I came back from AI4 in Las Vegas with an uncomfortable feeling.
While we spent the last few years discussing how quickly artificial intelligence was evolving, the real problem began to emerge on the other side. AI got fast. Companies are still slow.
AI4 is described by its organizers as the largest artificial intelligence conference in America. In 2026, more than 12,000 attendees, 1,000 speakers, 400 exhibitors and representatives from 90 countries gathered at The Venetian in Las Vegas.
The numbers are impressive, but they do not fully explain the importance of the event. What makes AI4 so relevant is the mix of industries, experiences and levels of maturity. On the same day, you could hear discussions about chips, infrastructure, agents, retail, marketing, security, autonomous mobility, small businesses, world models and applications already operating inside major organizations.
I also had the privilege of taking the stage to moderate the panel “Building Smarter Retail Experiences in the AI World”, with representatives from Ace Hardware (USA), Coppel (MEX), AREA15 (USA) and Vapi (FRA).
Being onstage gives you a different perspective. You do not simply listen to answers. You have to connect different points of view, notice contradictions and find a common thread while the conversation unfolds.
And this is my main takeaway from AI4 2026:
Artificial intelligence is no longer a conversation about possibilities. It is now a conversation about the ability to EXECUTE.
The curiosity phase is over
For a while, simply experimenting with AI was enough for a company to look innovative. Businesses created pilots, presentations and proofs of concept. The question was, “What can this technology do?”
At AI4, the question had changed: “What result is it producing?”
In one panel, someone used the expression “death by a thousand pilots.” Another company said it had been running 76 pilots at the same time. The problem is no longer creating experiments. It is deciding which ones deserve to scale, which ones duplicate work another team is already doing and which ones should simply be shut down.
In other words, a company can generate a huge number of initiatives without transforming a single important process.
AI is not a toy. It is a tool. And results come first.
That means measuring what actually matters: time saved, costs reduced, revenue increased, risk controlled, team productivity or a tangible improvement in the customer experience.
If an AI project cannot explain which problem it solves and which result it intends to change, it may still be nothing more than an interesting demonstration.
Technology accelerated. The organization did not
Content, code, analysis and prototypes that once took weeks can now be produced in hours. But the rest of the company has not gained the same speed.
Approvals, compliance, integration, security, legacy systems, budgets, internal disputes and resistance to change are all still there. In one of the panels, a speaker said that, in large companies, around 90% of a project’s time is not necessarily spent on coding, but on decisions, governance, security and everything required to put something into production.
This was one of the strongest points of the event. The bottleneck is moving.
When producing becomes easier, deciding becomes more important. When coding accelerates, integration and judgment begin to limit progress. When any team can create a solution, the company must learn how to prioritize, standardize and let go.
The risk is no longer just falling behind on technology. It is having fast technology trapped inside a company that continues to operate at its old speed.
What to buy, what to create and what to own
This discussion came through very strongly in the panel I moderated and connected with several other conversations throughout the event.
To me, an AI strategy does not begin with the question, “Which model should we use?” It begins with three decisions: what to buy, what to create and what to own.
Buy what has already become an available market capability. Mature solutions, generic resources and infrastructure that do not, by themselves, create a competitive advantage.
Create what differentiates the business. Processes, experiences and applications connected to customer knowledge, operations and specific details that no ready made tool can fully understand.
Own what the company cannot leave under someone else’s control. Data, context, rules, governance, intellectual property, evaluations and orchestration capabilities.
Creating does not necessarily mean developing everything internally. Owning does not mean building everything alone. A company can create with partners while retaining control over its intelligence and differentiation.
Perhaps the best summary is:
Buy capability. Create differentiation. Keep control.
The winners will not be the companies that develop everything, nor those that buy everything ready made. They will be the ones that know how to make this choice clearly.
The size of the company changes the role of AI
Another important conclusion is that there is no single path to maturity.
For a small business, AI can be a tool or an assistant for specific tasks. It can help organize a sales process, analyze customer interactions, produce content, configure a system or eliminate hours of operational work. Small businesses have an important advantage: they can make decisions and change direction faster.
