Gaurav Agarwal, COO of ClickUp, on how AI is redrawing the competitive map in productivity

We last covered ClickUp in October 2024, when Chief Business Officer Tommy Wang made the case for the all-in-one workspace as the antidote to SaaS sprawl, bundling tasks (Asana), docs (Notion), chat (Slack), and whiteboards (Miro) into a single app.

With AI agents now redrawing the competitive map in productivity, we reached out to ClickUp’s president & COO Gaurav Agarwal to learn how agents are reshaping competition, pricing, and go-to-market across horizontal SaaS.

Key points via Sacra AI:

You joined ClickUp in growth and led growth, and now you're president & COO of the company. We'd love to learn how your concept of the business, and your frameworks, changed as you went from optimizing acquisition and growth to owning all of the operations across the company.

I actually think my growth background makes me one of the more effective operators out there, and I can tell you both what's changed and what new perspective it's brought.

A lot of growth is about finding the truth as fast as possible through qualitative and quantitative insight, then acting on it quickly. The best growth operators build feedback loops and testing loops, and running a good business, whether it's sales led or otherwise, comes down to uncovering new surface areas quickly and then systematizing them.

When I took over the broader remit, not just the product led growth side but everything, what helped most was that I'd been trained to look for incremental wins quickly. Marketers spend a lot of time debating attribution: what created the value, was it a Facebook ad, a YouTube ad, and so on. About ten years ago, cutting edge companies like Netflix concluded that attribution is a myth and what matters is incrementality, the causal impact of the work you do, measured as rigorously as possible.

That shifts your frame of mind: doing things for the sake of doing things isn't enough, you have to do things that create incremental return. Once you can measure that, you can also understand the diminishing returns of doing something. This is true of everything in life. The first few things feel hard, then it gets easier, but that doesn't mean you keep getting incremental returns forever. At some point there's a diminishing curve, so you need to know when to stop doing something.

Most of the SaaS world has operated in a margin rich environment for the last fifteen years and built bad habits, doing more because they can, without an underlying framework for incrementality. What helped me in the broader role was understanding the true marginal impact of a given action. You do that through testing, or sometimes you just ask customers directly which of the ten people reaching out to them they actually find valuable versus just nice to have. Just because you can staff an account with an SDR, an AE, a CSM, a services person, a success person, and a support person doesn't mean you need all of them.

The biggest thing that helped me was already being trained in that school of thought: find incremental value, scale it to the right amount, and do it fast. That philosophy scales across the entire business, sales, success, services, everything. It let me cut fat quickly, reduce unnecessary overlap, and invest resources where we saw marginal improvement.

What I also learned in the process, has helped me evolve as a leader significantly. Deep within, I'm a data and programmatic person. I'm used to fast feedback loops where a deployed experiment is clean because a machine can replicate it with a high degree of repeatability. Humans are different. A human led system isn't as agile as a programmatic one, so change management takes time, and experiments are less clean because variance in execution increases. Did the experiment fail because the play was wrong, or because we didn't execute it well? I've built a much deeper appreciation for change management than I had before and how important it is to get buy-in from the teams. When you work only with machines, you ship code or a campaign and it's done. When you're training hundreds of humans, change management and habit building take time.

That becomes a more important skill as we grow, because AI will eat up the programmatic layer, and in a world full of AI slop, delivering great human experiences on a repeatable basis, using a real playbook, will be the edge. It's not easy, but I think that becomes a bigger moat going forward.

On AI and the agentic push: last time we spoke with ClickUp, in 2024, revenue was roughly half sales led and half self-serve, with enterprise already about a third of the sales mix. Has the AI and agentic push changed who's landing at ClickUp, or the persona of the buyer?

Our mix is more or less the same, maybe a bit more sales now vs self serve, but that's mostly because sales harvests accounts out of self serve. Our best accounts naturally graduate from self serve to sales assist, that just happens on its own.

