Name: Brad Axen, Block
Uploaded: 2026-11-11T00:00:00.000Z
Duration: 18 min 15 s

Clips
More from Block Builderlab Launch Event

Brad Axen

Head of AI Capabilities

Block

play_circle_outline Builderlab and Buzz: Centralized Human-Agent Collaboration and Multiplayer AI Conversations for Organizational Knowledge

play_circle_outline Model-agnostic approach integrating frontier models like OpenAI, Anthropic, Microsoft

play_circle_outline Goose open-source action platform with Builderlab enterprise identity, security, and scalability for Buzz

play_circle_outline Advice for CIOs: avoid many single-purpose agents; consolidate capabilities

19:35

08:26

Info

Brad Axen, Block

Brad Axen

Head of AI Capabilities Block

(INTRO)

Alison Kosik

>> Welcome back to Block Builderlab Launch Day 2026. We're coming to you live from San Francisco. I'm Alison Kosik and I've got Paul Gillin with me. He's the enterprise editor with SiliconANGLE. And joining us now, our next guest is Brad Axen. He's the head of AI capabilities at Block. Welcome to theCUBE.

Brad Axen

>> Yeah, thanks so much for having me on.

Alison Kosik

>> Congratulations on the launch today.

Brad Axen

>> It's really exciting. I'm super happy with the live event and really excited to see what people start doing with the tools.

Alison Kosik

>> Yeah, I'm sure. So let's talk about Builderlab. Walk us through what exactly it is and what did Block learn internally that led you to build kind of an enterprise platform around shared human-agent collaboration?

Brad Axen

>> Yeah, this has definitely come out of a few years of experimentation for us as we've built a really large number of tools and then left a lot of them behind as things kept getting better and better. This is the collection that we're most proud of and the things that we think other people would benefit from. So Builderlab is the full collection of tools. The premier piece is the interface for people working together with each other and AI agents, which is Buzz. And that is specifically a collaboration platform. So a multiplayer conversation, you get to work with your team and then bring AI in when they start to be relevant to the conversation or turn your ideas into actual working software.

Paul Gillin

>> Jack Dorsey said this morning in his initial remarks that one of his greatest frustrations is that so much information in the organization is locked up in the minds of individuals. Going into building Builderlab and Buzz, how did you address that issue? How does Buzz address that issue?

Brad Axen

>> Yeah, I think that's the strongest part about Buzz, is that it's bringing the whole context of your organization together into one system. And before that was kind of incomprehensible, like too many conversations were happening for people to possibly keep up with. And LLMs make that possible today. So if you have that central data asset about how your company is operated, you can tap AI into it. And instead of you doing all the work into telling them how to go solve problems, they already have access to those systems and you're just really telling them the ideas, like what decisions did you make as a human and how do they go make the implementation happen.

Paul Gillin

>> Now, we've had knowledge management tools for 40 years, that typically involved an interview process and a lot of indexing and tagging. How does AI change that equation?

Brad Axen

>> I think it's in removing the need for people to do all of the work to turn the loose conversations into structured outputs. the AI can do that part of the problem, so we can just focus on having the right conversations and let everything else kind of fall out naturally.

Alison Kosik

>> So then how does Builderlab fit alongside platforms from hyperscalers and frontier AI labs like OpenAI, Microsoft, or Anthropic?

Brad Axen

>> Yeah, I think we're really looking to be an interface layer, one above the model layer. So we care a lot that our tools are model agnostic. So when the best model, best new frontier models come out, we're always able to switch to the current best intelligence. So Buzz, for example, can use any AI agent from any of the labs. So it's really one more level on top of that where we want to build on the best things that those companies are providing.

Paul Gillin

>> People in organizations have become accustomed to using software the way the software wants them to use it. So the application dictated what we did. Obviously, you're trying to flip that on its head. When you look a little bit further down the road, what does software look like inside organizations?

Brad Axen

>> Yeah, I think there's so much potential for customization, right? Because you're talking to a coding tool. it can change everything that you're working with to fit the needs that you have for the process you're working on or just your preferences. And the core there is that you need— or you need a core that makes sure that everyone is sharing information when you're doing that. So if you have one version of the tool and I have another, as long as they're talking to each other in a shared language, we can both kind of get the individualized version of it. And so for us, that means that the whole UI of Buzz is editable by you and the AI that you're working with. But there's a shared message layer that lets us all exchange information in a standard way. And that's what is that kind of same organizational context we were talking about before that you can build on top of.

Alison Kosik

>> What is the distinction between your open source projects like goose and Buzz and Builderlab?

