Use AI To Reduce Toil - Not Layoff Employees (AWS Summit 2025)

The AWS Summit in NYC balanced product announcements with customer stories and demos, showing both AI's potential and its current practical applications. Despite being a leading hyperscaler, AWS took a realistic approach to enterprise AI adoption. Here are six key takeaways and their implications for customer experience:

1. Agentic AI Helps Enterprises Do Work No One Wanted To Do - Not Eliminate Jobs. "AWS's Erin Kraemer explained how developers use Amazon Q (their generative AI assistant) to handle tedious maintenance and upgrade work that often gets deprioritized. Amazon previously reported saving 4,500 developer years of work, though Kraemer emphasized this creates new value rather than eliminating jobs—the freed-up time allows developers to focus on higher-impact work, despite some tool errors requiring oversight.Accenture demonstrated similar efficiency gains with a B2B marketing client, reducing a 150-day process to 30 minutes while cutting manual tasks by one-third and dramatically improving time-to-market. Total cost savings reached 6%. Impact on the human experience: Employees will transition to higher-value work as AI handles routine tasks. Technology has always disrupted labor markets while creating new opportunities. "Hidden Figures" illustrates this pattern: women moved from manual calculations to operating NASA's first computers, demonstrating how technological change can elevate rather than replace human roles.
   

2. To Realize Value, Measure the Impact of AI on What Matters To Your Business. Accenture found that only 13% of surveyed executives achieved enterprise value from AI initiatives, with just 36% successfully scaling their projects. Key barriers included poor data quality, compliance issues, talent gaps, and complex integrations. As AWS's Erin Kraemer noted, "volume does not equal value"—a reminder that AI capability doesn't guarantee business impact. Natural language generation may sound human-like, but it doesn't ensure intelligent or accurate responses for customers. Impact on the human experience: My hotel routed calls from guest rooms to a natural language voice bot before quickly elevating to a human being.  The system was simultaneously friendly and useless to me as a caller, though it undoubtedly triaged calls and handled inquiries that would otherwise have required human attention. Online sources verified that this brand invested up to $2B in AI in 2024. Results include 50% call containment rate with voice AI (efficiency) and 85% CSAT (effectiveness) across all conversations. Natural language is not synonymous with effectiveness. (AWS was not the vendor.) Two personal and recent examples illustrate the bot's limitations: First, when I asked about the ice machine, it quickly stated "near room 61" but couldn't repeat the answer or provide directional context. A staff member immediately directed me to the machine steps from my room. Second, when I reported a smoke alarm going off—potentially signaling an emergency—the bot assumed I was asking about smoking policies and missed the urgent context entirely. In both cases, human staff resolved the issues efficiently.Intuit offered an enterprise example. Their AI agents bill more accurately and faster (efficiency). Their clients are also 10% more likely to pay in full (effectiveness).
   

3. Reliability is As Important As Accuracy. LLMs generate responses even when uncertain—it's how they're designed. However, users don't need perfection to find value. AWS's Erin Kraemer shared how developers embraced a tool with only 50% accuracy because it provided directional guidance for deeper research. The challenge lies in reliability. Agent performance varies daily, making enterprise scaling difficult without consistent high-quality data and robust model training. Business adoption will accelerate once AI systems reach approximately 90% accuracy—the threshold where human trust typically forms. (Paraphrasing AWS’ Rohit Prasad, SVP & Head Scientist for AGI) Impact on the human experience: There are two key elements here: trust and critical thinking. Trust requires transparency into AI decision-making and mechanisms to ensure unbiased, responsible outputs. Meanwhile, our skills must evolve from problem-solving to output evaluation—similar to how senior professionals already assess junior employees' work, but applied to AI-generated content.
   