The agentic push is interesting because ClickUp doesn't just have a technology-only customer base, that’s less than 20% of our revenue.. We're not one of those SaaS companies that sell to other tech companies, which is most of the SaaS ecosystem. We sell to businesses of all shapes, sizes, and verticals, so we haven't seen AI shift the mix as much, though early adopters in tech are naturally using it more.

Within the customers we serve, there are two types. One is at the edge of AI: they use MCP, they want to connect hundreds of MCPs into ClickUp, they're building agents. The other type says, don't talk to me about AI, I just want good project management, and thinks AI is still just a chatbot. Because we serve both ends of the market, we do see a divergence in how it's evolving. In some ways that's good, because it pushes us to build services, software, and capabilities for both ends, which makes us more robust. But I also think the people who are behind will catch up quickly. This wave is moving fast, but AI is still a very new thing outside the Silicon Valley bubble, and people are still wrapping their heads around it.

Are there wedges in the project management space that AI agents unlock, in terms of getting usage and getting into organizations, that weren't there before just because of the nature of the technology? By wedges, I mean a use case or feature that gets uptake quickly and lets you spread your footprint within an organization, basically an accelerant.

Yes and no. What you find with horizontal SaaS like ClickUp is that we don't have a single dominant ICP. Our biggest ICP is only 15 to 20% of our mix. We're used by marketing, creative, product, engineering, HR, facilities, operations, services, you name it. Any piece of modern knowledge work is effectively a project or a workflow: you have collaborators, an objective, a goal you break into smaller tasks, dates, risk, and so on.

I say that because ClickUp customers run the gamut, from field services teams to manufacturing assembly lines to creative shops. Some of the biggest agencies in the world run multi hundred million dollar campaigns on ClickUp. Some of the biggest Pharma companies use us for R&D and drug discovery. The use cases are so varied that there's no single wedge that works for every customer. But the underlying themes of these wedges tend to look similar: how do I organize work, how do I stay on top of it, how do I report what's happening, how do I allocate resources well, where are the idle or maxed out resources, what's the risk. That's the base layer that's relevant across everyone.

What's more interesting is that ClickUp's AI has gotten powerful enough that I've built my own agents to triage my email, handle scheduling, and run my own CRM. I use ClickUp for all my marketing use cases. We've built integrations with some of the best coding agents out there, and we've automated a lot of our own data analytics on ClickUp: we analyze data, find a bug, and a PR gets created straight from ClickUp into GitHub. That base layer of project management is working well with AI out of the gate.

What's more interesting now is that our AI is starting to do the work itself for customers. Customers are building agents that do end to end work with a human just reviewing the output, and that's where we're seeing a real explosion, not just in generic work management use cases but in things like reviewing legal documents. One of the biggest financial services companies in APAC uses ClickUp to process claims and check for fraud. So we're seeing a verticalization of AI on top of horizontal software. For each industry, we end up building a wedge highly relevant to that industry. I've built a personal CRM for myself that helps me stay in touch with peers, new recruits, investors, and board members, and that's good enough software that I could package and sell it.

So there's a horizontal set of features everyone can use, but I think what's coming next is an era where a horizontal platform with deep, fine tuned AI ends up beating vertical SaaS.

To dig into that a bit more: increasing footprint in a specific vertical must be something ClickUp thinks about. Are there verticalized industries where the workflows and context are specialized enough to resist a horizontal tool like ClickUp? Commerce, for example: Shopify is launching a lot of productivity tools, agents, and an app builder within its own platform. Given the specialized context of commerce, one argument is that it would resist a horizontal project management tool like ClickUp. How would you react to that?

Of course, I have a horse in this race, but here's what I think will happen. Shopify is best positioned to build the best vertical AI for ecommerce. But do people have to log in to Shopify to access that agent? I don't think so. Our goal is to become the harness for modern work, and part of that is making sure the best vertical AI is available within ClickUp so people don't have to leave it. I can see a world where Shopify's ecommerce website builder is available in ClickUp as an agent. We'd be tapping fully into Shopify's vertical expertise, but customers wouldn't need to leave ClickUp for it.