In larger and more complex operations, AI begins to move beyond being just a tool and starts functioning as part of the business infrastructure.
Ace Hardware showed that a business does not need to be born as a technology company to get started successfully. Its Hey ARMA assistant is already present in more than 2,300 stores, supported by an internal network of around 50 professionals who serve as AI Champions.
Coppel presented the other side of this curve. When a company brings together retail, banking and retirement services, serves millions of customers and operates with enormous complexity, there comes a point when simply adding AI to existing processes is no longer enough. The company needs to master data, platforms and technology decisions. Its internal Genius platform already has 15,000 registered users and 10,000 daily active users.
Both views are correct. They simply represent different points on the same curve.
You do not need to be a technology company to begin. But depending on your scale and complexity, you may need to build strong technology capabilities to keep moving forward.
AI is moving beyond the screen
Perhaps this was the most important development for anyone who still associates artificial intelligence with a text box.
Chat was the front door. It will not be the whole house.
AI4 showcased agents capable of executing tasks, updating systems, protecting applications, making decisions and completing transactions. It showed Waymo treating artificial intelligence as a mobility operation in the physical world. It showed Runway moving beyond generated video toward what are known as world models, which can learn relationships involving space, motion and physics. It showed AREA15 connecting ticketing, capacity, virtual queues and itineraries within a physical experience.
Voice also returned as an important interface. Not as the old menu tree in a call center, but as a conversation capable of understanding context, taking action and transferring the customer to a person when needed.
For many Brazilian companies, AI is still concentrated on producing text, images and analysis. At the event, however, the frontier had already moved somewhere else: AI that executes, interacts and intervenes in the real world.
There is no artificial intelligence without infrastructure
Behind a simple interface sits an increasingly complex structure.
Proprietary models, open models, chips, data centers, energy, security, connectors, inference and cost per task appeared in nearly every mature discussion.
The idea of choosing a single model for the entire company is beginning to make less sense. Different tasks require different combinations of cost, speed, security and performance. The work becomes less about finding “the best AI” and more about orchestrating the right capability for each problem.
There was also a clear shift in the way cost is measured. A cheap token does not necessarily produce a cheap outcome. What matters is how much useful intelligence and business value the company can generate from its investment.
This discussion is still limited in Brazil. We talk extensively about which tool is being used, but very little about the architecture that will allow it to operate safely, at scale and with economic viability.
People are still the hardest part
The further the technology discussion advanced, the more the human factor returned.
The companies that demonstrated consistent adoption did not simply distribute licenses. They created internal networks, involved business experts, trained teams, redesigned processes and established clear criteria for use.
AI without a change in behavior becomes just another forgotten tool. AI without business knowledge automates what should not be automated. AI without judgment accelerates bad decisions.
Technical knowledge remains important, but the value of people who deeply understand the problem, know how to ask good questions and can evaluate the quality of the outcome continues to grow.
At the same time, security, trust and accessibility are no longer being treated as later concerns. They are becoming part of the initial design. A fast but insecure agent is not efficient. An intelligent experience that excludes some people is not truly intelligent.
Why being there mattered so much
You can follow announcements, watch videos and read good summaries from a distance. But some things only become visible when you are actually at the event.
In the sessions, you get the content. In the hallways, you see the level of maturity. Onstage, you understand where the different perspectives converge and where they still conflict.
Being at AI4 allowed me to compare what many companies are still discussing with what others are already operating. More importantly, it helped me recognize which questions are beginning to feel outdated.
Perhaps the main question is no longer, “What can AI do for my company?”
The question now is:
What does my company need to change to do something relevant with AI?
That is why events like AI4 matter. They do not simply show us the future. They reveal the distance between your company’s present and the present of those that have already begun to operate differently.
Onstage, I joked that next year I want to see even more Brazilians in the audience and on the stage. After everything I saw, it no longer feels like just a joke.
Artificial intelligence is moving too quickly to be understood only through social media feeds, short videos and summaries. At some point, you need to be where the questions that will define the next few years are being asked.
I hope to see you at the next edition.
And if this article made you think, make it happen.
