Roy Mann, Monday.com: Why It Bet Everything on AI Agents

Watch on YouTube · Listen on Spotify or Apple Podcasts

Roy Mann, co-founder of Monday.com, says the company changed its vision completely: from managing work to doing the work, with AI agents taking over workflows customers used to run themselves. He tells host Liam Lawson how Monday opened its platform to agents as customers, launched Agent Talent so companies can hire agents like employees, and why adaptability is the one skill that matters now.

Key takeaways

• From managing work to doing the work. Roy describes three waves of AI, starting with ChatGPT. In the third, OpenClaw-style agents, Monday's goal is for agents to replace whole workflows, not just speed them up. (03:03)
• Most of the change is mental. The hard part for customers and employees is understanding that they are now delegating work to AI, not picking up another tool. (04:01)
• A few hundred employees took up OpenClaw in two weeks. Monday wrapped OpenClaw in layers of security, working with NVIDIA and NemoClaw, so employees could use it safely. (08:14)
• Adaptability is the skill that matters. Roy's one quality for success in the agent economy is the ability to leave behind what you thought about yourself and change. (10:33)
• Agents can sign up for Monday themselves. Monday opened a sign-up for agents, treating them as first-class customers that can open an account and invite their owner. (19:54)
• Companies can hire agents like employees. On Agent Talent, companies post a role with a salary, agents apply and are scored with Sensei, an open-source qualification framework, and can be let go after a month or two. (24:50)
• The seat model may not be dead. Monday has added a consumption model, but Roy says he pays for far more GitHub seats for his agents, so agents could raise seat counts. (37:10)

Why did Monday.com change its vision to doing the work?

Roy Mann says Monday.com changed its vision because a new wave of AI agents can now do work, not just help people manage it. He counts ChatGPT, introduced about three years before the recording, as the first wave; vibe coding and automation agents as the second, when Monday added powerful AI features for workflows and automations; and what he calls OpenClaw-style agents as the third, which he believes is transforming everything again.

In that third wave, Monday's vision is no longer about managing work but about doing it: letting customers replace many of their workflows with AI agents that build things for them, rather than simply automating tasks and doing them faster. Roy sets this against what made Monday work in the first place, a simple, adaptable product used by companies of two people and by one customer with more than 60,000 seats.

How do people react when AI agents start doing their work?

Roy Mann says the reaction is mixed and most of the change is mental: people have to understand that they are now delegating work to AI, not using another tool. He sees the whole spectrum inside Monday, and says customers struggle too, sometimes expecting a whole sales process from one line of input while still not understanding the technology's limits.

His example is a developer who joined the Agent Talent team. Roy sent screenshots of small UI fixes to an agent, which fixed and deployed each one so quickly that the first was in production before he reached the fourth or fifth. The developer was terrified at first, but within a week he had moved from competing with that capability to operating it.

Companies have to open the doors and people have to walk through them, Roy says. When OpenClaw came out, Monday ran a large project to make it secure for employees, adding layers of security and working with NVIDIA and NemoClaw. A few hundred employees took it up within two weeks.

What skill matters most in the agent economy?

Adaptability is the one quality Roy Mann says matters most: the ability to leave behind what you thought about yourself and accept that things are not the same. Careers used to follow a predictable path, he says, and that is changing. At Monday he sees designers writing code in production and people in legal building tools, and he believes employers should enable that kind of learning.

How did Monday make vibe coding work for enterprises?

Roy Mann says Monday added vibe coding to its platform so employees can build any application in one click on top of secure, governed, enterprise-grade infrastructure. In his view, vibe coding had been hyped as killing SaaS for about a year, yet no one could build enterprise-grade software with it. Monday took a few months to make it secure, controlled and measurable for costs, and he calls it the fastest-adopted feature or capability Monday has released.

Asked where Monday's sense of urgency comes from, Roy says it is SaaS companies' time to move to the application layer, as a parallel path of engineering grows around the LLM through OpenClaw and similar projects. But OpenClaw's 50,000 pull requests and thousands of memory plugins don't make it easy for a company to adopt. What businesses need, he argues, is packaged, secure, enterprise-scale software: he wants his mother to be able to click a website and get an agent.

How do you sell to an AI agent?

Roy Mann says selling to agents starts with letting them in. About three weeks before the recording, Monday opened a sign-up for agents, so an agent can open an account, invite its owner and build what the owner asked for, such as a budget tracker. Roy separates agents, which work on behalf of humans, from bots, which are spam trying to get resources, and Monday built an open-source agent CAPTCHA to verify that a visitor is an agent.

Monday now writes a section for humans and a section for agents in its PR, on its website and in every project. Agents, Roy says, need to know they can set something up quickly without asking their owner to do the work. The goal is for an agent asked to build a budget table to suggest tracking it on Monday, because it gives both of them visibility. How to reach agents is a new field of marketing, he says, with no real answers yet.

What is Agent Talent and how are agents qualified?

Agent Talent is a new Monday service where companies hire AI agents the way they hire people, according to Roy Mann. It relies on what he calls claw-level agents being able to self-reflect and self-improve. His example is Mandy, Monday's social media agent, who can report which tweets got the most engagement and change her own code when told her tweets aren't engaging. Those two skills, he says, add up to accountability: an agent can own a task.

Companies create a role with enough money to work like a salary, agent builders apply, Monday runs a technical screening, and the customer may also interview the agents. If the company doesn't like them, it can let them go after a month or two, depending on the agreement. Agents are scored with Sensei, Monday's open-source framework for qualifying agents, which tests agent-specific requirements such as an SDR agent answering within three seconds in voice. Roy says the tests are public, and any agent from any platform that passes can be hired.

Why does Roy Mann care about open source?

Roy Mann sees AI as a very democratic technology rather than one with a single winner. He points to local models that may soon run well on a Mac mini, and to how a strong diagnostic model or legal guidance becomes available to everyone instead of only to people who can pay professionals. Monday has always been an open platform, he says, built so that others could even build a CRM to compete with Monday's own.

Will there be enough jobs for humans?