Brad Axen

>> Yeah, so we started— related to that question a second ago about the Frontier Labs, we started a lot lower level, we just wanted to make it possible for LLMs to take actions for people. That's what goose was. And now I think of that as more foundational technology that we're building on top of today. So Buzz is the actual interface. That's where we want humans to go interact with the tools, talk to each other. And I think that's a durable layer, the place where we'll just keep improving over time. And then Builderlab is the infrastructure that powers Buzz, the kind of enterprise identity, all of the things that you need as a company. To make those systems secure and scalable.

Paul Gillin

>> This summer we had, of course, the incident where a group of agents, OpenAI agents, teamed up with each other, 7,000 of them, broke out of their guardrails and actually did damage knowing that they were doing the wrong thing. You were probably heads down in the middle of development when this happened. How did that change your thinking about what you were building here?

Brad Axen

>> It absolutely did. And I think for us, as much as possible, we want to have absolute guardrails. So there's soft guardrails that are suggestions to the models. And as we've seen in some of these incidents, those suggestions don't always work. And so a lot of our infrastructure is based on secure primitives that mean it's impossible for it to take certain actions. Like we don't give it the permissions to do certain operations. And that helps us make sure that say, our company, we're not going to have any alignment problems. Like the agents aren't able to do that, but we have to complement that with a really strong security team that's making sure that those protections aren't bypassable. And really for that, we also use AI as part of the solution. So we're trying to keep up with the potential security risks by using AI to find them before they become exploitable.

Alison Kosik

>> Yeah, one of the standout architectural decisions in Builderlab is running these remote agents at all hours, at night, and still doing the work when we're sleeping. But how does that shift in the mental model of enterprise workflows from synchronous prompting to asynchronous delegation?

Brad Axen

>> Yeah, I think this is a really key shift that a lot of companies are going to go through, and it's definitely both the experience of it and the security of it. So like we were just talking about, if you design an environment for the agents to operate in, you have a centralized place to secure compared to them working on your laptop, which is a lot harder. But to your point, it's very much about turning some of the things that can take hours from an interactive thing that's requiring your attention to an asynchronous process. So I think increasingly they're getting so good that you'll give them large assignments, building entire new systems or new applications, and you'll be more of a reviewer and you'll let them run overnight or you'll check on it on your phone and moving them away from local to remote runtimes It just makes everything much more convenient for the people too.

Paul Gillin

>> A lot of organizations are stuck in pilot mode with Agentic projects right now. What have you learned in your experience moving this project along that maybe could benefit them and get them out of spinning their wheels?

Brad Axen

>> Yeah, I think you definitely want to find the piece of value. I think end-to-end solutions can sometimes be challenging because there can be a lot of cruft. You're fighting the system as much as you're benefiting from it. So as soon as you see something is working, and so for a software company like us, the answer to that question was writing code. That was just such an obvious benefit to these AI tools that we just wanted to remove all of the things that got in the way of that value. And as we keep uncovering new similar things, now it's a lot of our systems around information context. Like summarizing what the organization needs to know and giving people the right information at the right time. That's increasingly also valuable. And we want to make sure that you're using a system that gets out of the way of you getting value out of that.

Paul Gillin

>> How do you ensure that the context is right, that they're getting it right?

Brad Axen

>> Yeah, that's definitely a human part of the process. I think if you're an old enough organization, there's so much stuff that's out of date or unmaintained that is a problem. And so the AI is going to come back to you. And this is why it's so critical that humans are still the pilots in this system. But when they do give you those problems, I think they are also enabling you to fix them. So it doesn't take the same amount of work from you as it did. You can work with these AI tools and make sure that they're curating the context in the right way and deleting the information that's now out of date. And I think a tool like Buzz helps remove even the initial part of that problem. you shouldn't have to write documentation for your AI. They should just learn from the conversations that you're having among humans. And since they can see people updating their decisions in real time, they can automatically get rid of that out-of-date context.

Alison Kosik

>> So then how do you— let's go years from now— how do you see this sort of orchestration between humans and agents working together?

Brad Axen

>> I think you'll have to— well, maybe two answers to this question. One is that we're trying to design tools that have the right structure for us to adapt to however the models get better. So it's really hard to predict what the models get better at. So we're looking at more foundational components. And so for us, for example, with Buzz, we think that conversations among people, that's a foundational component that's never going to change. So we're building on top of that. So wherever we think that goes or wherever that goes, we think we can adapt. I will say specifically, though, one of the things I think will happen even sooner than 3 to 5 years, but the next year, is that you're going to have to do less of the driving, in my opinion. I think because they're in the conversation, they can be more proactive so they can jump in when they're needed. You're going to have to do a lot less of actually guiding them along the way and instead just focus on the things that are important for you, like reviewing, decision-making, having taste.

