In a recent LinkedIn post, Luís Rodrigues explores the often-overlooked challenges hindering the successful implementation of AI within organizations, arguing that the focus on AI models themselves is misplaced. Rodrigues contends that the true obstacles lie not in the technology’s capabilities, but in the foundational elements of data, process, training, and change management.
Rodrigues uses a compelling analogy to illustrate his point:
“Four executives. One worker. A bridge that ends in air. The models stopped being the bottleneck. The pillars are the bottleneck. Data. Process. Training. Change.”
He elaborates that these critical pillars require significant time and resources, often a year and a budget that many companies are hesitant to commit to. This leads to a common scenario where businesses attempt to implement AI by purchasing the ‘span’ of the technology while neglecting the essential ‘foundations’.
The Missing Foundations of AI Implementation
Rodrigues, drawing from his experience deploying AI agents in banks, highlights the extensive effort involved beyond just the model development. He recounts instances where the actual implementation was dramatically slowed by:
- Data residing in disparate, conflicting systems.
- Core business processes that were undocumented and reliant on tribal knowledge.
- A lack of clear ownership and accountability, leading to stalled rollouts even after successful pilots.
As Luís Rodrigues notes, the model work might take weeks, but the surrounding elements consume far more time and effort.
The Crucial Pillar of Measurement
The LinkedIn post also identifies a fifth, often-ignored pillar: Measurement. Rodrigues criticizes the current trend of counting metrics like tokens used and seats deployed, which he argues do not correlate with actual business movement or success.
“They count tokens used and seats deployed. All of it goes up, and the business does not move. Because the work does not live inside one person. It lives in the gaps: the handoff to legal, the approval that waits three days for someone in another time zone, or the queue nobody has looked at since 2019.”
According to Luís Rodrigues, true progress is measured not by individual worker output, but by the efficiency of the entire workflow. He advocates for measuring key performance indicators such as cycle time, throughput, error rate, and cost per case.
Rethinking Workflow Redesign
Rodrigues suggests that many existing approval steps might be relics of a slower, human-centric process that are no longer necessary with AI automation. He posits that the real work and the key to successful AI adoption lies in critically examining and redesigning the entire workflow, rather than solely focusing on selecting the most advanced AI model.
“The real work is redesigning the workflow, not picking the model.”
In his view, organizations that bypass the foundational work of data integration, process definition, comprehensive training, and robust change management—while also neglecting workflow measurement—will find themselves standing on the edge of a cliff, wondering why their AI initiatives haven’t yielded tangible business results.
📝 About This Content
This article is based on insights shared by Luís Rodrigues on LinkedIn.
📅 Originally posted on September 1, 2026 | View original post on LinkedIn →