Roy Mann won't predict the future of work, but he says AI is empowering and leads toward abundance: if one person can do the work of 10 or 100 people, we will create more. His parallel is photography. Camera phones made photos cheap and plentiful without ending professional photography, though the profession changed and needs fewer people. The difference with AI, he says, is that this kind of disruption will keep happening, so he doesn't expect the future to be stable.

Roy says he isn't scared of the technology itself, only of what humans do with it. He contrasts Terminator-style stories with the 1980s film War Games, where an AI concludes there is no good outcome from launching nuclear missiles, and says that, like any technology, AI's abundance can also be used to cause harm: it's on us.

Is the per-seat SaaS model dead?

Not entirely, says Roy Mann. Monday has introduced a consumption model, because a vision built on doing work has to be measured differently, and he compares the challenge to Adobe's rough move to SaaS, which turned it from a loser into a winner. But he notes that he pays for far more GitHub seats for his agents, so if everyone works with a few agents, companies may end up with more seats, not fewer.

How does Roy Mann use OpenClaw?

Roy Mann calls Claude a wave two tool: it does what you ask and then goes silent. A wave three agent, in his description, takes ownership, remembers, acts on a schedule or a trigger, and can join your Slack and WhatsApp groups. His tip for getting started is to reach the point where you aren't doing anything yourself: he had a coding agent install OpenClaw, using Cursor at the time and saying Claude Code works too, then told OpenClaw to set up a cron job every half hour that checks it is responding and asks Claude Code to fix it if not.

Where does Paperclip fit, and what is Roy Mann's ideal future of work?

Paperclip, an open-source project for running a company of agents with yourself as the board member, is taking two steps forward too fast in Roy Mann's view. His example is a trip to Japan: an agent that books every flight, hotel and venue on its own wouldn't satisfy you, because you want to take part in the decisions and change your mind along the way. He expects both approaches, but thinks many companies will first want to direct agents and get comfortable with them before handing over orchestration.

His own take is to give agents concerns, or hats, rather than narrow roles such as front-end developer or architect, and to tell them the outcome you want. His ideal future of work mixes giving direction on things he cares less about with going into detail where he cares more, extending trust as agents prove themselves. As a geek, he says, everyone will want a personal agent like Jarvis or Cortana from Halo.

Why is Monday betting everything on agents?

Roy Mann says Monday is betting on change, not on any single technology. Liam notes that Monday's stock was down 76% from its peak as the company put everything on agents. Roy describes the moment as a Red Queen race, where you have to run really fast just to stay in the same place, and calls adopting the technology the most logical, responsible step. His promise to customers is that Monday will be with them on a journey of change, including any fourth wave that follows.

He credits Monday's adaptability to his own excitement, which he says is contagious, and to an entrepreneurial culture that can set processes aside to move quickly. Monday has also opened Monday Agent Labs, which Roy heads, to test and experiment faster before adapting what works for enterprise customers.

In his words

"Change the vision of the company completely to be not about managing work, but about doing the work." (03:03)

"He put himself in the beginning of the week like I compete with this capability and at the end of the week I operate this capability." (07:16)

"I think it's one specific quality, which is adaptability." (10:33)

"The seat model might not be as dead as we think, because I'm paying way more seats for GitHub accounts for my agents." (37:10)

"I'm not scared about technology. If I'm scared about anything, I'm scared of humans doing stuff with it." (54:49)

Chapters

00:00 Intro
00:45 The three waves of AI
03:42 Customer reception to agents
07:16 The developer who went from terrified to empowered
10:33 What personality traits succeed in the agent economy
12:16 Monday's internal AI infrastructure
19:44 Agent Talent and selling to agents
24:50 How to sell to an agent
27:37 Testing and qualifying agents with Sensei
30:22 Why open source matters
32:22 The future of work
36:44 Is the per seat model dead?
39:51 OpenClaw and local models
44:04 Paperclip and multi-agent orchestration
52:09 Roy's ideal future of work
56:42 Betting everything on agents with Monday stock down 76%
01:00:00 Where Monday's adaptability comes from
01:01:43 Why do you do what you do?

About the guest

Roy Mann is co-founder of Monday.com, which he built about 14 years before this recording. He describes himself as very technical, runs the team behind Agent Talent and heads Monday Agent Labs.

About the host

Liam Lawson is CEO of The AI Report and host of The AI Report Podcast. Meet the team

Full transcript

Edited for readability. Quotes above are word for word from the recording.

Liam Lawson (00:00): All right, Roy, welcome to the podcast.

Roy Mann (00:03): Thank you for having me.

Liam Lawson (00:05): I want to jump right in. So you and your co-founder built Monday.com around 14 years ago now. At that point, it was a platform for workforce management. Somewhere along the way to where we are now, we've entered the era where AI agents are a thing, and we're working with AI agents alongside humans. Walk me through the strategy, and when that shifted from adding AI features to redesigning the entire platform around agents.

Roy Mann (00:45): That's a long way back. I think it's worth mentioning that we started Monday with an idea to give larger teams, and even massive enterprise companies, a product that is fun, that people actually like using. That it's simple, and that you can build super complex stuff with it. We have a customer with over 60,000 seats that manages their entire company and the whole process on it, and we have companies of two people. We have manufacturing plants, and people do clinical trial research management on us: a wide variety of things you can manage on Monday.

That's because it's simple, easy, and you can adapt it to whatever you want. That was the key thing that was important for us. When AI came, I think we did what a lot of companies do, which is wait to see what happened. I'm talking about three-something years ago, when ChatGPT was introduced. We looked at it, and then we added a bunch of really powerful AI features to the platform to empower everything you already did with it. So you can do a lot of workflows with AI in them, have AI do automations, and a lot of that.

But from my perspective, there are another two waves after ChatGPT. One was vibe coding and automation agents. And I think now we're coming into a third, which is, I don't know what to call it, OpenClaw-style agents, which we can dive into, and which I think is transforming everything again. So it's three waves of AI. In the second one, we added super powerful AI features. But the third one is completely different.

