The Speed of Action: Bruno Aziza, IBM

Companies spend years building dashboards that told them what to know. Now they’re building AI that tells them what to think.

But what companies actually need is AI that takes action on their best ideas.

Bruno Aziza, VP, Enterprise Software at IBM, shares why so many AI deployments fail — and the frameworks that separate successful implementations from expensive pilot programs.

His insight: one customer now has "50 humans and 150 agents" on their team . The question isn't whether agents will reshape work. It's whether we're building agents that actually do the work.

Key frameworks from this conversation:

→ Agent Minus: The essential pre-agent infrastructure (rules, tools, workflows) that make agents work 

→ Agent Plus: Orchestrating hundreds of agents across vendors and systems 

→ FOMO vs. FOMU: The psychological traps killing AI adoption

The rule Bruno lives by: "Process useful: automate. Process not useful: eliminate."

Most organizations skip step two.

Subscribe to Humans of AI for more stories from people navigating the intersection of business and artificial intelligence.

Watch the full video interview on the WRITER YouTube channel for bonus content and deeper insights.

Learn more about WRITER at writer.com. 

Transcript

SPEAKER_02 0:04

This gentleman comes to me and he says, We have 20,000 dashboards that we built over the last year. And I said, Oh, how many of them are used by how many people? He said, Out of the 20,000 dashboards I have, 18,000 of them are used by less than eight users.

SPEAKER_01 0:19

20,000 dashboards. 18,000 of them used by fewer than eight people.

SPEAKER_02 0:25

The only thing that came through my mind here, I don't know that it's a quote by Einstein, but he said, don't let information just become knowledge.

SPEAKER_01 0:34

This was a few years ago before most companies were even talking about generative AI. Now everyone's building AI systems. And Bruno's asking the same question. Are we about to make the same mistake at scale? Are we building smarter intelligence when what companies actually need is action? Welcome to Humans of AI. I'm Alora Weaver. That's Bruno Aziza, group vice president of data, business intelligence, and AI strategy at IBM. Which means his job is talking to the companies actually running AI in production, not experimenting, running it at scale. Over the last year, he's had about 200 of these conversations. Companies like Heineken, PepsiCo, NatWest, the ones with real stakes, real budgets, real consequences if it doesn't work. And this story about the dashboards that nobody uses, it keeps coming up in different forms.

SPEAKER_02 1:34

And in a lot of organizations, we're really happy with the dashboards to scorecards, we do reports and all that. But if all that is is knowledge, you're not really advancing. The reason for why you had this dashboard in the first place is to understand what you should do, not what you should know.

SPEAKER_01 1:50

So all these dashboards, they're answering a question nobody asked. They're showing you what's happening, but they're not helping you do anything about it. And if you're listening to this thinking, we might have that problem. You probably do. Because the question isn't whether your reports are beautiful, it's whether they changed a single decision. Now, fast forward. Companies are building AI agents, intelligent systems that can actually do things, not just show you things. But Bruno's watching closely to see if we've learned the lesson. Because building smarter intelligence is easy. Building systems that drive action, that's the hard part.

SPEAKER_02 2:37

A customer I just met recently, and I was asking them about how do you accelerate the path to resolution and action? And we came to, well, how many employees do you have? And he smiled at me and he says, You know, I have 200. Oh, I said, What roles? He said, Well, I have 50 humans and I have 150 agents.

SPEAKER_01 2:57

I have so many questions about that company. What industry? What do the agents actually do? What are the 50 humans spending their time on? But even without those details, that number shifts everything. This isn't some distant future scenario. This is someone's Tuesday.

SPEAKER_02 3:14

I thought that the way that we then go to action changes, right? Because in the old method, we would have a system of record, a system of engagement, system of knowledge and intelligence, and they would stop there, and we forgot the most important thing, which is the action. Now the question is who takes the action?

SPEAKER_01 3:32

That's the shift. For decades we built the systems that stop at intelligence. Now we have AI agents that could actually take action. But here's Bruno's question: Are we building them to take action? Or are we just building smarter intelligence systems and calling them agents? Bruno didn't start out thinking this way. He's worked in data and AI for 25 years for startups, companies like Microsoft, Google, Oracle, now IBM. But here's something that tells you how he thinks about transformation.

SPEAKER_02 4:09

You know, I write this blog, it's called Nobody Cares, where I try to scale my leadership lessons. And it's really around that is how do we codify a way for ourselves to try and see beyond ourself and how we can move beyond. I've even written a user manual for myself. So people that work for me, they realize that I work for them and they should use me as a product. It's called How to Bruno. And I named it that because one of my employees told me to name it like that.