Your context lives in ClickUp: you know what your business is, your brand guidelines are there, your previous creative work is there, the experiments and analysis from that work are there. ClickUp connects to your meeting notes, it connects to HubSpot so you understand who your customers are, it connects to your inventory management system so you know how much inventory you have. I think we'll end up in a world where all of these vertical capabilities are available across different horizontal products, but only a few horizontal products will win, specifically the ones that can orchestrate across multiple workflows and models, deliver the right value at the right cost, and manage context and memory well.

So yes, deep vertical AI survives, but you won't have to log into it separately. You'll get it all through one interface that understands the work and lets AI handle the bigger picture. Context becomes king, and even vertical AI needs horizontal context to be more successful.

That's one part of it. The bigger risk, I think, is this: is Shopify sitting on proprietary data that horizontal labs don't have? If so, Shopify's vertical AI has a real advantage. But Anthropic's Claude has gotten so good that, four years ago, you had to prompt it with "you are a senior marketing person, this is what marketing is," and now you just say "write an email" and it knows how to write a great one.

Horizontal AI keeps getting better and keeps gaining access to more private, proprietary data over time. If that continues, the biggest risk to Shopify's AI isn't ClickUp, it's Anthropic. We don't care whether it's Anthropic or Shopify's own AI; we just hope ClickUp becomes the harness so that whichever AI wins has the best possible context to deliver the best outcome for our mutual customers.

Speaking of horizontal tools: everyone in the space, Notion, ClickUp, Monday, Claude Cowork, OpenAI Codex, is now saying they're the place you'll do work with agents and where your context lives. How do you map the fault lines between these tools, and what's actually different under the hood in terms of differentiation?

I have my biases, but I believe them to be true, otherwise I wouldn't say this in an interview like this. There are two levels to this.

The first is the category of collaborative work management: software tools from the past focused on getting more work done. ClickUp, Monday, Asana, Wrike, Smartsheet, we all used to play in that turf. What happened is that a lot of these solutions became narrow project management tools, because holistic work execution needs more than just tasks, it needs communication, it needs ideation tools like whiteboards, documentation tools like docs. ClickUp is the only tool with the broadest set of primitives needed for end to end work management. Otherwise, you run your projects in Monday, your documentation in Confluence, and your ideation in Miro, and AI hates that because it wants context in one place. Within that incumbent category, we have an edge because we have the broadest set of work tools, and we also connect to the broader ecosystem. We're one of the few platforms mature enough, with enough work primitives, to stitch work together end to end and keep context in one place. We don't transfer context from Monday to Notion to Wrike to Miro, it stays in ClickUp and gets better over time. Take Slack, for example: teams waste an enormous amount of time trying to keep Slack conversations aligned with their knowledge base. At ClickUp, our AI does that alignment dynamically, because AI performs best when all the information and context is accessible in one place.

The second level is that the battleground has changed in recent years. We're not just competing with those older, more archaic platforms anymore, a new competitor has entered the space: Claude Cowork. I used to use it a lot, and ClickUp's AI has gotten incredibly good over the last few months too. The main difference is that I don't have the work primitives I need in Claude. I can use Claude as a thinking partner or a research partner, but if I want to coordinate something like this interview in Claude, I'd need to create a task, assign it, hide it from people who shouldn't see it, and build an entire workflow myself. I could write code for a few workflows, but if I'm running a thousand person company, I have thousands of workflows running across different departments at any given time, and those workflows need primitives: approval chains, tasks with statuses, documentation of decisions, docs for sharing ideas, comments for giving each other feedback.

