CXO is being breached by constant changes in which AI time frames are compressed a few days before new developments take place. Breakneck Pace can freeze enterprise technology buyers because it cannot be spent on all new developments, must show returns and does not want to be debts for AI technology.
AWS Summit New York was to introduce a fundamental approach to providing businesses with the structure to employ AI agents.
Angie Ruan of NASDAQ's CTO Capital Access Platforms Division has put together the current AI situation. “The technology was running for over 10 years. If we didn't upgrade something after five years, we would have been late. It was later 18 months. Last year it was six months. Today, my way of thinking is five days ago, you're late, not knowing what's going on,” Ruan says. “I've never seen the pace as fast as AI.”
Luang added that there is a balanced act. “Be calm, strategic, agile and ready to take very practical delivery steps,” she said.
The practical real-world returns of AI projects – generations, agents, and everything in between were recurring themes of AWS Summit New York. AWS has deployed many updates and features, including many stories about Amazon Bedrock Agent Core, customizable NOVA models, and frameworks relating to reliability, security, observability and agility.
Eventually, Swami Sivasubramanian, AWS VP of Agent AI, returned to the fundamentals of the Enterprise using his keynote speech. The Shivas Bramanian speech in New York had a lot to do with the balance of innovation and basic approaches that can change models and underlying technologies. For AWS, a strong foundation and approach does not constrain it, but rather enables and accelerates innovation.
For businesses, focusing on fundamentals has been long been postponed. Do you really expect business to trade LLMs every time there is a newest best model that is 0.06% better in math, coding or reasoning?
“AI agents are structural changes. It's a change in the way software is deployed and operated, and the way software interacts with the world. Making it possible involves building on the foundations of today, Shivas Brahmanian.
That's a message that companies have accepted. “We're realistic about what AI can and can't do. This isn't a silver bullet, but that applies to all AI systems. Value comes when pairing with strong engineering practices.”
AWS has set up an AI agent production system that is sufficient for most use cases to have stability and models lasting for some time. As Agent AI is more enterprise-ready, there are fundamentals like identity, authentication, and stability.
“We've been working hard to get the most out of our business,” said Holger Mueller, analyst at Constellation Research.
“Amazon is stepping into the agent platform business with Agent Core. The good news for Amazon and its customers is that the traditional small-scale atomic services approach, derived from AWS DNA, could be accurate for companies to build their first agents. Vector's announcement is huge, as it makes digital assets stored in S3 available for AI to use.”
AI ROI mismatch
Rohit Prasad, SVP and Head Scientist at AGI at Amazon, said companies are struggling with the gap in expectations between AI deployment and actual returns.
“AI today is exciting, and the real world is the real benchmark in the end,” Prasad said. “We're hearing about these models that come out every day. If you're an enterprise CIO, you're thinking about real-world applications. How do you make real-world applications at scale?”
Prasad said the focus on AGI, a topic adjacent to obsessions in the AI industry, is often misdirected. “I want to level the set of AGIs. I think the whole conversation about whether to get to AGI first or get to it is pointless,” said Amazon chasing AGI and building a full-layer stack. “I don't think there's a switch when it comes to AGI. Let's focus on whether AI can be useful in real life. Can we simplify complex things?
AWS has announced the ability to customize NOVA models for enterprise use cases. AWS offers optimised recipes, distillation and customization of models that balance cost and performance.
Prasad noted that every company needs to think about AI returns from a workflow and process perspective. “It comes down to measurements. You can only improve what you measure. Look at the success criteria for all your workflows.”
He added that metrics cannot be stationary because your organization is constantly changing. “In many large applications, what you measure, what you measure every day, just go with a very open eye that it needs to evolve over a certain period of time,” Prasad said.