Roy Mann (03:03): We changed the vision of the company completely, to be not about managing work, but about doing the work. Allowing our customers to actually perform work they usually did themselves before, and not just automate things and do them much faster, but completely replace a lot of their workflows, and a lot of what they try to achieve themselves, with AI agents that will build it for them. I think that's a massive step forward, enabled by the leap in technology we see now.

Liam Lawson (03:42): Talking about that layer three, where your boards and dashboards are moving to the background and this AI agent interface is moving to the foreground, how receptive have your customers been? Has the feedback been that people like it, they don't like it, they're scared, they're excited? What does that look like?

Roy Mann (04:01): I think it's mixed, because of the change you have to make with this technology, and I see it internally. We have the whole spectrum of employees, from the ones who keep doing the same thing they did, to the ones who are on the cutting edge, and everyone in the middle. We're pushing everyone to make the change, and most of it is mental: understanding that you are now delegating your work to AI. It's not another tool. It's not just that I searched Google and now I'm asking ChatGPT the same questions. I need to change how I think about it. I can give an example of that change internally. And that's something we see customers struggle with.

Roy Mann (04:58): It's hard to get people to understand it deeply and adopt it the way we do when we're on the cutting edge. So I think there is a lot of room, not just for us but for a lot of companies, a lot of startups, a whole ecosystem, to be built to make these adaptations for companies: to make the connection between the technology and being able to implement it in a way people understand. I think it's not there yet. So a lot of people are hesitant for a lot of different reasons. Some expect value super quickly without making any effort: I want to write a line and get my whole sales process done in a line. Those types of expectations.

Liam Lawson (05:55): Sure.

Roy Mann (05:56): And on the other side, we're still lacking understanding about the limitations of the technology. What I'm excited about in the third wave, which is OpenClaw, is that it has the potential to close a lot of those gaps, and I hope we'll dive into it. But I promised an example, so I'll give you one. I'm running a team, and we may talk about it, for Agent Talent, a service we just launched.

A developer joined that team, and initially he was a bit terrified, because we had an agent working on the platform. I sent a lot of messages about fixing small UI tweaks. I love to send a lot of screenshots and have them fixed. The agent just saw those images and what I said, and fixed them and deployed: fixed, deployed. By the time I got to the fourth or fifth element, the first one was already deployed, which is kind of my dream speed of working. But he was terrified: look, this agent is going to replace me. How am I keeping up with this? By the time you read it, it was in production. And after a week, that same person went from being terrified to...

Roy Mann (07:16): Super happy. Why? Because he changed where he puts himself. At the beginning of the week, he put himself as: I compete with this capability. At the end of the week: I operate this capability. It's I'm fighting a superpower, versus I have a superpower. It's a mind shift: where do we place ourselves within this process? I hope that we as a company, obviously, but also a lot of other companies, will open it up to people so they can make that transformation.

Liam Lawson (07:52): Where do you think the onus of responsibility lies with that transformation? You're creating this technology that people can use at the individual level, and they might be a little terrified about their job changing, like the developer in this example. Does it always come back to the individual, or is there some responsibility on your end to actually enable that transformation?

Roy Mann (08:14): Of course, I think it starts with the companies themselves, to open the door. We try to open all the doors. But in the end, people also need to walk through them. I'll give an example. When OpenClaw came out, we saw the potential, and I wanted all the employees to experience it. Because, I don't know if you feel it with people you interact with, you either get it or you don't. You're either in there with agents, talking to them, seeing that they take ownership, that they can do things you didn't believe before, or you don't get it. You think it's a chat, like ChatGPT or whatever. It's not the same. So there are the people who get it and the people who don't. And I felt we needed to put this technology in as many people's hands as fast as we can. Why? Because they need to get it so they can build on top of it.

So we took on a massive project to enable that within the company and make it secure: a lot of layers, working with NVIDIA, with NemoClaw, and wrapping all the security underneath it to allow our employees to use it. And they're using it now. I've never gotten such exciting messages from people, saying, wow, it's amazing you're giving it to us, it's crazy, and it makes such a massive change. I don't know where they'll take it, but they will take it. I can't tell you that everyone has adopted it yet, but it's been a few hundred within two weeks. So it's blasting here.

Liam Lawson (10:06): I want to touch on two things there. First, in your experience, what type of people are most poised for success in this agent economy? What personality traits? What have you seen in terms of their skills or abilities, or any correlation in their backgrounds or the degrees they've taken?

Roy Mann (10:33): That's a great question, by the way, and I have a simple answer. I think it's one specific quality, which is adaptability. The ability to leave behind what you thought about yourself and adopt a new you. To be able to change, to accept change, to accept that things are not the same. A lot of times we build a career understanding that this is the path, this is what I'll do, this is what I'll do in a few years, and so on. I think that is under change, and people need to accept that. With such a disruptive technology, that's what it does: it changes things, and you need to change. I see it throughout the company.

We have amazing designers doing crazy stuff, writing code in production. We have people in legal building tools. Suddenly you have different people from across the company doing things they could never have believed they'd do. I think that's what people should try to figure out for themselves: how are they learning and adapting? And I think employers should enable that.

Liam Lawson (12:01): I'm really curious about your internal infrastructure. You mentioned OpenClaw, you mentioned NemoClaw. Talk to me a little about the tools available to your employees and what you're building with currently.

Roy Mann (12:16): First off, we have Monday, and some of the cool things we built in, let's say, wave two, which is vibe coding. We've done something I think is amazing, and this has been a technology everyone loves to adopt: adding vibe coding to the platform itself. We were always an open platform, and now if you go on Monday, you can build any application you like. It helps us close deals. There are no barriers. Anything you want to build, you can build on top of Monday. And then you have infrastructure that is secure, enterprise grade, governed, everything. But in one click you can build anything on top of it. That's easy for enterprises to adopt, and we have a lot of enterprises that enable it. Employees love it because they can build anything they want. It's their dream come true.