So with this new era of competitors, the Claudes and OpenAIs of the world, I think the real question is whether work moves to single player platforms where everything is conversational, or whether work continues to happen in workflows that need real primitives, with AI as an intelligent companion across those primitives. We're betting on the second. It's easy for me to ideate in Claude, but very hard to get real work done there once multiple people are involved. We believe teamwork will remain multiplayer, where multiple people need to engage with the same AI at the same time, and where work needs to be provisioned properly, with the right accountability and the right approval chains. So the older competitors struggle because they lack platform maturity and primitives, and companies like Claude are still single player AI built for research and thinking, not multiplayer AI built for complex work execution.

In a slightly adjacent category: ClickUp acquired Codegen, and we've all seen Replit and Lovable growing tremendously as vibe coding has gone mainstream. Do you see app building as a feature within an all-in-one platform, or as a separate, adjacent lane?

Within a year, most software platforms will offer vibe coding and app building. Claude itself has gotten good enough that you can prompt it to build apps, though it's not yet as good as Replit or Lovable because of the system prompting and orchestration they've built. But as AI improves, building simple apps will become table stakes. So yes, I'm confident app building becomes a feature readily available in a product like ClickUp, and not just ClickUp, a lot of good software companies will offer this.

We already have it in ClickUp, and we're piloting tools that let people build end to end dashboards and solutions using ClickUp's own work primitives. That means the solution inherits the right level of access control, shareability, and governance, you don't have to build a task list from scratch, you inherit all the building blocks, because these apps are ultimately just a doc editor, a list, a record you can click on and expand. For a company like ClickUp, it's much harder to create software from scratch every time than to customize strong existing primitives for a given use case. So even here, having the primitives gives us an edge: if you and I both work at the same company and both write code using those primitives, there's a high chance we get similar output, rather than two completely different things.

We've touched on a lot of primitives already, like tasks and lists. What about things like whiteboards, docs, and time sheets? Are there any new primitives worth thinking about?

Timesheets are a big one. Clips, audio and video clips like Loom. Whiteboards that become AI canvas. Docs and collaborative artifacts. Each of these primitives is a billion dollar SaaS category on its own, but for us they're just features. Our goal is to build every modern work primitive that exists, so we have a lot in the works. We're building spreadsheets right now, in part because the more native primitives we have within the same foundational database, the better our AI gets. That doesn't mean we won't also build a highly extensible platform: we connect with all the major MCP servers, so data can be transformed across the ecosystem efficiently and without losing context.

On measurement: what's the right way to measure an agent's ROI, based on what you've seen customers do or what you do internally? And bundled into that, knowledge work isn't as black and white as code, which has to pass tests and compile. How do you measure whether something has been done well in an area of knowledge work that's a bit more subjective, and what's the right way to measure ROI once an agent task is complete?

That's where a platform like ClickUp is powerful, and it's part of why our customers like us. There are two levels to think about here.

With ClickUp, you get access to all the frontier models, OpenAI, Gemini, Anthropic, through a copilot experience, chat everywhere, plus an agent building platform where people can build simple agents. That's powerful because it democratizes AI and lets employees build what they want. Some people build great use cases that drive real impact, you can see it, things that used to take weeks now get done in days or hours. That layer is working well.

But there's another layer companies need to move to, and we're beginning to move there ourselves. It's not enough to give employees AI tools and let everyone build their own version of a workflow. At some point you have to take what you've learned and build consolidated, certified workflows at the department level, so that everyone on a team uses the same workflow, say, for onboarding a customer, rather than everyone building their own version. You unleash people with the tools first, let them experiment, but eventually you have to bring that learning together so teams operate like a coordinated fleet rather than everyone freelancing.

So we have two AI efforts: empowering the masses, which is going well, and second, each team at ClickUp building what we call an AI native operating model. That means identifying the jobs to be done in your organization. In a support organization, for example, the jobs to be done are things like knowledge management, answering tickets, and triaging tickets, which break down further into ticket types like technical or billing related. For each department, you break jobs down to a granular level, in marketing, for instance, what are the 10 to 20 jobs to be done if you hired someone new? Our teams are starting to build AI at that jobs to be done level, which usually isn't a single agent or tool but an end to end orchestration workflow.

