The real barrier to AI in companies isn't the tool, it's the process nobody redesigned
- Reading
- 6 min
- Tables
- 1
While the market chases the next model, most Brazilian companies still haven't finished digesting the previous generation of technology. For the founder, that gap costs money every day, not just opportunity.
The hype runs, the process moves slowly
Every month a new agent appears, a new model, a feature that promises to eliminate an entire step of work. For anyone following this market closely, there's a permanent feeling of being one step behind. But a column published by entrepreneur Gilberto Barros, founder and CEO of Aion Holding, puts a finger on a wound every founder should look at before signing the next AI license: while the market debates the next generation of the technology, much of the corporate world still hasn't properly finished implementing the previous one.
Barros gives an example from his own operation, in the healthcare sector: some clinics are discussing artificial intelligence while others still run part of their management in spreadsheets. Systems that integrate scheduling, finance, medical records, and billing have existed for years and remain underused. This isn't exclusive to healthcare. It's the picture of any vertical where buying technology became easier than changing how people work.
What the Cetic.br numbers hide
The TIC Empresas survey, from Cetic.br (Brazil's internet research center), cited in the column, sets the baseline: in 2025, 17% of Brazilian companies with ten or more employees used some kind of artificial intelligence. Among small businesses, 15%. Among large ones, 50%. The breakdown by size is the data point that matters to the founder, because it doesn't just measure access to budget: it measures the capacity to absorb process change without stalling the operation midway.
A large company has a product squad, a data team, and a training budget to push a tool all the way to real use. A startup or an SME doesn't have that slack. For this second group, the bottleneck is rarely lack of access to technology (today any founder can open an API account in minutes). The bottleneck is the time and attention left over after the core product has already consumed the team's entire day.
The cost that doesn't show up on the SaaS bill
This is where the math comes in that every board should demand before approving one more AI line item in the budget. The subscription price is visible. The invisible cost is different: every new tool that requires a separate login, a separate flow, and a new habit competes for the team's same finite energy. If the tool doesn't replace a task, it adds to the pile.
In short: the return on an AI investment isn't measured by the sophistication of the model contracted, but by the distance between buying the license and the team actually changing how it works. That distance is where the money evaporates. A cutting-edge generative AI platform, used only to summarize emails or draft text, delivers marginal value next to the licensing cost and the training time invested. The same budget, applied to eliminate a manual and repetitive step in a process that already existed, changes the quarterly result.
The Aion Holding case: the right interface is the one the user already uses
The answer Barros found in his own operation is simple to describe and hard to execute: instead of bringing the patient to the system, bring the system to where the patient already is, WhatsApp. An MRI scan sent by message today depends on someone downloading the file, transferring it, and manually entering it into the medical record. The proposal is for a plain-language instruction to complete that path without repeated human intermediation.
The point for the founder isn't WhatsApp itself. It's the criterion behind the choice: technology wins when it reduces a step, not when it adds a channel. Any CTO proposing a new AI layer should answer, first and foremost, whether the end user (customer, patient, salesperson, operator) will need to learn something new or whether the gain arrives without extra friction.
The challenge of artificial intelligence isn't keeping up with every new release, but making technology solve real problems in companies' daily operations.
Gilberto Barros, founder and CEO of Aion Holding
The counterargument: isn't this just the natural adoption curve?
It's worth laying out the strongest argument against this reading. Every disruptive technology goes through a slow adoption curve before becoming the standard: electricity, the internet, cloud computing. In this view, 17% adoption in 2025 isn't failure, it's a normal early stage, and the founder's role would be simply to have patience and keep testing until the curve accelerates on its own.
The problem with this reading is that it treats time as an infinite resource, and it isn't. While one company waits for the curve to advance on its own, a competitor redesigns a specific process, measures the gain in hours or revenue, and pulls ahead before the rest of the market even realizes there was a race. Waiting for the technology to mature on its own is a valid strategy for those who don't compete on margin or service speed. For most founders reading this column, it isn't.
What to ask before the next AI budget
The question Barros raises at the close of his own column, about making a technology work on a Monday morning, is the test I would apply to any AI investment proposal that landed on my approval desk.
| Question | What it reveals |
|---|---|
| Does the tool reduce a step that exists today, or create a new one? | If it creates one, ROI depends on voluntary adoption, which is the most expensive thing to guarantee |
| Is there a before-and-after metric (time, cost, revenue) defined before the purchase? | Without it, the project's evaluation becomes opinion, not data |
| Did the team that will operate the tool take part in the choice? | A tool imposed top-down has a higher abandonment rate, even at large companies |
| Does the solution reach where the user already is, or does it require a channel change? | A new channel is new friction, even when the underlying technology is superior |
None of these questions require understanding the model behind the tool. They require understanding the process it promises to change. That's the work that falls to whoever approves the budget, and it's the work that the euphoria around every AI launch has been leaving aside.
What remains open
Barros's own experience with the WhatsApp project is still being tested, and he's honest about that: it isn't a ready-made solution, it's a hypothesis under validation. That's relevant information for any founder tempted to replicate the idea without adaptation: what works as an access channel for a clinic may not work for an e-commerce business, a bank, or an industrial operation, because the manual process being replaced is different in each case.
What doesn't change from sector to sector is the criterion. The right question for the CTO is never 'what's the most advanced model available.' It's 'what step of our process does this model eliminate, and how will we measure that in eighty days.'
Translated from the Brazilian Portuguese original · Read the original
OpenAI swaps its profit cap for a public benefit corporation structure
Since October 2025, OpenAI has gone from a capped-profit company controlled by a foundation to a public benefit corporation with conventional shares. The design holds direct lessons for anyone raising AI-native rounds in Brazil.
