Why AI Will Always Need Humans, According to SAS's 50-Year AI Chief

SAS at 50: When AI Should Decide and When Humans Must

Data analytics and AI company SAS is marking 50 years in business, a milestone few software firms reach. Udo Sglavo, vice president of applied AI and modeling R&D and a veteran of nearly half that history, says the company's longevity comes from a simple discipline: focus on the business problem first and ask how technology can solve it, not the other way around.

At the SAS Innovate conference in Washington, D.C., Sglavo made the case for what he calls 'boring' AI. Routine medical scans where there is no tumor, or a bank customer requesting a cashier's check, are repeatable, low-risk cases that an AI system can handle. The expensive human expert should be reserved for edge cases, suspicious findings and emotional judgment. That, he argues, is where the fastest return on investment lives.

The same newsletter put that human-in-the-loop argument next to bigger AI narratives. Meta CEO Mark Zuckerberg published a 6,500-word essay calling for superintelligence to be distributed through open-source releases and for closer government-developer collaboration, days after Meta disclosed an AI hacking incident. Google said its Gemini assistant crossed 1 billion monthly active users and became the company's fastest-growing product.

In industrial AI, Siemens, Schneider Electric, Honeywell, ABB, Emerson and Rockwell Automation are expanding through acquisitions and partnerships. Agility Robotics said its Digit V5 robot will begin deployment in December, designed to work near people in bulk material handling without safety fencing. Yet a separate survey found that 51% of software providers say fewer than one in four customers actually use their applications.

Zuckerberg, Industrial AI Deals, and the Real Adoption Gap

SAS's 'Boring AI' Is a Discipline, Not a Downgrade

Sglavo's examples point to a specific ROI logic: automate the repeatable norm, keep humans on the exception. He explicitly says humans remain decision-makers for ambiguous cases, and that AI systems are 'ice cold' decision engines built on historic data. His skepticism about running an entire enterprise with agents is also social: businesses depend on people as employees and as consumers.

Zuckerberg's Open-Source Optimism Meets Security Reality

Zuckerberg's essay is a strategic argument as much as a philosophical one. By framing open-source distribution and distillation as tools to empower individuals and maintain U.S. leadership, Meta is positioning itself ahead of likely regulation and trying to shift the debate from AI harm to access. But the timing matters: the company disclosed an AI hacking incident the same week, which makes security and model provenance harder for enterprise buyers to ignore.

Industrial AI Consolidation Is Quietly Raising the Stakes

The deals involving Siemens, Schneider Electric, Honeywell, ABB, Emerson and Rockwell Automation extend incumbents' control over industrial automation, risk intelligence and humanoid robotics. For startups, the deals offer scale. For buyers, the consolidation means a smaller set of platforms will shape how AI enters factories and warehouses. Agility Robotics' December deployment of Digit V5 is a concrete test: if a humanoid robot can work safely alongside people in a warehouse, the use-case door opens far beyond bulk material handling.

The 51% Gap Shows the Real Bottleneck Is Adoption

The survey finding that most software providers see fewer than one in four customers using their applications challenges the idea that model capability is the main constraint. Sglavo's prescription is direct: connect business and IT, start with the business question, get the data architecture right, and build decision architecture that can explain why an AI system made a call. If a regulator asks why 10% of credit applicants were declined, 'we don't know' is not an acceptable answer.

What Technology Leaders Can Act On This Week

For technology leaders, this newsletter contains several specific signals rather than a single directive:

• Apply the 'boring' test to AI pilots. Sglavo's medical scan and cashier's check examples suggest low-risk, high-volume processes are the fastest path to measurable ROI. Keep a named human owner for edge cases and emotional decisions.
• Review open-model deployments against Meta's own disclosure. Zuckerberg's call for open-source superintelligence and government collaboration does not eliminate deployment risk. Given Meta's disclosed AI hacking incident and the newsletter's warning about vulnerabilities in AI-generated code, document model provenance and add security checks before rollout.
• Watch Agility's December Digit V5 deployment as adoption evidence. A cooperatively safe humanoid robot working in bulk material handling without fencing would be a meaningful proof point for warehouse automation, but safety in a difficult workspace is the claim to test.
• Treat the 51% usage statistic as an internal audit prompt. Before buying more AI tools, follow Sglavo's sequence: define the business question, align business and IT, fix data architecture, and ensure any automated decision can be explained to a regulator.