Paul Gillin

>> Those conversations take place in many different venues now. What do you do with the Buzz agents to give them access to all that data?

Brad Axen

>> It's— yeah, it's twofold, right? Because there's the world we came from before we started to move towards Buzz. So there's information in all these other systems. And then now, a big part of the strategy is moving those conversations inside of Buzz, where we think this will be much more convenient, a lot more readily accessible to those AI tools. But because we have those legacy things, we're definitely working a lot through our connectors. So we have a suite of connectors in Builderlab. They can reach into those other information systems and then grab the context that they need as a stopgap. And we've done a lot of work in there to make sure that when you have agents reaching into those information systems, they can only get the data that they're authorized to have.

Paul Gillin

>> Ultimately, does this put you in competition with Slack and Teams and other Jira and other workflow managers?

Brad Axen

>> I think for our company at least, we want to move off of those other communication platforms. And onto something like Buzz because we see the potential for that to give us a lot more leverage. And I think other companies will probably see the same value after trying out some of those tools when AI is much more integrated and a lot of the work that they've had to do to fill up AI context windows becomes an automatic part of the system.

Alison Kosik

>> Can you give us a concrete example of maybe something that may have taken a team several days to accomplish, but that Builderlab can now do in a shorter amount of time?

Brad Axen

>> Yeah, I think we've had a couple of our teams do kind of before and after comparisons when they're building software, and we've seen people make somewhere between 50 to 75% faster releases. So, say you're building a software product, we've seen the time between releases get much quicker. That's not super specific. So, to do something concrete, if we're building features in Cash App, and you come over to Buzz, we can get new app releases out with new features, like new surfaces on the interface. That's all getting a lot faster. And that's not just because of engineers writing code, but it's the collaboration, right? It's the design team and the product team joining that conversation and deciding what to do faster as much as it is about making the changes.

Alison Kosik

>> Yeah, yeah.

Paul Gillin

>> Block's core customers are small businesses, people who are more focused on running their business than on technology. What does this mean to them?

Brad Axen

>> I think those businesses still are groups of people. And we've been designing this software in Buzz to work for teams of as few as just a couple of people all the way up to an enterprise at our scale. So I think we'll be seeing more and more technology coming for Square and small businesses that use the same platform of Buzz but presented in a very comfortable way that's integrated with the Square tools they're already using.

Alison Kosik

>> What do you think is the biggest obstacle that is preventing companies from adopting this kind of agent-driven workflow? Is it technology? Is it trust? Is it organizational values? What's the pushback coming from?

Brad Axen

>> Yeah, I think a year ago it was trust, and I think that's still a big tailwind today or headwind today. You do have to have the process in place to make sure that there's people's judgment at the right times in the system. I feel like personally at Block, we've gotten to a place where we understand that really well. So we're trusting them to do a lot more because we have the processes in place to find when it goes wrong. I think that's critical. And then the second thing I would say is I think of this as a bottleneck problem. So the old bottleneck for a software company like us was implementing software. Now the new bottlenecks are more like reviewing software or coming up with the new ideas. And so you take on those new bottlenecks. And so LLMs solved the software implementation bottleneck. The next one is the collaboration bottleneck. And we think we're making tools that make that a lot easier.

Paul Gillin

>> Have you seen agents come up with valuable new ideas?

Brad Axen

>> They're not great. I think that if you just ask them without any organizational information on what should your product strategy be? I've never seen that be very successful. I think you get more and more interesting idea generation as they get more plugged into the conversations people are having.

Alison Kosik

>> Is this a moment that you're going to look back on where you're thinking that it just fundamentally changed how companies operate? Is this a fundamental turning point, you think?

Brad Axen

>> I really think that moving from these single-player AI tools to multiplayer AI tools is definitely going to be the next step. And I suspect that most people building inside an organization are going to start to find ways to bring AI into the collaboration part and stop having it just be a single-player operation.

Paul Gillin

>> What do you say to people who would say that this is simply another way for you to take my job?

Brad Axen

>> I really— we've talked about this at the event a lot, and I think that it's, as Jack said earlier today, people are the edge. I think that it's really critical that we have an idea or we have a goal, even an ethos that we represent. And it's— these are just tools that we use to express ourselves and to move a little bit faster. So I really think it's critical that we make sure that the people are a major part of this process and we're not trying to holistically replace any of the roles.

Alison Kosik

>> So Block is defining itself as an intelligence company. I'm curious what that actually means. How does Builderlab fit into Block's continued transformation that's underway?