4. Enterprises Will Realize Value With AI When They Imagine New Things. Following Conway's Law, enterprises build processes that mirror their communication structures. Today, employees primarily use AI agents for two purposes: gaining "superpowers" to accomplish previously impossible tasks, or adding "convenience" to existing workflows. We review AI output like managers evaluating junior staff work. In 2025 we’ll start to imagine new ways of working. According to Erin and others at AWS, AI’s augmentation of teamwork will be a substantial leap forward this year. AI native work processes will follow. All of this might sound mundane or progressive, but it ultimately will change how we work. Today, we consider the limitations of what humans can do when we design workflows - not the potentially unlimited abilities and capacity of machines.  SaaS solutions design interfaces to support humans with eyes and two hands - not machines with the ability to crawl indefinitely.  Humans will still train employees and AI agents. Humans leave organizations. Will AI agents? System Integrators echoed this sentiment. More than one said they were hired to automate existing processes. Almost always mid-project, clients would begin to imagine how they might work differently with agents. Impact on the human experience: Small teams of humans will be able to “do more” (not yet defined) and do new things if they can learn to work differently. Initially, employees may evaluate which of their high frequency tasks are best suited to agents and shift-, manage-, and review that work.  Individual contributors will become managers of perhaps large teams of agents almost immediately. Ultimately individuals may feel more akin to the conductor of an orchestra if they can become a master of a broader set of domains. 
   

5. Generative AI Won’t Always Be the Answer - Today. As one AWS executive said, “you need to find the right problems for the right technology at the right time to provide the right value.” GenAI will make sense when you need to scale human reasoning or creation, for example. AWS’ VP of Amazon Connect, Pasquale DeMaio (PQ),  offered a very grounded perspective, “Our focus is on practical AI implementation that delivers real customer value in context.” Applications don’t need to make up answers when they can look them up (e.g., balances, fees, dates). Use must always be balanced against the risks. PQ also emphasized that their focus is on solving tangible problems within the constraints of running a business, rather than pursuing AI for its own sake. Impact on the human experience: Know the phrase, “when you have a hammer, everything looks like a nail”? While AI’s abilities are exciting and mesmerizing, they are not always the right technology solution. The creation of value depends not only on the outcomes, but also on the costs and risks of achieving them. Regulated industries add another layer of complexity. Most of us won’t get out a calculator let alone a spreadsheet application to add two numbers together or calculate the tip on a check. They are simple, one-off tasks.  In July of 2025, I asked a tool how much it cost to send a single Email. Rather than offer me an answer, it conducted deep research and generated a 22-page report. A Google search would have been a better approach.
   

6. Time Between Competitive Advantage And Tablestakes is FAST. I’ll offer an example. In his opening keynote, Swami Sivasubramian initially focused on the challenge of “getting agents to scale” and detailed the new features of Amazon Bedrock AgentCore. He spent time reviewing the capabilities including: security, agent memory and ability to learn, agent identity and access controls, code generation and execution, agent’s ability to discover and connect with custom tools, observability, etc. I was absorbing the vast capabilities when he nonchalantly mentioned that many of these abilities are undifferentiated tools to do heavy-lifting. A year ago, agents seemed like just an idea. Impact on the human experience: Through interviews with a breadth of executives working on AI products, a common theme of “speed of change” is emerging. More than one product marketer told me, “if I don’t use or read about AI daily, I feel behind. It is likely drinking from a firehose … everyday.” Humans will need to adopt a curiosity and discipline to use tools daily. One executive even suggested that “tool usage” might be the most important metric today. 

Background: I had the opportunity to attend the Amazon Web Services (AWS) Summit in New York City on July 16, 2025. While there, I listened/spoke to AWS executives, spoke with partners, watched live demos from SI partners, and attended Swami Sivasubramanian’s keynote. I am especially grateful to Pasquale DeMaio who patiently answered my questions about the impact of AWS technology on customer experiences within the context of the much broader impact AI will make on businesses. I would also like to thank Tanya Shuckhart, Annie Weinberger, Paul Weiss and the entire AWS analyst relations team for their hospitality and the opportunity to learn so much about the technology and its applications.