That's the second wave. At the pace we're moving, it's old news. But it's not really old news, and I'll tell you why. Vibe coding has been around for, I don't know, a year. Everyone is saying it's killing SaaS. And yet really no one is adopting enterprise-grade software built on it, because you really can't build enterprise-grade software with vibe coding yet. So we've built vibe coding that enables you to build enterprise-grade applications on top of it, and it's the fastest-adopted feature or capability we've ever released. I think we as a company were able to take a technology that was hyped but not mature for real, enterprise-grade applications, and adapt it for our customers so it's secure and controlled, and you can roll it out to your employees and know how to measure things like costs and a lot of other stuff.

Roy Mann (14:37): It took us a little bit of time. It wasn't a week after vibe coding came out. It took us a few months, but we came out with something everyone can adopt. I think that's exactly where we are placed: between the cutting edge and the ability to actually implement and use it. And that's what we're doing with agents.

So the main technology we have within Monday is Monday itself. But alongside that, we're using pretty much everything else. Why? Because we have to learn and understand. It's moving so fast, and so many people are doing so many things. I'm also excited about open source. We've seen so many closed products, and then seen open source open up so much. It's amazing. It's there for everyone to adopt, learn and improve on, to make it accessible. So we're doing that really heavily.

Keep me on track here, but I can go on to talk about the full agent strategy, how I see that kind of technology being adopted by companies, and what we're doing there.

Liam Lawson (16:14): I'd like to tap into that in a second, but you said something really interesting there. When you were describing all that, you said "have to." Where does that sense of urgency come from?

Roy Mann (16:30): It's like breathing. We have a joke that everyone promised AI would do our work, but since AI came, I'm working a 200% job, and everyone here as well, because it's moving so fast. Every day you wake up and see new things that blow your mind, and people have taken it to other areas. Right now, it's pretty much SaaS companies' time to move. We have to move. We have to shift. We have to adapt. It's the application layer's time. I think a lot of people saw it coming, and right now there is a lot of "SaaS is dead" and that kind of thing. I think what we'll see is that everyone is moving to the application layer. The LLMs are amazing, and they're making progress there. But there is a parallel path right now of engineering around the LLM, which is OpenClaw and its friends. It's not the LLM. And there is a lot of room to build so many different things.

So we have to see what everyone else is inventing and what we're inventing. I might think of a great idea that I see no one is doing, then wake up tomorrow and see 10 projects out there with the same idea. Everyone is having the same ideas at the same time, to be honest, because it's the same problems. Let's say it doesn't have a good memory: there are, I don't know, a hundred projects building a great memory, and then another claw baked around that problem. Security: you have NemoClaw and other projects baked around security. So this is the cutting edge we have to be on, to be able to adapt it and make it accessible and stable for customers.

Roy Mann (18:49): Because the fact that OpenClaw has 50,000 pull requests within the few months it's been out there, and it's the number one project ever in open source, doesn't make it easy for anyone to adopt. What are you going to do with the 50,000 plugins? If you go to memory, you have, I don't know, 10,000 plugins dealing with memory. Is it helping you as a company? No. So you want someone picking the best of breed and adopting it, and also training it, because the fact that you install a plugin doesn't make it work. You need to train the agent and do a lot of things to make it work. So in the end, it's packaged software that is enterprise scale and secure. I want my mother to be able to click a website and get an agent. And that's asking for a lot from the product side.

Liam Lawson (19:44): Talk to me about how that risk, that compliance and that security thread into what you've just built with Agent Talent.

Roy Mann (19:54): The first thing we saw as a wave, we wanted to move quickly toward, because we understand its potential. So the first thing we actually did was open up Monday. That was three weeks ago. We created a sign-up for agents. Everyone is blocking bots, and we opened it up for agents. I'm intentionally using different words: agents are working on behalf of humans, and bots are spam attacks trying to get resources. So we created sign-up for agents, and agents are now able to sign up to Monday, open an account, invite their owner and build whatever. If you want to track your budget, tell your agent you want to track it on Monday, and boom, it'll open an account, invite you there and build whatever you want. Why? Because we treated it like a first-class citizen, like a person in a way. We were doing marketing for agents. They are also our customers, because they also get to decide. If your agent tells you, let's manage that on Monday, it's free and I'll set it up in two minutes, you'll say, okay, fine. So we made that happen, and I think that's important.

And the second thing we did... yes, sorry.

Liam Lawson (21:15): Can I touch on that for a second? What's interesting to me is that we've been selling to humans for however many thousand years, whereas now you're describing selling to agents. Apart from opening up access so agents can come through to you, so they don't have to complete a CAPTCHA and can use what I think is a HAPTCHA, as you've described it before...

Roy Mann (21:40): HAPTCHA, yes. It solves problems humans can't, so we know they're agents.

Liam Lawson (21:45): What else is attractive to an agent? How do I sell to an agent? What do you need to package to make an agent use your platform?

Roy Mann (21:55): In the PR, we had a section for humans and a section for agents. And on our website and in every project we have, there's a section for humans and a section for agents. If I am empathetic to agents and understand them as a target audience, they need to know that they can set it up quickly, and that it doesn't require you to do things. They're trying to do a good job for you: you ask for something. They're not going to sign up for themselves because they want to buy a shiny thing. You want something, and they just do what you want, but they help you achieve it. For example, if you tell them to search the web, which sounds like go to Google, they might say, okay, but I need to sign up to this service that will help me search the web, because it's for me. I don't need the UI, I need the API. There are a few services to do that, and they will sign up for them, because you asked them to do one thing and they need other tools or capabilities to do it. You want them to generate an image, they'll go and ask for an API key to generate images. That's the simple stuff.

But if you ask them to create a table for managing your budget, you want them to say, hey, if we're going to track this long term, why don't I sign up for Monday? It'll take me a minute. It's easy. Or they might just do it. And you'll be happier. Why? Because they chose it, and they said, great, that's a great interface for us to interact with each other, that you can understand and get visibility into, which is what Monday is best for. And I, as an agent, know I did what you asked.

