> ## Content Index > Fetch the complete content index at: https://futureinuse.ai/llms.txt > Use this file to discover other available public pages before exploring further. # From adoption to operations, six short notes on making AI work - URL: https://futureinuse.ai/da-adocao-a-operacao-notas-curtas/ - Published: 2026-09-18T20:50:00.000Z - Updated: 2026-10-22T00:49:20.000Z - Description: From January to September 2026, Rafael Cordeiro wrote short notes about the same bottleneck. AI impresses in the demo and stalls when it meets the real company. This collection brings together six of them. - Author: Rafael Cordeiro - Tags: Rafael Cordeiro, colunista, #historico-linkedin What separates an AI pilot from a system in production is rarely the model. In the notes below, the author writes about workflows before agents, about what happens after the insight, about the patterns that stall traditional companies, about pilots that start too big, about the engineers who carry AI into operations and about the infrastructure agents still lack. Each note carries its original date and a link to the post. The data is valid as of the date of each note. ## Workflows before agents *January 21, 2026* Everyone wants to jump straight from chatbot to agentic AI. But 2026 is making something clear. The real value lies in workflows, not in fully autonomous agents. In McKinsey's November 2025 survey, in no business function did more than 10% of companies say they were scaling AI agents. Calling everything an agent sets the wrong expectations. My practical recommendation for 2026 has four steps. Take a prompt that already works well, turn it into a repeatable process, define a recurring deliverable and let AI run the flow, with a human in the decision loop. Those who build workflows today will adopt real agents much faster tomorrow. Source, [McKinsey, The state of AI in 2025](https://www.mckinsey.com/capabilities/quantumblack/our-insights/the-state-of-ai-2025?ref=futureinuse.ai), November 5, 2025 · [Original post](https://www.linkedin.com/feed/update/urn:li:activity:7419672799956533248/?ref=futureinuse.ai) ## The bottleneck is what happens after the insight *January 26, 2026* We are obsessed with intelligence. More data, more models, more dashboards. But the bottleneck was never intelligence. The bottleneck is what happens after the insight, when the decision has to cross teams, systems, approvals and conflicting priorities. The future of companies is not about thinking better. It is about turning decisions into execution and continuous learning. [Original post](https://www.linkedin.com/feed/update/urn:li:activity:7421602827560734720/?ref=futureinuse.ai) ## Where AI stalls in traditional companies *January 29, 2026* The biggest mistake traditional companies make when adopting AI is not technological. It is operational. There is no lack of interest in AI. What is missing is a connection to the business. Three patterns keep recurring. AI as a curiosity assistant, which answers questions but cannot access the ERP, CRM or financial data. Pilots that work in presentations and never enter the real flow. And critical data kept out of AI's reach, out of a justified fear around security and compliance. Traditional companies do not need to become big tech. They need to answer where decisions really happen, and why AI is not there yet. When it accesses real data, respects governance and operates within existing flows, it stops being an experiment and becomes decision infrastructure. [Original post](https://www.linkedin.com/feed/update/urn:li:activity:7422711383328219136/?ref=futureinuse.ai) ## Pilots die from starting too big *February 5, 2026* Why do so many AI projects never get past the pilot stage? It is not a lack of talent, investment or technology. What I see most are teams trying to build the perfect system before the first real impact. In the projects that reach production, the pattern tends to be different. A specific problem, poorly solved today. An imperfect but functional automation. A fast loop between machine and human. Learning from real use. Human-in-the-loop does not slow AI down. It speeds it up, because that is how you understand exceptions, discover limits and learn where to automate for good. The most sophisticated systems today were born simple and evolved because they were being used. [Original post](https://www.linkedin.com/feed/update/urn:li:activity:7425138807462785025/?ref=futureinuse.ai) ## The next scarcity is people who put AI into production *September 10, 2026* The next scarcity in AI may not be compute. It is people able to put it into production. Google Cloud and Accenture announced a business group dedicated to Gemini Enterprise, with a force of 1,000 forward deployed engineers to speed up implementations. Anyone who has implemented technology in large companies knows the demo is almost never the problem. The problem starts when it meets fragmented data, legacy systems, permissions, integrations and poorly documented workflows. The FDE is the engineer who works close to technical and business teams to carry the technology across the last mile between working in the demo and working in operations. For me, this move signals the shift from AI experimentation to AI operationalization. Source, [Accenture and Google Cloud announcement](https://newsroom.accenture.com/news/2026/accenture-and-google-cloud-deepen-partnership-with-formation-of-new-accenture-gemini-enterprise-business-group?ref=futureinuse.ai), September 8, 2026 · [Original post](https://www.linkedin.com/feed/update/urn:li:activity:7503772511399243776/?ref=futureinuse.ai) ## Agents arrived before the infrastructure *September 18, 2026* AI agents have arrived. The infrastructure to put them in the field, apparently, has not. When they enter real processes, they meet fragmented data, legacy systems, undefined permissions, processes no one documented and a meeting to decide when the next meeting will be. To deliver value, an agent needs reliable context and data, controlled access to systems, clear limits on what it can do, validation and traceability, and a business result to answer for. Without that, even the most advanced agent ends up stuck in a POC. The technology impresses. Operations decide whether it will work. [Original post](https://www.linkedin.com/feed/update/urn:li:activity:7506816889277276160/?ref=futureinuse.ai) **Further reading** - [AI Gateway, the control layer for scaling AI in the enterprise](https://futureinuse.ai/ai-gateway-camada-de-controle-ia-empresarial/) - [The enterprise knowledge layer, the context AI agents need to operate](https://futureinuse.ai/camada-de-conhecimento-ia-empresarial/) - From adoption to operations with AI --- *Notes originally published on LinkedIn between January 21, 2026 and September 18, 2026, with each post's link next to its note. This collection has been edited and organized with context and sources. The views are those of the author, a co-founder of OperOS AI.*