've empowered them with AI-first tools to be able to glean insights and analytics and run experiments faster. So we've brought PLG into product and made it a core job to be done with a general manager holding a revenue number inside product reporting to me. We intentionally made that choice because for us, PLG is an acquisition funnel.

Even if you are a Fortune 500 company, a lot of our Fortune 500 companies started off being a 5 or 10-person team just signing up to get their work done, and that helped us expand later. The second thing is for PLG, we also made an intentional choice. In this day and age where you can get so much done with tools like Cowork and Codex, as an operational workflow company it is difficult for us to provide value to extremely small businesses. If your business is 10 people or smaller, you would probably be able to hack around the limitations of spreadsheets and email and Slack with these agentic tools.

But when it becomes slightly larger than that, like if you're truly an SMB in the sense of a 50 or 100-person company, then it's a sweet spot for us. So we've intentionally focused the PLG business in that area where we're looking at 50-plus all the way to Fortune 500 companies because that's an acquisition funnel for us. FDEs, GMs, and the New CPO Org Structure Carlos González de Villaumbrosia | Product School 00:13:00 So is that a separate business unit for you? Arnab Bose | Asana 00:13:05 Yes, that PLG GM is a separate business unit reporting into the CPO.

The second thing is for our newer products, and if you're here physically with us today you can go check out the Asana demo booth outside. You'll notice we've got newer products which are all about AI agents and human beings collaborating with each other. In an enterprise, that requires a forward-deployed engineer to figure out what are the right integrations for that AI agent, what should the intake for that be, because that agent can get trained with behavior, with external data, things like that.

And so we have an incubation team of AI specialists who are forward-deployed engineers reporting up to the general manager for Asana AI, who reports up to me. We've brought that FDE talent into the product org as well because not only do they help us deploy our first sets of customers, all the learnings get fed directly back into engineering, and they are way faster at helping us achieve product-market fit than if they were a separate team in the revenue organization who talk to us once a month.

They're literally sitting with our PMs side-by-side, helping us sell, close the first few deals, and helping those customers adopt. Carlos González de Villaumbrosia | Product School 00:14:30 That's a pretty unique org design from what I've seen. You as CPO have multiple GMs owning P&L and reporting to you. So how many GMs do you have?

Arnab Bose | Asana 00:14:45 We have a GM for Asana AI, we have a GM for our PLG product, and we have a GM for another product that hasn't yet been announced and is secret. I can't tell you what it is, but it'll be announced shortly. Empowering PMs and Designers to Ship Like Engineers Carlos González de Villaumbrosia | Product School 00:15:00 As we go layers down, we were talking this morning about the intention to collapse managerial roles to empower even managers to build. How are you thinking about structuring your own pods and empowering non-engineers to ship?

Arnab Bose | Asana 00:15:10 The way we're doing this is I'm encouraging our product team as well as the FDEs in two slightly different ways. For the product team, I want them to go fast. PMs and designers are going faster in two areas: one is going from ideas to prototypes and then suggestions for the engineers to review and ship, and the second is addressing voice of the customer issues. The thesis we have at Asana is we want to use this data structure we have, the Work Graph: mission, goals, portfolios, projects, to build a self-learning brain for the enterprise, where as you contribute ideas, decisions, and approvals back into it, AI can operate faster on that and help you make decisions and get to outcomes faster.

That's the macro goal. One of the ways you can manifest that for voice of the customer is when we get feedback requests from the field in Slack, or it comes directly to us as an Asana task, we can pick that up, put it into a queue for one of our AI agents to review based on all the other pieces of feedback we've been collecting, prioritize it, and assign it to a PM for their taste-making. If the PM says, "Yeah, this is my final proposal for it. Let's see if we can fix this issue," we have an agentic AI coder that is connected via MCP to that task inside Asana that can then go build out the PR, and then it gets routed to the engineer for review going forward.

These kinds of flows are possible because there's a ton of data in that context graph about historical decisions and things that are in the product backlog, and that context graph is constantly being kept up to date. The who does what by when and how is being constantly kept up to date. One more step beyond this: yes, we are doing this right now to improve product building and eliminate some of the tax that engineers were paying in the past where they were personally having to make decisions and pick up those tickets.

Now they're just getting PRs to review that have been approved by the PM from a strategy perspective. You could apply that same process to a number of things. If you're thinking about marketing operations and campaign launches, having a context graph that's up to date and plugging an AI agent into that means that you can go from an intake form to a campaign brief to a preview of your website to finalized copy that your CMO loves, way, way faster. And that keeps compounding because you can take those decisions and feed them back into the compound graph.

Why Enterprise AI Productivity Gains Are Still at Zero Carlos González de Villaumbrosia | Product School 00:18:00 To me, that is the promised land for non-engineering teams, because we've seen engineering teams using shared code bases and collaborating across technical tools to ship code. But as we think about the rest of the world, the 99% of people who don't code, the work has still been very fragmented. You have your to-do list, you have your document, but there's no shared place where you can actually build stuff.

