"AI replaces programming for decision-making"... AWS accelerates the shift toward a model-centric design

Embracing the era of AI, traditional programming methods are facing limitations. Especially in the design and operation of AI agents, existing fixed workflows are no longer the optimal choice in terms of flexibility or efficiency. To address this, Amazon Web Services (AWS) is accelerating its shift toward model-driven design that actively leverages the reasoning capabilities of artificial intelligence.

Claire Liguori, Chief Engineer at AWS, emphasized this strategic paradigm shift at the “AWS re:Invent 2025” conference in Las Vegas. She explained that because AI agents operate in unpredictable and dynamic environments, the traditional method of developers pre-coding all processes can actually be counterproductive. Therefore, she advocates for a model-driven approach, allowing AI models to autonomously construct logic and lead the execution process.

Liguori stated, “In the past, we created standardized workflows using traditional coding practices to solve complex tasks, but ultimately this is a fragile method. Returning to model-driven design and letting AI take the lead makes it possible to accomplish innovative tasks.” She added, “This mindset is especially intuitive for the younger generation of developers.”

The biggest technical obstacle in agent development is the so-called “structural glue code”—that is, orchestration code and defensive logic. Liguori pointed out that these fixed-cost components cause significant inefficiency, accounting for 90% of total development time and cost. She specifically emphasized that for high-performance models like cutting-edge AI that can reason autonomously and select tools, excessive engineering around these models can actually undermine performance.

Liguori explained, “Today, managing agents is no longer that complicated. You can simply swap out the model without modifying the entire software system, and performance will improve significantly.” In line with this trend, AWS has introduced TypeScript into its open-source framework “Strands” to lower the entry barrier. This move aims to enable even non-AI experts to easily implement AI agents.

She said, “We want anyone to be able to create agents with just a few lines of code. In fact, some product managers have come to me and said, ‘After trying to write it myself, it’s amazing.’”

Experts predict that, in terms of reducing system costs, simplifying code, and rapidly applying AI innovations, the model-driven design approach will become the core infrastructure strategy of the AI era. Liguori’s remarks clearly indicate that this shift is not merely a marketing slogan, but a genuine trend in technological evolution.