In 2025, I stopped asking “which AI model is the best?” and moved on to “how do we put AI to work in the business with clear metrics and without putting data at risk?” That’s the conversation I’m having today with manufacturers, wholesalers, channels, and telcos in Latin America.
From demo to agent that does the work
The novelty is no longer a chatbot that responds: it is agents that understand a goal, plan steps, and execute them in your systems. End-to-end customer service, IT agents who open and close cases, or operational agents who adjust network parameters. This trend marks the leap from testing AI to operating it in production.
AI on the device: PCs with NPUs and local decisions
The other wave is bringing AI to the endpoint (AI PCs). What does the business gain? Less latency, more privacy, and more predictable costs. Analysts project more than 100 million AI PCs in 2025 and a strong renewal cycle; the total PC market is around 274 million units this year. For us, this means moving some of the AI workloads to where the work happens, not just to the cloud. Security tip: Some endpoint features (such as Recall in Windows) raise privacy concerns. Recommendation: Enable opt-in, define exclusion lists and policies per device from day one.
AI-ready networks on site: Wi-Fi 7 and 5G-Advanced
If intelligence runs at the edge, the network cannot be the bottleneck. Wi-Fi 7 incorporates 320 MHz channels, 4096-QAM, and Multi-Link Operation (MLO) to increase capacity and reduce latency on campuses and in offices. 5G-Advanced (3GPP Rel-18) enhances uplink, positioning, and automation for industry and campuses. In practice: smoother experiences, less waiting, and more predictable operations.
Data: from “uploading PDFs” to reasoning systems
To prevent agents from hallucinating, data must be governed. We move from basic RAG to pipelines with quality, deduplication, observability, and controlled memory. Simply put: without organized data, there is no reliable AI.
Security with a 10-year outlook. Two priorities: post-quantum cryptography (PQC)—plan the migration starting today—and more secure software—reduce classic memory failures. Security that protects without slowing adoption.
Regulation: preparing to sell to the world
The European AI Act already sets out obligations and deadlines. If you operate or sell to the EU, you will need transparency, risk management, and traceability. Better to design it well today than redo it tomorrow.
What are we running at Gen3sis?
TransformatIA 3.0 (diagnostics with numbers) We see where an agent really helps: people, steps, and tools. We deliver a short plan with expected savings and dates. Key question: Which process gives you the most time and money back in 90 days?
ImplementIA 3.0 (putting agents to work) We set up an initial agent to work in your day-to-day operations and measure its impact from the outset. Key question: What repetitive task would you eliminate tomorrow if an agent could do it well today?
ProtectIA 3.0 (security that doesn’t get in the way) We set clear rules and responses to incidents, and close weak points before they escalate. Key question: How do you increase defense without slowing down AI adoption?
OptimIA 3.0 (make it perform well and get people to use it) We make AI cheaper per use, faster, and easier to adopt for every role. Key question: If the cost goes down and the experience improves, what new use case becomes profitable?
AdminIA 3.0 (operating at the service level) We operate AI every day with clear dashboards and service goals; everything is measurable and auditable. Guiding question: What level of service would allow you to sleep soundly?
ResolutIA 3.0 (self-resolving support) We combine agents + knowledge + approved steps to resolve repetitive issues without delay: fewer tickets, shorter wait times, and more satisfied users. Key question: How much support could you automate without losing control in 90 days?
Our goal is for AI to cease being a pilot project and become a living, measurable, and auditable business system.
As CEO of Gen3sis, my position is simple: 2025 is the year to industrialize AI. We are not looking for the “newest” tool, but rather the right architecture: useful agents, reliable data, fast networks, and security that will last for the next decade. With TransformatIA, ImplementIA, ProtectIA, OptimIA, AdminIA, and ResolutIA, we are already implementing this with our customers and partners in the region.
SOURCES:
Gartner — Tendencias 2025 (Agentic AI, PQC, gobernanza). Gartner
McKinsey — State of AI 2025 y agentes en empresa. McKinsey & Company+2McKinsey & Company+2
Canalys/IDC — dinámica de AI PCs y renovación 2025. Canalysmy.idc.com
Apple — Apple Intelligence y Private Cloud Compute. Apple+1Apple Security Research
Microsoft — guía de privacidad de Recall; alerta operativa (UPenn). Microsoft Supportisc.upenn.edu
Wi-Fi 7 (Cisco, Meraki) y 5G-Advanced (3GPP Rel-18). CiscoCisco Meraki Documentation3GPP
RAG 2.0 — encuestas y buenas prácticas (arXiv/Google Research). arXiv+1research.google
PQC (NIST FIPS 203–205) y memory-safe (CISA/NSA). NISTwolfSSLU.S. Department of Defense
AI Act — cronograma y obligaciones para GPAI. Digital Strategy EU
Juan Sebastian Gomez,
CEO of Gen3sis