Thinking in terms of workflow makes outcome tracking much easier, because you're not deploying AI for its own sake. You can say customer onboarding takes 30 days, and with AI we can bring it down to 22. What you can't do is immediately tie that to a claim like "revenue increased 20% because of this."

Instead, we measure ROI based on the lowest level KPI we know is highly correlated to the top or bottom line. If you onboard reps faster, their productivity and sales performance should improve, and ARR should follow, but since I'm building AI to onboard reps faster, I measure success in terms of rep attainment and how quickly they get there, not directly in top line impact. Across each team, we break the workflow down to find the lowest level KPI we can measure that we know correlates with the end outcome.

It reminds me of the British cycling team story. They were losing, a new coach came in, and instead of obsessing over winning the race, the team broke cycling down into about a hundred input variables: how you pedal, how you hold the handlebars, what you eat, how much you exercise. Instead of fixating on winning, they focused on perfecting each of the lowest level variables that they knew, in aggregate, would move them closer to the outcome. Some of those, like pedaling faster, tie directly and quickly to time and ROI. Others, like eating well and staying fit, matter but are harder to correlate directly and quickly.

I think of this as the era of autonomous business, where AI workflows accelerate both how work gets done and how much work gets done. What's been eye opening for us is what happened with our own data analytics and tracking. It used to be a slow, reactive process where an analyst could analyze three or four things a week. Now we're doing close to 600 analyses a week, the equivalent of unlocking many more analysts than we actually have, and we're doing it proactively. We've built a system that tracks 60 KPIs end to end.

You could ask what outcome that's driving, and the answer is that it helps us fix the business proactively rather than reactively, by identifying trends that can then be operationalized. And the marginal cost of that is very low: at very little additional cost, we've improved a team's throughput and made it proactive instead of reactive. That's the workflow layer I'm describing. The more of these AI driven workflows we build, the faster we can work and the more volume we can handle.

I'd like to move on to growth, then business model, then the future, probably three more questions. On growth: companies like ClickUp and some of your competitors are known for mastering product led growth and paid acquisition. I recently saw statistics on how much Product Hunt and App Store launches have exploded, and I'm sure product launches generally are exploding too as everyone ships more. What's effective today for launching a new product into this market? What are some of the new go-to-market launch practices and growth loops that actually work now?

When it comes to launch, in a world where the product layer gets commoditized, perceived differentiation becomes extremely important. You can differentiate either by building a net new feature so compelling that people want to use it, or by being excellent at packaging, standing for something people believe in, an unmet need that existing solutions haven't fully solved. Either way, a differentiated product experience and the story around it become critical.

Look at consumer products: they're commoditized, and what wins is either something genuinely cool that catches attention, or a brand built around a set of shared principles with customers, serving a slice of the market that deeply believes in the same thing.

As more software gets commoditized, true product differentiation paired with a strong, community focused brand becomes important. When you launch, you need both of those in mind, and you need to collaborate with an ecosystem that believes what you believe, like influencers. Twitter has become an extremely important channel for launching new products, and early customer evangelism matters a lot.

We also live in a time where AI customers are growing because their customers are automating workflows, and usage is compounding within the same accounts, which makes the product itself the kingmaker. In consumer, a good ad with a celebrity can get someone to buy once, but it won't get repeat purchases. At some point, using the product has to feel genuinely great, you have to deliver on the value and exceed expectations. In SaaS, broken buttons and slow loading websites used to be tolerable.

That doesn't work anymore. Your product has to delight the customer around its core value, and that becomes the biggest moat, with your launch anchored around it. I borrow a lot of these principles from consumer, because that's the direction the whole industry is heading: software is no longer an 80% gross margin business by default, AI will compress margins and commoditize software further, so you have to stand out. You have to make it extremely easy for people to sample your product, and if they like the sample, they'll keep coming back, the same way a new drink or a new candy earns repeat buyers.

