What changes when AI gains tools to act

Originally published on NewHack, on November 9, 2026.

The opening of Web Summit Lisboa connected agents, public services, and robotics. For entrepreneurs, the discussion comes down to redesigning processes, earning trust, and delivering results that justify the investment.

The opening of Web Summit Lisboa, on the evening of November 9, brought together subjects that might seem unrelated. Agents capable of executing tasks, state transformation, Chinese companies, internet commerce, and humanoid robots shared the stage.

To me, there was a connection between these conversations. Artificial intelligence is gaining access to the places where work happens. It can consult information, operate tools, communicate with other people, and, within defined limits, take actions.

This changes the question for anyone building a company. Beyond choosing the model, we need to understand what work can be delegated, how to monitor its execution, and what makes the result trustworthy.

When the agent gets an email address

Tao Zhang's participation, co-founder of Manus, brought this shift into an everyday situation. A person starts the day with several pending items and needs to turn each one into organized work.

Delegating requires gathering context, clarifying the expected outcome, identifying dependencies, and tracking what happens afterward. When part of this is handed to an agent, the potential gain goes beyond the time needed to write a message.

Manus is exploring this direction with Cue, a personal agents app introduced alongside Manus 2.0. According to the company, agents receive their own identity, with an email, phone, wallet, and computer, and can make payments within the budget set by the user. Official presentation of Manus 2.0 and Cue.

These features bring the agent closer to the tools we use to work and consume. They also make more pressing the questions a company needs to answer before expanding autonomy.

Who authorized a purchase? What information can be shared? When does the agent need to ask for help? How do you correct an action that produced an unexpected result?

In practice, trust becomes part of the product. The user needs to understand what they delegated, track progress, and be able to intervene.

Simplify the process before putting AI into it

One of the most applicable messages from the opening came from the discussion about the State. Portugal's Minister for State Reform, Gonçalo Matias, placed process simplification ahead of digitalization.

This order matters inside a company too.

Imagine a registration form that asks for the same information across three departments. Each area checks a different version, and someone spends the day routing documents between them. An agent can speed up that routing. But we can also ask why the customer needs to repeat the registration at all.

When designing a solution, it's worth reconstructing the entire path. What triggers the request? What information already exists? Who can decide? Which steps meet a real need and which ones simply stuck around out of habit?

The proposal of agents coordinating public services, such as those related to the birth of a child or the opening of a business, points to an experience organized around the person's need.

For startups, I see a similar opportunity. Knowing a process deeply allows you to find steps that can be merged, moved earlier, or eliminated. That knowledge helps turn technical capability into a delivery the customer can actually perceive.

The next customer may arrive accompanied by an agent

In the panel with Matthew Prince, of Cloudflare, Anton Osika, of Lovable, and Richard Socher, of Recursive, another consequence of this shift came up. The internet is also being used by agents that look for information and carry out tasks on behalf of people. Participants in the opening.

Prince raised a concern about how this movement might affect company discovery and favor businesses that are already established.

My reading is that this deserves the attention of any founder who depends on digital acquisition.

Today, a company tries to show up in search, build reputation, and convince someone to visit its site. When the user delegates a search or purchase, part of that journey runs through a system that compares options according to the instructions it received.

Price, availability, commercial terms, and evidence about the product need to be accessible and consistent. The ability to integrate services also gains importance.

For a smaller business, this can open up new ways of being found. It can also increase dependence on whoever decides which options get presented to the customer.

I would start by looking at a concrete journey. If someone asked an agent to find a solution like yours, what information would be available? What could it verify? At what point would it need to call in a person?

China broadens the repertoire of those who build

Paddy Cosgrave highlighted the role of Chinese companies in the future of technology. That presence also showed up in the program, with Manus, robotics companies, and other firms.

For entrepreneurs in Brazil, I see value in broadening the repertoire of references.

There are different lessons in building models, in integrating software and hardware, in distribution, and in the pursuit of efficiency. Observing how a company works under certain constraints can help formulate better questions about your own business.

The exercise requires context. A result shown on stage depends on the type of task, the environment, and the test conditions. When analyzing a case, I want to understand how much it costs to deliver with the required quality and how much effort still falls on the team.

That's a more useful comparison for making decisions than tracking a model's position on a ranking in isolation.

Robots take the discussion into physical operations

The opening also made room for humanoids and for the proposal of a vehicle with a modular aircraft. These presentations make visible an ambition that runs through the event: taking artificial intelligence beyond digital interfaces.

In the physical world, delivery depends on a larger combination of factors. A robot needs to perceive its environment, manipulate objects, and execute movements. It also needs to operate safely, receive maintenance, and keep functioning for as long as needed.

To evaluate an opportunity, I would look for a well-defined task. Where is there repetitive, heavy, or hard-to-fill work? How is that work done today? Is the environment predictable? How much human effort will be needed to accompany the machine?

A demonstration helps you understand a capability. The routine of an operation reveals what it will take to turn it into a business.

There's room for companies that build applications, integration, maintenance, training, and monitoring. Whoever knows a specific operation well can find opportunities in that gap between available technology and everyday use.

The evaluation of people is also starting to change

Tao Zhang described a way of evaluating candidates by their ability to solve problems while working with AI. That brings the discussion about agents into hiring itself.

When tools start carrying out part of the implementation, it becomes more important to observe how someone defines the problem, provides context, judges alternatives, and checks the result.

A delivery that looks ready can hide fragile decisions. That's why following the work process helps you understand what the person noticed, what they questioned, and how they reacted when the first attempt failed.

I see in this discussion a competency we can already develop within teams: learning to delegate with clarity and to review with judgment.

Where I would look for value after the opening

I left these conversations thinking about whoever manages to connect a need to a complete delivery.

It could be a company that reduces the effort of starting a business, a tool that prepares a task with enough context for someone else to execute it, or a solution that solves a physical problem in a specific environment.

In all these cases, there are questions that help guide the first project. Who lives with the problem? How much does it cost to solve today? What needs to change in the process? How will the customer recognize that the solution worked?

That's where I want to focus my attention in the coming days of Web Summit. On the startups that can explain the work they're taking on, the conditions needed to do it well, and the evidence that it makes someone's life better.