There's a shared place where you can communicate, there's a shared place where you can keep track, but ultimately being able to orchestrate across different tools would be a huge unlock. So especially for a company that is measuring productivity gains, how do you actually prove that all of this orchestration and these new ways of working drive value for the business? Arnab Bose | Asana 00:19:00 If you take a look at all of the research that came out last year from Goldman Sachs and McKinsey about real productivity gains in the enterprise after AI spend had gone up, all those results came back at 0%.

And the reason why this is happening is because what people are getting access to is this super genius genie in Claude or ChatGPT, but they're chatting with it. Each individual is copy-pasting data in and out, but then all of the rest of your processes are still gummed up, and none of those decisions are making it back into a context graph that will create that compounding benefit. Within Asana, we are all bought in on the Asana Kool-Aid. We run all of our processes in Asana, whether it's customer success or escalations, executive briefings or product launches.

With the product launch stuff, what we've been able to do is take our Chorus or Granola meeting recordings, pump the transcripts into Asana, which gives us more real-time data. So let's say you're doing a stand-up about a PRD review. There's a PRD that's linked in Asana. There's the call notes linked in Asana.

You can go back to the Spec Writer agent and ask it to revise the PRD based on it. Then you can ask the agent that is doing your launch planning to break up that PRD into tasks that the coding agent can pick up. And all of these things, as you're evaluating them, human beings are becoming like evals, right? We are evaluating the quality of the output.

As you say, "Yes," "No," "Approve," "Change," that's feeding back into the shared memory for the agent that gets written back into the context graph. Carlos González de Villaumbrosia | Product School 00:20:45 Automatically. Arnab Bose | Asana 00:20:46 Yes. Because it's writing it back into Asana, right?

It's writing it back into that graph database, and so the next time you run it, it's already got all of those latest decisions inside. And if you were using these tools in 2016, like Jira, Asana, Trello, Salesforce for CRM, the big challenge at that point was getting human beings to record all of these activities inside the system. No one wanted to do it. And then even if it was recorded, a human being's ability to go and read all that data, grok it fast enough, and come up with the next best action is limited just by time and how much context we can hold.

Whereas AI agents don't have that problem. In fact, they love this. Even if you start out with a shallow graph, if you can create this loop where it's proposing ideas or proposing specs or proposing PRs that are terrible, but you as a human being are giving it feedback and you're correcting it, and that correction is getting automatically recorded, you will get way, way, way faster every week, every month, every year. The Compounding Context Graph: Headless vs.

Multiplayer Carlos González de Villaumbrosia | Product School 00:22:00 Salesforce recently made this announcement that they are going headless. They're exposing their data through MCPs, a well-structured API, CLI, and kind of allowing other interfaces to harness that data. How are you thinking about that for your own product? Arnab Bose | Asana 00:22:15 We have thought about two different modalities of work.

One is we want to express the Asana Work Graph in as rich a way as possible via MCP and MCP UI into all of the apps. We have an Asana app for Claude, we have an Asana app for ChatGPT, Google built one for Gemini. And I think that is amazing for individual productivity. We want to meet people where they're at.

Because I want to drop the tax of collecting the information and putting it back into Asana. What I mean by that is, today, if you have Claude hooked up to Asana, Gmail, and other apps, you can say, "Hey, what are the top things I should focus on today?" and based on your email inbox, "I don't have these things tracked in the right project." You can just talk to Claude and it will update the project data for you.

I'm fully sold on that particular flow. The other flow that's very interesting is what happens for multiplayer use. How do you ensure that all that knowledge and memory is not stuck in your Claude instance, but can actually operate inside the graph itself and work across multiple human beings? AI Agents as Teammates Inside the Work Graph Arnab Bose | Asana 00:23:15 We have made AI agents that are powered by the latest frontier models, and over time you'll be able to bring your own agent inside of that, where they operate as individual actors within the system.

They're not Carlos's agent or Arnab's agent. They're a teammate. And you can think about what is the role-based access control, or what projects or portfolios that teammate should be assigned to. And when they're there, they watch it like a person on your team.

They can react to a question from you, they can react to a question from me, they can react to a status update, they can automatically and proactively take action. Those are the two different modalities of work. One is, how do you get data in and out of the context graph as fast and as efficiently as possible? Headless is good for that, MCP is good for that.

The second is, how do you exploit the context graph for multiplayer, near real-time collaborative work between humans and agents? And that's when you work inside Asana. Carlos González de Villaumbrosia | Product School 00:24:00 Arnab, thank you for your time and giving us a sneak peek into life after the SaaSpocalypse. Arnab Bose | Asana 00:24:10 Of course.

Happy to be here. Thank you for having me.