The days of AI as a differentiator are numbered: AWS’ Olivier Klein

In the last decade or so, cloud computing has moved from being an architectural layer to a foundation for modern businesses, and artificial intelligence is now following the same path at a far greater speed.

Nowhere is this shift more visible than in the Asia-Pacific region, where enterprises are accelerating experimentation and real-world deployment of AI—fundamentally changing the way software is built, operated and experienced.

The primary driver for this accelerated AI adoption is natural language, which is becoming a practical interface for systems and data, helping autonomous agents move beyond assistance into execution.

Olivier Klein, Chief Technologist at Amazon Web Services (AWS) for Asia Pacific Japan (APJ), says one can now use natural language pretty much as a programming language, and also as a language to just understand and query your data.

“Large language models now let people interact with systems and data using plain language, almost as if it were a programming interface. This suddenly allows business users, not just engineers, to prototype ideas and move much faster,” says Klein, drawing on his work across global markets.

This has brought in a level-playing field and is changing the dynamics around who can build, test and refine ideas inside organisations. AI can no longer be touted as a great differentiator, he notes.

“The days where you say AI is my differentiator are numbered, because everybody has access to the same latest and greatest models and tools. It is really about bringing in your data into those AI models to derive value,” says Klein, in an exclusive interview with YourStory, during the AWS re:Invent 2025 conference in Las Vegas.

Klein offers a grounded view of how cloud, data and AI are converging in practice, particularly in fast-moving regions such as APAC. He also speaks about security guardrails to chips and energy efficiency.

Edited excerpts…

YourStory (YS): You joined AWS more than a decade ago. How has your role, and cloud itself, evolved over that time, particularly with AI now in focus?

Olivier Klein (OK): When I started in 2013, we were a much smaller business at the time. AWS was still early in its growth. The first services were basic building blocks for storage and communication, and only later did large-scale computing become central. The focus then was explaining why cloud computing made sense at all. Over time, that conversation shifted. Cloud became the default, and the question changed from whether to use it to how to use it well.

The platform broadened from developers to enterprises across every industry, with services expanding to support scale, security and resilience.

That same pattern is now repeating with AI. Machine learning itself is not new, but what is changing is broad, production-grade adoption, particularly with generative and agentic AI. Three years ago, many organisations were experimenting. Today the focus is on running these systems reliably in production.

Agentic AI is starting to change how software is built and operated. Instead of spending time maintaining infrastructure or manually managing systems, teams can focus on building value for the business. Just as cloud removed the need to manage servers, AI is beginning to remove layers of operational toil, creating a new baseline for how platforms are designed and run.

YS: How are things showing up on the ground in the Asia-Pacific region?

OK: One of the most visible shifts is how natural language interfaces are changing who can use technology effectively. Large language models now let people interact with systems and data using plain language, almost as if it were a programming interface. That is opening access well beyond traditional engineering teams.

This means business users can move much faster. For example, Axis Bank in India uses AWS solutions so business teams can query data directly and gain insights. They rolled out over 70 dashboards across various business functions. By migrating to an AWS solution, Axis Bank realised cost savings of 80% on certain costs. Another example is Union Bank of the Philippines, where business teams were able to achieve 3 to 5 times faster data request turnaround times.

This is not limited to financial services. Across industries, teams are prototyping ideas without deep software engineering backgrounds, then formalising them later. That inclusivity matters in APAC, where large developer communities and fast-moving markets create strong pressure to iterate quickly.

Agentic AI builds on this by changing how software is developed and operated, allowing teams to focus less on maintenance and more on delivering business value at speed.

YS: When it comes to adoption, what should companies do differently to ensure AI reaches markets like India effectively?

OK: It is less about geography and more about culture. The companies that are most successful are those that allow people to use AI tools effectively and embed AI into everyday work. There is a tendency to think in terms of regional differences, but the real difference is whether an organisation encourages experimentation and gives teams access to the right tools and data.

AI itself is no longer the differentiator. The real value comes from bringing your own data into those models and using it well. That is why so much focus is on data foundations, fine-tuning, and cost-effective ways for storage and retrieval.

Services such as Amazon S3 Vector are designed to make this practical at scale, allowing customers to load their data into AI workflows efficiently and at significantly lower cost than traditional approaches. Tools like Amazon Quick then allow business users to query that data using natural language across multiple sources.

The combination of culture, accessible tooling, and strong data foundations is what turns AI from a capability into a competitive advantage, regardless of geography.

YS: How do you see AI changing developer workflows and the broader developer ecosystem?