SPEAKER_01 4:39

A user manual for himself called How to Bruno. So his team could use him like a product. That's not just clever. That's someone who's serious about systems thinking, about orchestration, about making sure all the pieces, including himself, work together effectively. And that mindset shows up in everything he's learned about AI transformation.

SPEAKER_00 5:08

Building enterprise-grade AI shouldn't be complicated. It should just work the right way. At writer.com, we don't do everything. We do one thing. We build enterprise AI that unites business and IT. Business teams build your own AI agents. No code required. IT teams manage just one platform, not a plethora of point solutions. Writer.com creates AI tools that are safe, scalable, and smarter every time you use them. That's why Accenture, Qualcomm, Vanguard, and hundreds more aren't just doing enterprise AI the right way. They're doing it the writer way. Book your demo at writer.com slash demo.

SPEAKER_01 5:53

Gartner just released a prediction. 40% of agentic AI projects will fail by 2027.

SPEAKER_02 6:00

40%, I actually think it's pretty low. I would have expected it to be higher. Because most things that are worth doing are not easy, right? So if it was 100% a success, I would be worried.

SPEAKER_01 6:12

Wait, he thinks 40% is low?

SPEAKER_02 6:15

Second, is what is the timeline on which you're measuring this failure and success rate? Uh if you look at the history of the internet, it took us, I think, 10 years to adopt the internet. Truly adopt it. So right, it goes back to this idea that we completely underestimate the transformative power of a technology over 10 years, we overestimate it over a short period of time.

SPEAKER_01 6:35

Which means if you're measuring AI success on a two-year clock, you're measuring the wrong thing. You're going to see a lot of so-called failures that are actually just slower than expected.

SPEAKER_02 6:48

Most organizations I talk to, they are stuck between two acronyms. The first one is FOMO, fear of missing out. And so you got to resist that, right? We're trying to be efficient, useful, and then if we can be cool, that's the benefit. And then the second one is FOMU. F O M U is the fear of messing up.

SPEAKER_01 7:09

So you're either rushing into AI because everyone else is, or you're paralyzed because you might pick the wrong thing. And either way, fear is making the decision instead of strategy. And if that feels uncomfortably familiar, if every leadership meeting oscillates between those two poles, Bruno's saying that's not a character flaw. That's the actual strategic problem you need to solve.

SPEAKER_02 7:34

If I'm talking to you as a leader and many of the CIOs and technology leaders I talk to, is the truth is somewhere between these two, right? And you gotta go out and then identify what use case is gonna be most valuable for you.

SPEAKER_01 7:47

How do you find that middle ground when you're caught between those fears?

SPEAKER_02 7:51

In a lot of these conversations, it's more about the emotional attachment to what's going to make your career and what is not. So my conversations with leaders are let's define the job. What is it that's going to drive value for you in your organization? And what's going to help you here is is your principles.

SPEAKER_01 8:11

Not the latest model, not what your competitor is doing.

SPEAKER_02 8:15

A great example is model strategy today, right? 18 months ago, everyone's talking about large language models, and now we're starting to see organizations first having a hard time deciding which large language model to go for, and people realizing wait, small language models could be helpful. The principle that would have helped you at the beginning is how do I build with agility? If I don't know what the best model is tomorrow, what size, which provider, how do I work in a way where I can mix and match, change quickly and adapt.

SPEAKER_01 8:50

If you'd started with the principle, build with agility, expect things to change, you wouldn't be stuck now trying to unwind a decision you made 18 months ago.

SPEAKER_02 9:00

The ones that just did not have that principle might have gone down a silo, and now they're figuring out how to get out of it. It's really hard.

SPEAKER_01 9:07

This idea that principles matter more than tactics, Bruno didn't learn it at IBM or Google. He learned it when he was one of six people at a startup, five developers and him. No customers, no database, no website, just code and an idea.

SPEAKER_02 9:26

It was very much a David and Goliath. In fact, last week I was at an event with Malcolm Gladwell, and I reminded him that 15 years ago, Malcolm and I worked together at MIT. In fact, I have the video, so I pulled up the video, and when I did the startup I'm about to talk about, I sent him a message and I said, I just bought your book about David and Goliath. Would you sign it for me? So he signed it.

SPEAKER_01 9:51

Malcolm Gladwell signed his copy of David and Goliath because Bruno was about to live it. The startup was the big data space. Think massive databases that companies use to process huge amounts of information. Three massive competitors dominated the market. Bruno had zero customers to survey about what the market needed. So he did something counterintuitive.

