A comic strip went viral a few weeks ago and caught my attention. A little figure asks, "Who are we?" and several others reply, "CEOs." "And what do we want?" They respond: "AI!" "Artificial intelligence to do what?" And they reply, "We don't know!" "But when do we want it?" And they respond, "Now!"
The joke perfectly reflects reality, not just what we see in our daily lives, but also what emerges from the report "The GenAI Divide: State of AI in Business 2025," published by the MIT initiative Networked AI Agents in Decentralized Architecture (NANDA).
The study reveals that while generative AI holds promise for agility, problem-solving, and even profitability in the corporate world, most initiatives have yet to reach significant levels of success.
The report states that only about 5% of the AI pilot programs achieved significant revenue acceleration. Most initiatives stalled, resulting in little or no measurable impact on profits or losses.
In an interview with Fortune, the lead author of the report and MIT's NANDA project collaborator, Aditya Challapally, explained which pilot programs of large companies and younger startups stood out using generative AI in recent years. "They choose a problematic point, execute well, and make smart partnerships with companies that use their tools," he added.
For the 95% companies included in the report, simply implementing a generative AI solution wasn't enough. The core issue wasn't the quality of the models and tools, but rather the "learning gap."
The Fortune article states that, while executives blame regulation or model performance, the MIT investigation points to flaws in business integration.
In other words, the AI solution does exist, but problems were found in the workflow of the other parts of this process: the humans.
More general-purpose tools, like ChatGPT, are excellent for individuals due to their flexibility, but they're not a miracle cure for business use because they don't learn or adapt to workflows, Challapally explained.
I've talked about this in several presentations and conversations. AI tools are a great support, but not a shortcut. AI is excellent for accelerating tests, refining ideas, verifying data, or even performing complex tasks, like mastering a dedicated software or application.
As the report noted, companies that successfully selected their AI agent chose a problematic issue or point of friction and managed to resolve, or at least accelerate, that process, naturally leading to increased productivity and profitability.
A good question to ask, before joining the chorus of "We want AI in everything now," is: what tools and solutions are available that can help meet the company's needs?
There may not be ready-made products and applications for the biggest questions yet. However, if you can speed up a process or provide a stronger foundation for your employees to make better decisions, that will be the best AI support for your company right now.
The final decision is always human, including in determining how far artificial intelligence will assist. And like all technology, we are in a period of evolution and refinement. Therefore, your decision may also change within six months.
Henrique Calandra He is the founder of WallJobs, a Brazilian tech company offering automated solutions for internship contracts, author of the book "Generative Artificial Intelligence for Beginners," columnist for ABStartups, and speaker at major ecosystems like InovaBRA and Distrito.
