This week, people called for slowing AI. Here is what we think at Wiwo.

It is worth reading the fine print. The call to slow down concerned the frontier race: self-improving models and swarms of autonomous agents with nobody in charge. Nobody asked to stop the AI that reviews mammograms today, teaches alongside teachers or gives hours back to a small business.

And there is a detail that got lost in the headlines. The researcher Nuria Oliver said it clearly: this week's incidents do not demonstrate destructive intelligence; they demonstrate the danger of allowing autonomous actions without effective human supervision. In her words, no AI should independently execute tasks affecting millions of people.

In other words, the remedy proposed by the critics themselves is human supervision. Welcome. We never let go of it

In Wiwo we believe the future can also turn out well. This is not a hope: it is what AI, with people in charge, is already doing today.

Health. In Sweden, the MASAI trial followed over 100,000 women and published it The Lancet in January 2026: with AI support, mammogram screening detected more cancers, with 27% fewer aggressive cancers and a 44% reduction in radiologists' reading workload. Radiologists are still there, making decisions. A model published in Gut this year detects early pancreatic cancer at twice the rate of current methods; it is still undergoing validation, but that is the direction. AlphaFold made more than 200 million protein structures available to science and earned them the 2024 Nobel Prize in Chemistry to its creators; Isomorphic Labs, its spin-off, takes its first AI-designed cancer drugs into human trials. At MIT, a generative AI designed new antibiotics against resistant gonorrhea and MRSA. And in UC Davis, an AI brain-computer interface restored natural speech to a man with ALS.

Science. The "co-scientists" of AI already shorten biological analyses from months to days, propose drug candidates and solve mathematical problems left open for decades. Always with scientists validating: the people at DeepMind acknowledges that human validation is now the bottleneck, rather than idea generation.

Education. In Harvard, physics students learned more than twice as much in less time with an AI tutor designed by their teacher. In Nigeria, the World Bank measured six weeks of teacher-guided AI tutoring: the equivalent of almost 1.5 years of typical learning, better than around 80% of educational interventions evaluated through controlled trials in developing countries. With teachers in the loop, not replacing them. And from Chile, Latam-GPT: Latin America's first major open model, involving more than 60 institutions from 15 countries, designed for education and public administration.

Justice. The Supreme Court of India translates its rulings into 18 languages with AI, and states in the same document that AI makes no decisions. In Colombia, PretorIA helps the Constitutional Court process more than 2,700 constitutional protection claims daily and prioritize urgent ones. In Argentina, Prometea reduced rulings from 190 days to one hour, with an official reviewing and signing. In Brazil, AI cut case-file review time by more than 40%. In San Juan, the Court of Justice transcribes hearings and summarizes cases "under human supervision." It is the phrase repeated in every case that works.

State and democracy. In Bowling Green, Kentucky, over 8,000 residents built a shared vision for their city with AI, and today that vision guides municipal planning. In Catalunya, 210 public procedures were simplified with AI this year. Fewer lines, less paperwork, more time.

Climate. Google Flood Hub provides flood forecasts up to seven days ahead in more than 150 countries and territories, including Chile. ForestCast predicts where deforestation will occur so action can come first, rather than counting losses afterward.

Economy and SMEs. In Chile, the 86% of entrepreneurs already uses generative AI regularly; 39% save more than four hours a week, and one in five recovers an entire workday. A workday. To sell, to think, for their family. The Dallas Fed found that AI replaces codified knowledge and complements tacit knowledge—experience: wages grew 8.5% in the most exposed industries, compared with 7.5% nationally. The IMF estimates that new skills pay between 3% and 3.4% more. And the most cited study on productivity, in customer service centers, showed that an AI assistant increased performance by around 14% on average and more than 30% among less experienced staff. AI raises the level for people just starting out.

Culture. AI recovered John Lennon's voice from a 1970s demo for "Now and Then", the first AI-assisted song nominated for a Grammy, and it won. It did not invent Lennon: it recovered him. It is reading carbonized papyri by Vesuvius 2,000 years ago without opening them. It found hundreds of new geoglyphs in Nazca that had been in plain sight for centuries. And in Chile there are AI translation projects for Mapudungun and Rapa Nui.

AI did none of this alone. It worked with a person in charge: the radiologist, teacher, judge, scientist, entrepreneur or music producer. In each case, someone reviews, decides and signs off. That is called doing its part. And it proves that "non-replacement AI" is not a slogan: it is how the things that work actually work.

To those who raced and those selling fear

To those who raced without brakes: apply your own brakes. Put in place the audits, pauses and limits you failed to establish before. Welcome to human supervision; we never gave it up.

To those selling fear: here are the data. The apocalypse is short on evidence. Selling the end of the world is easier than building. We prefer to build.

In WiWO we chose our path on day one, and we say it plainly.

We work toward a AI that does not replace people: amplifying people instead of replacing them. Every agent we build exists to help a team do more and better, not to make the team smaller.

We work toward a Safe AI: with defined boundaries, data protection, and levels of autonomy agreed with each customer before anything is activated.

And we work toward a AI that waits for human approvals: our agents propose, but no decision that matters is executed without a person's sign-off. A budget, a campaign, a message to a customer: AI prepares, a person approves.

AI proposes. A human signs off.

That is how innovation lasts, rather than a brake on innovation.

Optimism does not mean denying risks. The Dallas Fed itself warns that pressure falls on entry-level positions, and the IMF speaks of polarization without training.

Our optimism demands responsibility: doing our part also means training teams, redesigning how people learn a craft when AI handles repetitive work, and measuring the impact on people as seriously as we measure the impact on results.

Since 2008, we have been told our industry was ending three times: with social media, with mobile phones, and with programmatic advertising. It was never the end. It was change. And change worked out well for those who did their part

The future can turn out well, too. It is a decision, not a promise. And it is made every day, in every company, every team, every approval.

We have already chosen. Let's do our part.

What is yours?