So you'll still have to do some heavy lifting to get people through the door, similar to PLG and SLG today, but the product then has to be good enough that usage compounds, and AI is well suited to that when you're solving the right problem.

To bring these ideas together: product differentiation drives true consumption, and true consumption is what lets you spend more at the top of the funnel. LTV gets built through genuine consumption, which is where AI products are leading the way right now. If your product drives real consumption and higher LTV, you can afford more CAC and more advertising spend. Brands, on the other hand, get built at the top of the funnel, through great brand advertising, storytelling, and community evangelism, the way great consumer CPG products are built.

I'm drinking LaCroix right now: they don't do performance marketing, they do brand marketing. As more products get commoditized, marketing shifts from bottom of funnel demand capture to top of funnel awareness creation, and companies that can drive higher LTV and more consumption will increasingly own the top of the funnel, billboards, TV ads, and so on, since the bottom of the funnel is largely commoditized.

Paid marketing isn't going away. If you have good LTV from real consumption, you'll spend more on paid to build awareness and take the market faster. But great products also drive word of mouth, which becomes another channel. Sales becomes an important channel too, especially human led experiences, not AI slop, but genuine, high touch enterprise sales for high value items where customers want a human to walk them through the experience. And the last piece is AEO and GEO, since the way people discover products is changing.

The way to think about it: if your product is commoditized, there's no point building a brand around it unless you actually have a story to tell. In that case, you win through AEO and GEO, or better bottom of the funnel performance marketing—showing up where people search. Nobody builds a brand around a nail cutter, you just want to be the best nail cutter that shows up in Amazon and Google reviews.

For differentiated products with unique stories and the LTV to back it, we will see a deeper emphasis on building a real brand (content, advertising, experiences) and enabling more human related distribution channels (WOM, community, sales, referrals), with an emphasis on delivering great experiences.

Distribution is changing fast, and you need to know whether you're selling a premium, story driven experience or a commodity like a nail cutter. If you understand which one your business is, you'll know which distribution channel to lean into.

Let's do a future question, and I don't want to keep you too long past the mark. Everything is changing quickly, but if we push you to imagine a world where everything goes right for ClickUp over the next three to five years, what is the company in three to five years, and how has it helped companies, humans, and AI work together differently?

We become the harness for modern work. Implicit in that is that any great harness has humans training AI on what to do: AI does the work, and humans build the agentic systems, review the work, and constantly fine tune that harness. Our goal is to become the harness for knowledge work.

To backtrack for a moment: how does that change your business model? If you're the harness for work and a lot of that work is being done by agents, where do you think pricing and business models are heading? Is everything moving toward usage based pricing, or will there always be a place for seat based pricing too?

We'll most likely land on hybrid pricing, though I think the bulk of revenue will ultimately come from usage. As a species we'll do a hundred times more work than we do today, but that hundred x increase won't come from more humans, the population isn't expanding a hundred x, it'll come from agents consuming credits to do the work. That's where you'll still need some seat based or platform pricing, because there's a lot of plumbing required to enable this, enterprise readiness, governance, security, though you can amortize some of that through consumption too.

So I think it ends up hybrid: some seat or platform cost, plus token consumption on the upside. As customers win and do more and more work on your platform, driving more outcomes because of it, that creates a consumption increase, and that increase shows up through token usage. If a customer isn't finding value, they won't consume tokens, so they won't be charged for what they're not using.

This transcript is for information purposes only and does not constitute advice of any type or trade recommendation and should not form the basis of any investment decision. Sacra accepts no liability for the transcript or for any errors, omissions or inaccuracies in respect of it. The views of the experts expressed in the transcript are those of the experts and they are not endorsed by, nor do they represent the opinion of Sacra. Sacra reserves all copyright, intellectual property rights in the transcript. Any modification, copying, displaying, distributing, transmitting, publishing, licensing, creating derivative works from, or selling any transcript is strictly prohibited.