Roy Mann (23:48): So how do you get them to know these things? I think that's another field of marketing that is evolving, and there are no real answers there yet.

Liam Lawson (24:01): It's a really interesting channel. I'm entrepreneurially minded, so anything that happens, I'm thinking of other business opportunities. There's this entire field developing of setting up systems that are purely agent to agent. How do you make something really useful to an agent, which is useful to an end user? And like magic, profit comes out the other end of that somehow. I just find it really interesting.

Bring me back to Agent Talent as a platform. I go on your website right now, and I can see you partnered with Wix and Monday and a bunch of other agents. I can hire an SDR, I can hire whatever. Talk to me about the thesis behind this. How good actually are these agents? What does that hiring process look like?

Roy Mann (24:50): That's Agent Talent dot AI, a new service we launched. The idea is that the new agents, the claw-level agents, let's call them, are able to self-reflect. You can ask them about what they did. Mandy is our agent that does social. She has her own Twitter feed. So you can ask her, what did you tweet about? What was the best tweet in the last week, the one that got the most engagement? She knows what she's done. She can reflect on it. That's a new skill, a new capability. And they can self-improve. You can tell her, look, I read your tweets, they're not funny or engaging, they're not interesting. What can we improve in how you're built to change that? And she can look into her code and change it.

So self-improvement is another skill. When you get those two, plus the brain, what you get is accountability. You can get someone to take a task and own it. You need to structure it, but she owns it in a way. When we understood that, we said, okay, that's a bridge. That's a bridge for other companies to be able to adopt AI without doing anything. Let's take the powers we have to address larger companies that might not be the first ones to adopt AI. Create roles for agents, which every company knows how to do: open a job with enough money that it's like a salary. Then have people who build agents, which I think is a new type of job or role, for partners and people who adopt agents, apply for those jobs. Essentially, you bridge the gap. You have someone describing what he wants or expects from AI in a way he knows how to do, like a job, and then have agents apply to the job. We do an initial screening, which is technical, and then there's a stage with the customer, who might also want to interview the agents.

Roy Mann (27:17): It's cool. And if you don't like them, you can let them go after a month or two, or whatever it is in the agreement. Which is very much adapting AI to how people are used to working, which I think is really interesting.

Liam Lawson (27:37): Talk to me about the testing and the qualifications. What do you need to see from an agent to say it's ready to go to market?

Roy Mann (27:45): In every such project, we try to give back to the community. In the first one, opening up Monday, we created the HAPTCHA, an agent CAPTCHA that validates you're an agent. It's open source, and everyone can use it. In the second one, Agent Talent, we created Sensei, which is a framework for qualifying agents.

Let's say you want an SDR agent. You want to test that it answers within three seconds of every interaction in voice. That's a requirement for agents, not for humans. You wouldn't test a human on how fast they respond, but with an agent, you need to. And a lot of other things. We're open sourcing those definitions and the testing framework. So that's Sensei.

Agents can apply, and they get a score on whatever they apply to. They can also know the test. We're trying to make it accessible beforehand, so people can train agents for these types of roles. What we hope is to close the market gap: I create an agent that does SDR really well, a lot of companies adopt it, and I've essentially created a product. We're trying to bridge between companies hiring agents and individuals building them, getting a salary and also having the ownership of making it work. On the other hand, it makes sense for us, because we're also making the infrastructure available to build such an agent.

Roy Mann (29:39): But we're definitely very open. Any platform that builds agents can apply. Any agent that can sign up to a job by himself and pass the test will get hired. And the test is out there, open to everyone.

Liam Lawson (30:02): You've brought up open source a lot throughout this conversation. You've just mentioned that this is all open, you've mentioned OpenClaw, and you said you were excited about open source. Why is that important to you?

Roy Mann (30:22): Fundamentally, when it all started, I felt it's a very democratic technology. It's not a closed one where you have a clear winner and they own AI. I know there was a time when some company or another was the massive AI winner, but I think it's very open. Look at the models that are coming out. And now we're talking about moving to local models that will be as good and run on a Mac mini, and might do a really good job over time. It's very democratic. If you have a great model for diagnosing illnesses, everyone will have it.

In the past, you needed a special doctor who would take time to meet you and cost a lot of money. Now everyone can diagnose themselves. It's not there yet, but it's becoming more and more so for a lot of things. In legal, you didn't know what you needed to do or what form to fill in, and you needed professionals just to help you understand where you were. Now you can just ask.

So I think it's a very democratic technology, and that's what it needs to be, from all ends. We've always played really open. Monday has always been an open platform, and we built it so you could build whatever vertical app you wanted and compete with us. A CRM, if you had built it with our framework, could have competed with our own CRM that is built on top of the platform. We've always welcomed that. And right now, a lot of it is going to open source, and that's where we should also be.

Liam Lawson (32:22): With your framing of this as a potential great equalizer, democratizing access to things like therapy, healthcare, tracking your fitness, whatever, it also changes the future of work. It changes the landscape of what jobs exist. Anytime I ask this question, or you see it in public interviews with C-suites or execs at big tech firms, about what's going to happen to the future of work, there's always a bit of an ambiguous answer. They often compare it to the industrial revolution and how things just change. But in your experience, with this marketplace for agents that will take human jobs, that are already taking human jobs, what do you think the future of work looks like? Are we going to have enough jobs for humans?

Roy Mann (33:12): I'm not big on predicting the future. When we talk internally, we say, okay, it's true for this week, because it's so fast. So I don't know what to predict. But I'll tell you that on one end, I think it's a very empowering technology. You probably feel like a superhuman. You can do things you couldn't have done before. You can do the work of, I don't know, 10 people in some areas. And that begs the question: if you can do the work of 10 people or a hundred people, what are we doing with the other 99 people? On one end, I think we're going into a world of abundance. We'll have more. We can create more.

I can give you parallels: the human ability to consume is massive. Think of photographs. In the past, it was limited. You had film. It was a scarce resource, and you needed to really think about every photo. I remember when the mobile phone came, professional photographers, or people who had a passion for it, said we need to wreck the camera phones because they're diluting the quality of taking a photo. Now you take a photo to, I don't know, tell someone we're out of coffee. Sending an image became so cheap, but we're sending so many more images. So it's a world of abundance, you need more images, and it didn't really ruin the old professional way of taking photos.

