Generative AI: What's Coming Next | SUMāTO

For years, artificial intelligence was, for most companies, a background promise: something that sorted emails, recommended products or detected fraud, but rarely surfaced in the boardroom conversation. That is about to change. A new generation of models capable of creating text, images and code is moving out of the research labs and, at remarkable speed, into everyday use. Anyone paying close attention to what systems like GPT-3 or DALL·E 2 already do will realize this is not a technical curiosity but the start of a phase shift.

In short: Generative AI refers to models that produce new content from natural-language instructions. The technology already exists and is maturing fast; what's coming is its move into the mainstream. Companies that start now to organize their data, define usage rules and test concrete use cases will arrive far better prepared than those who wait for everyone else to move first.

What generative AI actually is

The AI we have known until now is mostly predictive or classificatory: given a data point, it decides which category it belongs to or what value to estimate. Generative AI does something different: it produces original content that was not present in its input data. You ask for something in natural language and it returns a drafted text, an image, a piece of code or a summary.

Behind this capability are two families of models that have matured in parallel:

• Large language models: trained on vast amounts of text, they learn to continue and complete instructions. They serve to draft, summarize, translate, answer questions and generate code.
• Diffusion models: they learn to transform noise into coherent images guided by a description. They are the foundation of tools that generate illustrations, sketches and visual concepts from a single phrase.

What is truly new is not only the quality of the output but the interface: for the first time, the way you ask an advanced machine for something is the same language you speak. That removes the technical barrier that kept AI in the hands of specialists.

Why it will move from the lab to mass adoption

Several forces push in the same direction and explain why adoption will stop being experimental:

• The barrier to entry has fallen: you no longer need to know how to program to get value. Anyone who can describe what they need can use these tools.
• Distribution is getting simpler: the models are increasingly offered as accessible services through simple interfaces, which makes it possible to embed them into existing products without building everything from scratch.
• The return is immediate and visible: in tasks such as drafting first versions, synthesizing documents or generating creative variants, the time saved is felt from the very first use.

When a technology combines a low barrier to entry, simple distribution and an obvious benefit, its diffusion stops being linear. That is why it pays to prepare before the wave arrives, not while it breaks.

Concrete opportunities for companies

The value is not in "having generative AI" but in applying it to processes where time is currently spent producing or processing content. Some fertile ground:

• Service and support: generating responses, case summaries and drafts that a human agent reviews and approves.
• Marketing and communications: first versions of copy, variants for testing and visual concepts that accelerate creative work without replacing judgment.
• Software development: assistance in writing, documenting and reviewing code, freeing technical teams to focus on higher-value problems.
• Internal knowledge: synthesizing manuals, contracts or lengthy reports so people can find answers in seconds.

The common pattern is the same: AI produces a draft or a proposal, and the person contributes judgment, context and final accountability. That division of labor is, for now, where the real return lies.

Risks worth facing head-on

The very ease that makes this technology attractive brings risks that leadership must understand before scaling its use:

• Errors that look like certainty: the models can generate incorrect statements written with complete confidence. Without human review, an error can propagate.
• Data confidentiality: feeding sensitive information into external tools without clear rules exposes the organization.
• Intellectual property and authorship: the origin of generated content raises legal and reputational questions that are still evolving.
• Inherited bias: the models reflect the data they were trained on, and that can carry bias into the results.

None of these risks is a reason to stay out. They are reasons to enter with governance, defining from the outset what can be used, with which data and under what supervision.

How to get ready now

The advantage will not belong to whoever has the biggest model, but to whoever has their house in order when the technology becomes accessible to everyone. Three areas of work:

• Data: your company's differentiating value lies in its own information. Organizing it, cleaning it and making it accessible is the investment that pays off most, because without good data any model performs poorly.
• Governance: defining clear usage policies, information classification and human oversight. This gives you the confidence to experiment without exposure.
• Use cases: choosing one or two bounded, measurable, low-risk processes to learn in practice, rather than waiting for the perfect strategy.

An organization that adopts an AI-first mindset does not automate to follow a trend: it redesigns how it works, starting from the question of which tasks these models can support and which will always demand human judgment. That shift in approach is the real preparation.

From curiosity to capability

It is worth distinguishing between trying a tool and building capability. Everyone will do the former within a matter of months. The latter—embedding generative AI into real processes, with your own data and clear rules—is what separates the companies that gain a sustained advantage from those that merely accumulate scattered experiments.

At SUMāTO we see this stage as joint work across consulting and technology: organizing the data, designing the governance and taking use cases into production. If you'd like to explore the full picture, you can review our perspective on artificial intelligence applied to business in LATAM.

Frequently asked questions

Will generative AI replace people?

On the visible horizon, the most useful model is collaboration: the machine generates drafts and proposals, and the person contributes judgment, context and final accountability. What changes is the content of many jobs, not their immediate disappearance.

Do I need a team of data scientists to get started?

Not for the first steps. The technical barrier has fallen precisely because these models are operated with natural language. What you do need is to organize your data and define usage rules, two tasks that depend more on decision and method than on a large specialized team.

What about the confidentiality of my company's information?

It's a legitimate and central concern. The recommendation is not to feed sensitive data into external tools without a clear policy, and to evaluate options that let you control where the data resides and how it is processed.

Is it too soon to invest?

It's too soon to bet everything, but it's the ideal moment to build foundations: organized data, defined governance and a couple of use cases underway. Those foundations pay off now and become decisive when the technology goes mainstream.

The first step

Generative AI does not demand a big initial bet, but rather a first, well-made decision: identify a process where content is the bottleneck, organize the data that feeds it and test with supervision. From there, capability is built on evidence.

If you want to prepare your organization before this wave goes mainstream, let's talk. At SUMāTO we help you move from curiosity to real capability. Write to us and let's take the first step together.