SPEAKER_02 10:17

Because I didn't have a single customer in my database, I looked at the industry and I looked at identity differentiation. Why did we want to start this company? A lot of it was about neutrality, is that we had access to information that was cutting across most vendors inside that ecosystem. I created the SurveyMonkey survey. But what happened is I came up with 20 questions that I felt everybody should have the answers to. And then I went to the CMOs of these three companies and I said, you can collect that information by yourself because if you send that to your people, they will know this is biased because you're one of the three vendors. But if I do that for you, one, you will get the answer naturally.

SPEAKER_01 10:59

He convinced his three biggest competitors to let him survey their customers. Think about that for a second. He had nothing. They had everything. And he got them to give him access to thousands of potential customers by being the only one who could ask the questions everyone wanted answered without looking biased.

SPEAKER_02 11:21

They taught me a few things is that it really doesn't matter how small you are. What matters is where are the breaking points in the market? How can you serve your customers the most? Because while it's a value to my partners, it's only also incredible value for the people in this market. Everybody's wondering, are you behind? Am I ahead? Well, I think most entrepreneurs they start with who do I compete with? And sometimes they get a little too obsessed about that. And for me, competition is a data point. The data point I care about the most is what's the customer problem, what's the job to be done, how do I become the company that is most helpful to them.

SPEAKER_01 11:59

Not how do we beat the competition, not how do we prove we're smarter, what's the actual problem people need solved.

SPEAKER_02 12:09

My job is to make sure that I can cooperate with the ecosystem and provide the best value to the customer. So in a way, I'm really not competing against anybody else. I'm just kind of competing against myself in a way and who I was prior to the version of the product I have.

SPEAKER_01 12:25

So here's where Bruno sees people getting stuck with AI right now. Everyone's obsessed with building agents, making them smarter, making them autonomous. And that's not wrong exactly, but it's incomplete.

SPEAKER_02 12:39

Every new technology looks like an excuse to try something new. The first question you should ask is, what are we trying to accomplish? So the rule should be if process useful, automate, if process not useful, eliminate. And we forget that rule.

SPEAKER_01 12:57

Most people hear AI and think, what can we automate? But maybe the question is, what should we stop doing altogether?

SPEAKER_02 13:06

Most vendors out there will tell you it's a revolution. You gotta erase everything you got and completely reinvent. And I actually think the truth is not that at all. My observation with customers is that it's an evolution. When I worked at Microsoft, Bill Gates had this thing he printed, he said, We always overestimate the change that will occur in the next two years, but we underestimate the change that will in fact occur in the next ten.

SPEAKER_01 13:30

Evolution, not revolution. That's not about going slow. It's about not destroying the things that actually work. And if you've been drowning in everything is different now narratives, this is permission to stop. You don't have to blow everything up. You have to build on what works.

SPEAKER_02 13:50

This idea of mix and match, it's kind of what I call agent minus. It's the agent, but there's also a whole set of technologies before the agent, like rules tools, workflows that need to marry with it.

SPEAKER_01 14:03

Agent minus. Approval chains, diagnostic processes, routing logic. And those aren't broken.

SPEAKER_02 14:15

There's also a whole set of technologies before the agent, like rules tools, workflows that need to marry with it.

SPEAKER_01 14:22

Heineken didn't start with agents. They started by harmonizing 100,000 data elements, getting their existing systems to talk to each other. Then they added agents on top. The agents made the existing work better, but they didn't replace it. Then there's agent plus, and this is where Bruno thinks most companies are headed for trouble.

SPEAKER_02 14:46

While everyone today is obsessed about how easy it should be to build agents, and we are going to get there. If you're listening to us and you're trying to build agents and run them into production, I know you could probably build the first agent, but the minute it runs, now you're starting to get into the meat of the issue.

SPEAKER_01 15:05

Building one agent? That's the easy part.

SPEAKER_02 15:07

Agent Plus is this idea that if you live in a world that has hundreds of agents, agents that might be built by multiple vendors, some hyperscalers, some that are not, maybe an application vendor, maybe your developers or hand coding uh agents, then the difficulty is going to be on top of that, which is going to be orchestration of these agents. How are they supposed to talk to each other?

SPEAKER_01 15:35

Because in real company, you're not going to have one agent. You're going to have agents built by different teams running in different clouds using different models, and somehow they all need to work together. This agent minus, agent plus framework, this might be the conversation you need to have with your team this week. Because right now you're probably asking, should we build agents? But the question is, how do they orchestrate with what already works?

SPEAKER_02 16:03

In order to enable the system of action, you're you're probably going to need three big areas of capabilities. The first one is autonomy. So do these agents have the right permission to operate with high autonomy? What are the feedback loops? What are the risk profiles? The second bit is agency. Let's say that they have autonomy and their compliance. Is it purpose aligned to an objective? And do they have the agency to complete the task with optimizing for efficiency, optimizing for safety, uh satisfaction, revenue, outcome, and all that? And then the third bit is accountability. An action that's been taken. Is this action auditable? Was it done with transparency? Was biased, you know, checked?