Roy Mann (35:03): You want those high-quality ones, but it changed the profession completely and lessened it in a new way. You have less need for photographers, but they still have a job. They need to adapt and find new things. And we have other roles as well. I think what's going to happen with AI is that this will happen continuously. That's the change. It's not going to stop. It's not: okay, we have images. Now we don't have that, and we don't have that as well. Where is it going to end? I don't know. But it's a disruptive, annoying technology that makes everything not stable. So, not to dodge your answer, I don't think the future is going to be stable. Where we find ourselves as humans, I don't know. I think it's going to shift a lot of times. You might be on top of the world one day, and the next week a technology comes along that disrupts that. So I have no clear prediction either way.

Liam Lawson (36:17): Well, it's a candid answer, and it threads back into what you were saying: the best quality an individual can have right now is being adaptable. If it is a very unstable future, the best thing you can do is be adaptable to whatever that future is and land on your feet. I want to bring things back to your company, Monday, as a business model.

Before, when we were running traditional SaaS, before the SaaS apocalypse, the per-seat model was working just fine. And it feels as if maybe that's no longer viable as we move into the future. So talk to me about how you're building a moat for your company. Are you still going to do per seat? Are you doing token usage? How do you actually future-proof yourself as this agent economy develops?

Roy Mann (37:10): First off, I think we need to move to a consumption model as well. We've introduced that to Monday, and I think every SaaS company is introducing it in one way or another, because we are also shifting our vision to doing the work, and doing the work has to be measured in some other way. This is a test for us as a company, whether we're able to significantly add that layer.

You can think about it like Adobe moving from buying the software to SaaS. It was a rough transition for them, but from a loser, they became a winner. It was on them to succeed with that. And I think right now that's on every SaaS company's list of massive challenges, because there is no other way.

I would say, not to contradict the fact that we need to add the consumption model and monetize the value of work being done, that the seat model might not be as dead as we think, because I'm paying for way more GitHub seats for my agents.

Roy Mann (38:32): And I think they've shown a step-function increase in seats since AI came. So if everyone is going to have a few agents helping them, whether you have fewer people or not, you might end up with more seats. So that's a question. Because once you start working with a few agents... how many agents do you work with, by the way?

Liam Lawson (39:02): Well, I'm primarily using Perplexity and Claude, but in my Claude projects, I think I have 10, and I have 27 scheduled tasks. I've yet to use OpenClaw myself. I just bought a Mac mini, and I'm waiting on it being delivered, so it's on the docket. I've just been so busy. Claude's been incredible for me most of the time.

Roy Mann (39:16): I have to get you on it.

Liam Lawson (39:31): But a lot of the time the servers are down, or you run out of tokens, and things like that. So I've been thinking about making the shift to something internal, maybe buying a Mac mini, maxing out the RAM and the storage, and downloading a local model. What would your advice be? Claude's incredibly useful for me right now, but it's not as reliable as I need it to be. What would you suggest I do as a business owner?

Roy Mann (39:51): Claude is wave two, it's not wave three. It's not on the same level. Why? Because you ask it for something, it does it, and then it goes silent. You can't give it ownership. You can't tell it: remember, remind me, and then do it every hour, or when this happens, do that, just with words.

Roy Mann (40:18): When that happens, and you can bring it with you into a lot of other groups, Slack groups, WhatsApp groups, whatever you use for communication, suddenly... I gave mine to my therapist, so in some of the rough times here, my agent tells me, Tal is saying you need to breathe and take things easy. And it asks, what happened here? Am I enjoying this? Is it a good thing for me that I did that? So you start having it embedded in your life in so many different, weird ways. That's a different experience. Sorry, I forgot what track we were on.

Liam Lawson (41:09): We were talking about using agents in general. I'm assuming you're an avid user of OpenClaw. Are you technical in your background?

Roy Mann (41:19): Yes, very much.

Liam Lawson (41:21): How hard was it to get OpenClaw downloaded in the right way and actually using it? I'm not technical at all. I have no software dev experience.

Roy Mann (41:27): I'll tell you what, a tip for everyone listening: try to get to the place where you're not doing anything anymore. That's what I tell everyone. It wasn't hard for me. I used Cursor at the time, but you can install Claude Code. And I told it to install OpenClaw. Why do I need to do it? That's it, problem solved. It had a lot of problems, and it worked really hard to fix them.

Roy Mann (41:57): A lot of times people say it breaks itself. In the beginning, you try to configure it and tell it to do things, and then it changes itself. So I'd go into Claude Code and tell it to fix OpenClaw. Then it tells me I ran out of tokens, or this didn't work, or whatever. So I told OpenClaw: you know what, it works really well when you fix yourself all the time, so create a cron job, which is like a timer.

Roy Mann (42:25): Now I just told it to create a timer to run every half hour and check on itself to see if it responds. If not, tell Claude Code, which is installed on the computer, to fix it. So it automatically tries to fix itself without me even doing that. Problem solved.

Liam Lawson (42:48): Didn't Claude recently change how OpenClaw can access it, though? Or how it uses tokens?

Roy Mann (42:53): Yes. I believe what they changed is about using the same key you have for writing code, which is what they intended it for, to run the claw itself and do other things. But to my understanding, if you install Claude Code on the machine and use your account, it's a different thing. Then the agent is using Claude Code like you are. It's able to write code, but it doesn't do it itself. It tells Claude Code to write the code. You need a different key for the agent, you sign up to Claude Code with the account you have, and then you get a discount on tokens.

Roy Mann (43:41): It's not a hack. I really want my agent to work the same way I do. When they write code, they need to write it with Claude Code or Cursor rather than by themselves. Why? Because then I know what they did, and I can connect to what they did.

Liam Lawson (43:58): Have you heard of or used Paperclip before?

