Amazon’s Peter DeSantis at VivaTech: AI’s biggest breakthroughs still haven’t arrived

Amazon AI executive Peter DeSantis used VivaTech 2026 in Paris to argue that the era of truly transformative artificial intelligence is still ahead, not here. In remarks published by Amazon, he framed today’s AI progress as the opening phase of a development cycle, saying the most important breakthroughs have not yet happened and that the field is still close to the “starting line.”

DeSantis’ core claim was that AI needs substantially better performance before it can deliver the kind of real-world impact people typically associate with “agent-like” or human-level intelligence. He pointed to the magnitude of improvement required, describing it as needing “a couple more orders of magnitude” before AI becomes truly transformative. The comment sets a high bar for future capability, implying that incremental gains from current systems will not be enough by themselves.

The executive also suggested that the next wave of progress will come from architecture changes rather than only training data or scaling. He said “new model architectures” are expected to emerge beyond today’s widely used transformer approach. Transformers are the neural-network architecture behind many modern language models, built to process sequences of data efficiently, especially text. DeSantis implied that surpassing transformer limitations will be necessary for the next step in capability and responsiveness.

One concrete performance target he highlighted was speed and conversational responsiveness. DeSantis said future architectures could let AI respond “as fast as humans talk.” In practical terms, the goal is to reduce the latency and turnaround time that often make current assistants feel slow or segmented, particularly in back-and-forth conversations where timing affects usefulness.

Across the remarks, DeSantis emphasized that even as systems become more capable, people will remain central to the most complex innovations. He said humans will be at the center of AI’s most complicated innovation, positioning AI as an accelerator and enabler rather than a replacement for human judgment, creativity, or problem framing. That framing is especially relevant for enterprise adoption, where decisions depend on domain context and accountability.

DeSantis’ comments also arrive as AI competition intensifies across the industry, with cloud providers and model builders each pushing deeper into large-scale model deployment, tooling, and infrastructure. Amazon, via AWS, has been a major player in offering AI and machine learning services to customers, giving it an incentive to see clear paths to better performance and more reliable AI experiences. While the remarks do not announce a specific new model or product, they align with a broader message common in AI circles: scaling helps, but architectural and system-level changes determine the next leap.

Amazon’s position can be read as a caution against premature expectations. By tying transformation to orders-of-magnitude improvement and a shift beyond today’s dominant architecture, DeSantis suggests that organizations evaluating AI should separate near-term experimentation from longer-term operational transformation. His view implies that many current deployments may remain limited until the next performance thresholds are reached.

At the same time, Amazon did not lay out measurable milestones in the published remarks. There was no specification of what “orders of magnitude” means in terms of accuracy, cost, throughput, latency, or reliability, and there were no timelines for when new architectures might be available at scale. The remarks also did not identify particular techniques, model families, or AWS-specific services that would deliver the anticipated leap.

The lack of technical detail is notable given how much the statement implies. If DeSantis is right that the next breakthrough requires a fundamentally different approach, then near-term gains from current tooling might look smaller than some market narratives suggest. For editorial review, the key uncertainty is what combination of architectural change, training methods, inference optimization, and product integration will actually produce the speed and responsiveness he described, and how quickly that can reach real users. Investors, partners, and customers may ultimately focus less on big claims and more on concrete system benchmarks and deployed experiences.

Looking ahead, the most important thing to watch is whether Amazon and the broader AI industry converge on new architectures that deliver the responsiveness target DeSantis described, and whether those systems can be implemented reliably in production settings. Another watch item is whether the market recalibrates its expectations for “AI transformation” around measurable capability thresholds, not just model releases or headline performance scores. If future systems begin to approach human-like conversational speed and show clear reliability improvements, it would support the premise that the major breakthroughs are still ahead rather than already completed.