Measurement is important when it comes to AI agent values. Prasad said:
“The bars for evaluating agents must be the same as those used to evaluate humans in terms of safety reliability. I think that in a reliable human, you want it. It means it's a function of reliability, consistency and robustness. On the same bar.”
https://www.youtube.com/watch?v=7eksg8uyubs
Measure AI values
Erin Kraemer, Senior Principal Technical Product Manager at AWS Agent AI, said that AI could fundamentally change the way value is delivered.
problem? Most companies do not properly measure the impact of AI. “One of the mistakes I see is how we are measuring the impact of AI right now and how we talk about it,” says Kraemer. “We don't know if we're doing that the right way. Organizations that understand how to apply AI to meaningful issues and measure success are things that quickly adapt and position themselves for the future.”
Amazon's approach is to solve problems such as infrastructure scaling, product catalog management, and administrative tasks, with a focus on controlled input and continuous improvement.
Important takeouts:
When it comes to AI, we will focus on business outcome metrics on volume, she said. If there's too much conversation about AI volumes, it revolves around volume-based metrics, especially when it comes to code.
Kraemer said business achievements have a Trump volume. “I'm going to argue that value, not volume, should ultimately be an indicator of our success. So my mind, volume, it's the focus of the output, and perhaps not even the right outcome.”
Certainly, Kraemer said that the statistics that ridden her were generally written by AI. “The number is 30%. I really hate this number. It's a fundamentally flawed number. It tells me very little about what's going on, the systems, the customers,” Kraemer said.
Focus on bottlenecks. She said companies need to see AI through business outcomes. Specifically, AWS will consider AI to address process bottlenecks. “If the bottleneck tends to be centered around human reasoning, there's a reasonably good chance that AI is a well-placed solution to it,” says Kraemer.
Specifically, human bottlenecks have been an issue on Amazon throughout its history. She said:
“We love automation. We love streamlined processes. We have fairly large and complex systems to handle these processes, but it's humans who end up getting stuck. Human reasoning ability was a bottleneck. We literally couldn't build a system that could be expanded through a bottleneck.
Problem-specific metrics indicate actual value. Regarding code-related AI, Kraemer asked: “Would you fix the defect faster? Are you improving your security attitude?
Amazon is looking at AI through its customer experience. Look at the specific metrics AWS uses and measure AI returns.
Software Development:
- Defect resolution speed.
- Development speed.
- Reduce infrastructure costs. AWS saved “approximately $260 million on an AI-assisted Java upgrade.”
- Time savings for developers. AWS saved an estimated 4,500 developers years of effort with Java upgrades.
Customer Experience:
- Catalog quality improvements.
- The contact information for each order will be reduced.
- Customer satisfaction.
Knowledge work:
- We saved time using AI to answer over a million internal developer questions.
- A survey of decision times and decision times.
Amazon's approach to AI internally
The panel, representing technology leaders from various Amazon units in Amazon Ads, Alexa, Technology Infrastructure and other areas, spoke about how AI is integrated with products for success.
Here are some examples:
- Amazon Connect uses Genai to enhance customer engagement and automation with data context and entity resolution.
- AI generates images and videos of Amazon ads and its AI services.
- Amazon Business uses AI to automate business verification, improve accuracy and reduce manual review times. Search relevance and bulk purchase reviews are also designed to improve the sourcing experience for Amazon's business customers.
- AWS Marketplace uses AI to approve boarding and funding approvals to sellers, providing a comparison engine for product insights.
- Alexa is getting a more natural interaction and reconstruction for agent AI actions.
The metrics for these projects revolve around cost, friction elimination and customer experience. Maintain and iterate an experimental-based mindset focused on customer needs when deploying these key performance indicators and metrics.
“We're looking for a way to help you get started,” said Lak Palani, senior manager of product management technology at Amazon Business.
“My recommendation is simple. Don't use AI just to use it. Find the right business case where AI actually adds value. Small, measure the results and remember that it's an iterative process.
There is a way AWS meat and potatoes focus on Agent AI and the basics. Enterprise adoption of AI agents continues to talk about advances in technology and vendor marketing. AWS meets current customers.