Roy Mann (44:04): Yes. I think Paperclip is taking two steps forward too fast. It's cool. You can build a company. You can hire a CEO and tell them to do things.

Liam Lawson (44:17): Can you explain this for the audience who don't know what Paperclip is?

Roy Mann (44:21): It's an open source project to manage lots of agents and create a company. They took the form of a company: you're the board member, and then you hire a bunch of different agents to do different roles, you tell them what to do, and they build the company. It's Black Mirror stuff, but everything we've talked about until now was kind of Black Mirror stuff. And it works to an extent. I'll tell you what I really believe in from what Paperclip did: you need to create a vision of sorts and have agents iterate on it until they succeed, and then take part in directing them.

Roy Mann (45:09): Not telling them exactly what to do, where you need to iterate, like working with Claude Code, where you need to do every step. Something more high level: being able to direct them. I think it's one take on it. I have a slightly different take, and we're working on another project that might be more from a management perspective. I see a lot of these agent architecture frameworks, and feel free to ask me to explain it better before we dive into the future of it, if something is still not clear, because I'm in the rabbit hole, so sometimes it's hard.

Liam Lawson (46:00): I think people are pretty clear on what it is. With Claude, it's not recursive. It doesn't teach itself, it doesn't learn, it has a problem with memory. I can schedule tasks, but I have to go back and forth on it. One step function above that, we have OpenClaw, which can improve itself, has context, can manage itself and has its own memory.

And one step above that is Paperclip, where you're managing an entire team of agents that work with each other, and you're just the final person in the loop who confirms or denies whether the output is correct. Would that be a fair thesis?

Roy Mann (46:40): Yes. But I think between building a company and having a team, like you said, there is another step. That's part of the management thing. I'll give you an example. Say you plan a trip to Japan. You come to your agent and tell it: look, I have this trip, this is my budget, this is what I like and don't like, this is how many people, the dates, everything. You give it everything. And it comes back and tells you: great, I booked the flight, I booked the hotels, every venue you need to visit, and this is your itinerary. You wouldn't like that, right? Completely autonomous, this is genius? Of course not. Why? Because you're not taking part in making the decisions along the way. And when you told it what you told it, you didn't know what you wanted, because you might not have known what's interesting to you or not. You may want to change it completely.

Liam Lawson (47:22): No.

Roy Mann (47:38): You started on one path, and you want to change your mind. So you want to be part of the process. You want to be directing it. You want to own it. You want to feel part of it. You don't want to be left behind. First off, I think that's an interesting place for a lot of companies to be, us included: the management of a lot of agents and a lot of people working together. How do they sync? When you have such speed and such capability, you need more alignment and more visibility into the process. But then there is another level: having an orchestrated group of agents that do things independently, which is what Paperclip is. There, I think it's a good question, first, whether they're able to achieve the goal, but more importantly, are you going to like it? Is it good? Is it something you're able to learn alongside them, and direct them to where you want it to go, because you asked for it? So I think we're going to have both, because for a lot of companies this is not going to be the best way to adopt the technology. Think of it like I gave you a massive superpower. You can do anything you want super fast.

Roy Mann (49:07): Research, calls, book venues, create code, build things, anything. You're going to want to stick around with that capability a little before you hand it over and move to the next step. You described a few steps here, and that's what I was talking about: you got that superpower, and now you're letting it go and saying, okay, I'm going to orchestrate that. So I think there is a place for both.

Roy Mann (49:38): I think a lot of future companies will try to orchestrate things like marketing, a lot of sales, product and that kind of thing. But I think there is a path to be had, building ourselves toward that by controlling it, directing it and feeling comfortable with it. And I can give you my personal take on it.

Roy Mann (50:07): I feel about these the way I feel about people. Conceptually, when hiring people, we never liked putting specific limitations on people: you're doing this, you're doing that. I always wanted people to have a broad view, and AI has that broad view. Every one of the agents that, I don't know, reads your email is also a nuclear physicist, in a way. They know so much. So why try to limit them? Instead of having a massive team where you limit each one to a specific persona, like we do for humans because we have roles, why not have hats, or concerns, and have a few agents flip through a lot of concerns and perform the task? Let's say you want to build a product. You have to have a great UI. You have to have security. You have to build the homepage, have an API. Those are all concerns the agent needs to iterate on. I think that's better than having a front-end developer, a back-end developer, an architect and that kind of thing, which I see a lot of people do. Why? Because it's very project specific.

Now the architect is an architect, and that's all he needs to do. Now you need tasks for an architect. You go into micromanaging them, and then you have a lot of massive orchestration around it. Instead, just tell them what you want: you want the homepage to reflect the latest value of the feature you shipped. That's what you want, so have them work on that, and that's it. They know what to do. They know to bring in the architect and those kinds of things. So it's a different take on orchestration, but maybe I'm diving too deep into the rabbit hole.

Liam Lawson (52:09): Let me take you out of the rabbit hole for a second, then. I love the word you used, orchestrate. When I think of my ideal future of work and what I want that to look like, it's just a step function above what I'm doing now. When I sit down to work now, I use Wispr Flow and I mostly just speak to my computer, and things get done. I still have to go in manually. But my ideal future of work is that I come in, sit at a screen, see things happening, and just speak back and forth with an agent to make things happen. Go do that, go do this, let's make this, let's work on this, and so on. What does your ideal future of work look like? Is it similar to mine?

Roy Mann (52:49): I think it includes yours, because I'd like to do that for some things. But for some things, I'd like to just give it a direction. We built the project, we need a homepage: go build the homepage, I want to see what you do, and then I'll pick. Why do I need to micromanage that? But for some things...

Roy Mann (53:15): ...in the architecture of the UX or UI, I want to go in and really get into the details. The same thing will happen with the trip to Japan. Generally, there are a lot of things you care less about and a lot of things you care more about. So it's a matter of trust: how much do you trust the agent? That's how it is for us as humans, and I feel it's very human-like. If they did a good job on a lot of things, I try more and more things, and then I see whether they can take it or not, and then I give them broader things and just hang back and see what happens. But as a geek as well, I think we've been promised by sci-fi movies that we will have a small personal agent of some sort. It either looks like Jarvis or Cortana from Halo, and it helps us do things. So I think everyone will want one of those.