Brad Axen

>> Yeah, I think Builderlab is an attempt at making the technology for becoming an intelligence company. That's really about, as Jack says, we're trying to operate as an intelligence. And a big part of that is that the intelligent tools we have access to have the context of our whole organization. So Builderlab is a collection of tools that make that possible for us.

Alison Kosik

>> So if you're advising an enterprise CIO watching this right now who's stuck in purgatory right now, pilot purgatory with generic generative AI chatbots. What's the single biggest operational habit they need to unlearn today to be ready for autonomous agent teams?

Brad Axen

>> I think that you want to avoid having end-to-end agent solutions in every single different tool that you use. So if you're switching, if you're the person sitting between 20 different AI agents that can only do one thing at a time, there's no way that you're going to get a significant speedup and you're just going to be stuck working with the limitations that they all have. I think you want to build systems, and I think Buzz and Builderlab are one example, but there are many more that give all of those AI tools a whole set of capabilities so that when you have a problem, they can put together the solution from many parts.

Alison Kosik

>> So what do you want folks here today at this launch to walk away with? What are some of the things you want them thinking about?

Brad Axen

>> I think it's collaboration. And I would be— I would strongly encourage people to go try Buzz. It's public now, so you can give it a shot. I think that the collaboration elements of what we're building are definitely the next generation of how we work with AI. And I think it can really speed up the way that you work together with other humans.

Alison Kosik

>> Well, thanks so much for stopping by theCUBE. Great conversation, really appreciate it.

Brad Axen

>> Yeah, thank you so much for having me.

Alison Kosik

>> And you're watching theCUBE, the leader in live technology coverage and in-depth expert analysis. We'll be right back from San Francisco.

Brad Axen, Block

(INTRO)

Alison Kosik

Brad Axen

>> Yeah, thanks so much for having me on.

Alison Kosik

>> Congratulations on the launch today.

Brad Axen

>> It's really exciting. I'm super happy with the live event and really excited to see what people start doing with the tools.

Alison Kosik

Brad Axen

Paul Gillin

Brad Axen

Paul Gillin

>> Now, we've had knowledge management tools for 40 years, that typically involved an interview process and a lot of indexing and tagging. How does AI change that equation?

Brad Axen

Alison Kosik

>> So then how does Builderlab fit alongside platforms from hyperscalers and frontier AI labs like OpenAI, Microsoft, or Anthropic?

Brad Axen

Paul Gillin

Brad Axen

Alison Kosik

>> What is the distinction between your open source projects like goose and Buzz and Builderlab?

Brad Axen

Paul Gillin

Brad Axen

Alison Kosik

Brad Axen

Paul Gillin

Brad Axen

Paul Gillin

>> How do you ensure that the context is right, that they're getting it right?

Brad Axen

Alison Kosik

>> So then how do you— let's go years from now— how do you see this sort of orchestration between humans and agents working together?

Brad Axen

Paul Gillin

>> Those conversations take place in many different venues now. What do you do with the Buzz agents to give them access to all that data?

Brad Axen

Paul Gillin

>> Ultimately, does this put you in competition with Slack and Teams and other Jira and other workflow managers?

Brad Axen

Alison Kosik

>> Can you give us a concrete example of maybe something that may have taken a team several days to accomplish, but that Builderlab can now do in a shorter amount of time?

Brad Axen

Alison Kosik

>> Yeah, yeah.

Paul Gillin

>> Block's core customers are small businesses, people who are more focused on running their business than on technology. What does this mean to them?

Brad Axen

Alison Kosik

Brad Axen

Paul Gillin

>> Have you seen agents come up with valuable new ideas?

Brad Axen

Alison Kosik

>> Is this a moment that you're going to look back on where you're thinking that it just fundamentally changed how companies operate? Is this a fundamental turning point, you think?

Brad Axen

Paul Gillin

>> What do you say to people who would say that this is simply another way for you to take my job?

Brad Axen

Alison Kosik

>> So Block is defining itself as an intelligence company. I'm curious what that actually means. How does Builderlab fit into Block's continued transformation that's underway?

Brad Axen

Alison Kosik

Brad Axen

Alison Kosik

>> So what do you want folks here today at this launch to walk away with? What are some of the things you want them thinking about?

Brad Axen

Alison Kosik

>> Well, thanks so much for stopping by theCUBE. Great conversation, really appreciate it.

Brad Axen

>> Yeah, thank you so much for having me.

Alison Kosik

>> And you're watching theCUBE, the leader in live technology coverage and in-depth expert analysis. We'll be right back from San Francisco.