OK: A big shift is moving from simply using AI to generate code towards using AI across the entire development lifecycle. That is why we introduced AI-Driven Development Lifecycle (AI-DLC), an openly accessible methodology to enable accelerated adoption of AI by businesses. AI-DLC places AI at the centre of the development process, with tools and frameworks such as Kiro, Amazon Q Developer, and Strands Agents, and reduces implementation timescale from months into days or hours.

With tools like Kiro, developers can still vibe-code to prototype ideas quickly, but there is now a strong emphasis on spec-driven development. That means starting with clear acceptance criteria and specifications for what the software should do, and then using AI to help build, validate and maintain it.

This approach recognises that most developer time is not spent writing code. A lot of time goes into documentation, fixing issues, testing, and maintaining systems. AI agents can take on much of that work. Security agents, DevOps agents, and autonomous coding agents help developers focus on building value for the business rather than managing repetitive tasks.

This changes what it means to be a developer. The role becomes less about producing lines of code and more about defining intent, validating outcomes, and iterating with the business. In regions with strong developer communities, such as India, this model is particularly powerful because it allows teams to scale faster while maintaining quality and consistency as they move from prototype to production.

YS: What advice would you give startups building with generative AI as they try to move fast without losing focus?

OK: Speed is a huge advantage for startups today, but it is not a differentiator on its own. Everyone has access to powerful models and tools now, often at low cost. What really matters is where the value comes from, and that is your data. The question to ask is whether you have built the right data foundations and whether you are using that data effectively when you build with AI.

AI agents can help startups iterate very quickly. You can build an MVP, test an idea, get feedback, and pivot faster than ever before. That is incredibly powerful. But if the AI workflows are not grounded in your own data, you will struggle to create something defensible. Models will converge, but your data, your customers, and your domain knowledge will not.

That is why fine-tuning, bringing your data into models, and using cost-effective ways to store and retrieve embeddings are so important. Startups that combine speed with strong data foundations are the ones that will create lasting value, rather than just impressive demos.

YS: Agentic AI is gaining a lot of attention. Where do you see it having the biggest impact, particularly on customer experiences?

OK: The early phase of generative AI was about assistants. You would prompt a model, get a response and then decide what to do next. That still requires a lot of human oversight. The shift now is towards embedding AI directly into workflows so agents can make autonomous or semi-autonomous decisions and execute tasks end to end.

This has implications far beyond software development. Agentic AI will change how customers interact with businesses. Instead of keyword-based searches or static workflows, experiences become conversational and goal-driven. In retail, for example, customers can describe what they want in natural language and get personalised results rather than long lists of products. The same applies to customer support, insurance claims, or service requests, where agents can proactively guide and resolve issues.

This is not limited to one sector. Across industries, agentic AI can handle routine tasks at scale, allowing humans to focus on judgement, empathy and complex decision-making. Productivity gains matter, but the bigger shift is the ability to reimagine customer experience itself, rather than simply making existing processes faster.

YS: Infrastructure and custom chips are less visible to customers, but how important are they in the AI era?

OK: Chips are a very important part of the equation, but they are only one part. What really matters is how everything comes together. That includes how data centres are designed, how cooling works, how energy is delivered, and how efficiently models run on top of that infrastructure. All of those factors affect cost, scale and sustainability.

We have been investing heavily in our own silicon with AWS Trainium, and we are now in multiple generations of that chip. The goal is to make AI workloads more cost-effective and more energy efficient. Most customers will never interact directly with the chip layer, but they benefit from lower costs and better performance as a result.

At the same time, infrastructure choices need to remain flexible. Some customers want the latest NVIDIA GPUs, and some want to run parts of their workloads on premises. That is why we focus on giving choices across the stack. Chips matter, but so do the model layer and the tooling that sits on top. All of it has to work together so customers can run AI reliably, securely and at scale, without being locked into a single approach.

YS: As AI systems become more autonomous, how should organisations think about security, governance and guardrails?

OK: The fundamentals are consistent across regions. What matters is having the right guardrails in place and designing AI workflows with governance built in from the start. As AI agents move towards autonomous or semi-autonomous execution, organisations need clear controls that determine when an agent can act on its own and when a human must be involved.

This is why evaluation and policy become critical. There are built-in evaluators and the ability to add custom evaluations so models and agents can be continuously assessed against defined criteria. That allows teams to design different classes of agents. A public-facing agent might be allowed to interact openly, while internal or regulated workflows require much stricter controls.

The key is flexibility without losing compliance. Some workflows benefit from full automation, others need human-in-the-loop oversight. The goal is not to slow adoption, but to make autonomous systems reliable, auditable and aligned with regulatory and business requirements. Governance is not a blocker to agentic AI. It is what makes large-scale production deployment possible.