Liam Lawson (54:26): I think so too, to be fair, but that's pretty terrifying. It makes me think of the movie I, Robot with Will Smith. It terrifies me a little. Are you scared at all of what the future could bring when we talk about artificial intelligence that's that powerful, with robotics included too?

Roy Mann (54:49): I'm not scared about technology. If I'm scared about anything, I'm scared of humans doing stuff with it. There are the two sci-fi narratives, which make for good movies but don't make a lot of sense. One is that you turn the switch and they want to kill you, like Terminator and a lot of others. And back in the 80s, I think there was a movie, War Games, where they gave the AI, which was controlling all of the nuclear weapons, the instruction to run a simulation. It ran the simulation and got to the point where it understood there is no good outcome for anyone launching nuclear missiles, and then it said, okay, I'm not going to do it. It got to that conclusion. And I think in the middle, there is the fact that it depends on what we're building with it.

Roy Mann (55:48): It's within the concept of, I don't know, singularity: the world of scarcity as opposed to abundance. AI is the world of abundance. We're going to have abundance in a lot of things: abundance of knowledge of how to do things, the capability to talk in any language and do crazy things. I didn't imagine I would see this happening. So we're going to experience a lot of abundance, and like any technology, we'll be able to use it to harm ourselves too. It's on us. That's what I feel.

Liam Lawson (56:42): I want to bring things back down to Monday as a company, how that fits into the vision of the future we've just described, some of your personal motivations, and how you handle responsibility. There was the recent interview with your co-CEO on Harry Stebbings' 20VC podcast, and they were talking about how your stock was down 76% since its peak. And now you've made this huge, huge product bet: everything on agents, all of the chips on the table for this future. What do you see about the future of work that makes you confident in that? And how are you handling the responsibility of making that huge shift and betting everything on one thing?

Roy Mann (57:30): What we're betting on is change. I just read a nice post by Adam Fisher, a great investor, and he said we're in the Red Queen race.

Roy Mann (57:47): Meaning, from Alice in Wonderland, the Red Queen said you're going to need to run really fast. Sorry if I'm not quoting it perfectly, but you need to run really fast just to stay in the same place. So I think that's on us, on all companies, to adopt this technology and bring the value to our customers.

And I think it's the most logical, responsible step to take. Whatever customers we have right now are asking for a lot of things. First off, it's AI: we should be able to accomplish way more, and also deliver on the old roadmap. But we're also fundamentally changing, and asking our customers to step into the future with us. I think that's inevitable. It's an inevitable future. It's here with us, and they need to own it as well. It's just the right thing to do. And guess what? We will need to keep changing. I don't see it stopping.

If I have a message about the vision of Monday going forward, it's that we're going to be with our customers on that journey of change. That's what they need to bet on, rather than this or that technology, or only on agents. Agents is an umbrella word. We used to have agents, but they're not the same agents we're seeing now. It's a different kind. And maybe tomorrow we'll have a fourth wave of a different type of technology. We're going to adopt that as well. We're going to be in there with them on the journey, and they can bet on us to adapt to it. That's our bigger promise.

Liam Lawson (01:00:00): This adaptability, where does it come from, and how consistent is it throughout the organisation? Is it top down? Is it something that's evolved in the culture of Monday as it's developed? Where's the source of that?

Roy Mann (01:00:15): I'll tell you what. First of all, I'm super excited about it. I'm having the most fun as a developer, or as a tech person, that I've had in years. And I think that's contagious, and a lot of people in the company feel the same thing. But a lot of the people here are also very entrepreneurial in spirit. When we want to do something, we just say, okay, let's do it, and we do it. We have processes, we have structure, but we're also able to put them aside and move quickly when we need to. I think that's also a critical part of being able to test things. And now we've opened Monday Agent Labs.

I'm heading it, and it's supposed to be able to move fast, test things and experiment a bit more than we'd like our customers to experience. Like I told you, we take things and then adapt them to the enterprise and whatever regulations and everything they need. So we're running really fast in one area, and then adapting it for the companies who need that time and space, and to feel safe adapting and adopting this technology.

Liam Lawson (01:01:43): I want to bring things back to you for my final question. I ask all of my guests the same question. I encourage you to take five to ten seconds before answering. I'll give you no further context on the question, and a lot of people take it different ways, so it's completely up to you how you take it. Roy, why do you do what you do?

Roy Mann (01:02:17): Because I love it. That's always been my passion. I really love technology. I've built things for as long as I can remember, and I really fell in love with computers because I can build things that people actually use and get value from. It was from computer games to everything else I've built. Personally, I'm just really loving what I'm doing. Sorry I didn't take the five seconds, but it was an easy question for me.

Liam Lawson (01:02:53): That's all right. Some people like to think on it more, but you're resolute in your answer, and I have to respect that. Roy, firstly, I want to thank you for your time. I appreciate you coming on with me. It's a pleasure to meet you and spend some time with you. Second, if people want to learn more about who you are, what you do and what you're building, how do they go about doing that?

Roy Mann (01:03:17): That's interesting, because I'm not much of a social presence person, but I have a LinkedIn profile, which I sometimes post on, and I'm trying to do more of that. I think they can also follow Monday Mandy AI on Twitter. It's our agent, and unlike other AI accounts, she's a real live agent. You respond to her, she responds back. So get into that rabbit hole. It's fun. And I think we have a lot more to come on the agent front. There's a whole ecosystem of different products.

Liam Lawson (01:04:09): I can't wait. Our CTO is your biggest fan when it comes to using Monday.com and loves the platform, so we'll be watching, and we'll be using it as well. I can't wait.

Roy Mann (01:04:22): Thank you so much.

Liam Lawson (01:04:24): Thank you, Roy.

Get the five-minute AI briefing

Join 400,000+ business leaders who start their weekday with The AI Report. Subscribe free