SPEAKER_01 16:55

So it's not enough to ask, can the agent do the thing? You have to ask, is it allowed to? Is it solving for the right outcome? And can you explain what it did to someone who's going to hold you accountable for the result? These aren't just governance questions. These are the questions that keep you out of the news for bad reasons. Because when an AI initiative goes wrong, someone's going to ask all three.

SPEAKER_02 17:21

Don't be overobsessed about the agent creation, because the next thing here is the orchestration of those agents, the operations of those agents, and to run a system of action effectively, you're gonna have to come up with some principles, rules. Of course, there's technology that's required, but the idea is if winning means designing a system of action, there's some rules. Uh and uh you should be writing them right now.

SPEAKER_01 17:47

Because the rules about how agents work together, that's where the value is. Companies like those using writers agentic compact aren't focused on making individual agents smarter, they're focused on the layer above that, the orchestration. So, what does leadership even look like when your team is mostly agents?

SPEAKER_02 18:10

The best leaders do three things. One, they identify the superpowers of their employees. They really know, okay, when I need to solve this problem, the best person on the team, hopefully the best person in the world, is this person, this team. Second, is they create an environment in which that superpower could be applied with maximum impact. And that's where I think a lot of leaders fail, is that they have the great players, but they have not figured out a way to just optimize. Either they can't collaborate, either the company does enable you to create an environment, or maybe you're unable to communicate the coaching feedback that individual, that team needs. And then the third idea is the inspiration, is it's one thing for me to see what could get someone to the next level, sometimes things they can't see themselves.

SPEAKER_01 18:58

That third one stops me because it's not about motivation or cheerleading, it's about seeing potential that the person can't see yet. The same way Bruno created how to Bruno, helping his team see how to use him effectively, even in ways he might not have articulated on his own. And those same three things apply to agents.

SPEAKER_02 19:21

In the case of an agenc situation, what you really have is you really have to identify where is this agentic system going to be the best in the world? What is the superpower of this agent? Sometimes it might not be make the decision all the way, sometimes it might be checking on the decisions of other agents or checking on the decisions of other humans.

SPEAKER_01 19:40

Maybe the agent's superpower isn't autonomy, maybe it's validation or pattern recognition or connecting things humans miss.

SPEAKER_02 19:49

Can this agent in fact drive the process all the way? If not, what are the places where it's the best at? And what environment can I create in order for that agent to have full autonomy and full accountability? And ultimately you have to decide, right? Because if the accountability is to the human, the human needs to make the decision.

SPEAKER_01 20:06

That's not about whether you trust the AI, it's about organizational reality. When something goes wrong, someone has to be accountable. And right now, that someone is human.

SPEAKER_02 20:17

The science of leadership does not change, right? It's still the superpower, the environment, the inspiration. The art of it might evolve a bit.

SPEAKER_01 20:27

The fundamentals don't change. You still need to know what your team is good at, give them room to do it, and help them see what they're capable of. It's just that now, some of your team members might be agents. And if that's grounding in a way that makes you feel like you can actually do this, that's the point. You don't need all the answers. You need a methodology for navigating this that builds on what you already know. Bruno writes a blog called Nobody Cares. The name is deliberate. Because when you're trying to scale leadership lessons, when you're trying to help people see what they're capable of, what matters isn't whether people care about your story. What matters is whether they can use what you've learned. He spent 25 years in data and AI. And what he's learned is this. The problem isn't that we don't have smart enough systems. It's that we keep stopping right before the thing that matters: action. Bruno's not here to make you panic about AI transformation. He's here to show you there's a methodology for navigating it. You don't need to pretend you have all the answers. You just need to know what question you're actually solving for. Bruno is our last guest of season four. And what a season it's been. We've been in the rooms where this transformation is actually happening. With the chief marketing officer of Writer, who talked about the metagame of AI that most leaders miss. With Logitech's global head of AI who trained 800 colleagues across 46 countries and learned that adoption isn't a technology problem, it's a human one. With the CEO of Front, who discovered what it really means to lead a company through this moment. With leaders from IBM, Meta, Forester Research, and more. If you've been listening and something landed, share it with someone who needs it. The leaders figuring out AI transformation aren't reading the same vendor white papers. They're having conversations like this one. Subscribe so you don't miss the next one. And if this episode gave you a framework worth keeping, tell us. Leave a review. It takes 30 seconds and it's how other people like you find us. We read everyone. Thanks to Bruno Aziza for the conversation. This is Humans of AI. I'm Alora Weaver.