|
Hello Reader, Something shifted last week that most people in tech are not paying attention to. Anthropic CEO Dario Amodei, the same person who said AI would eliminate all coding jobs and potentially all white-collar and blue-collar work, came out and said we should slow down on AI. His stated reason is that Anthropic's models are getting so good they can improve themselves without human intervention. Do not confuse marketing with virtue. Anthropic has an IPO coming. What he is actually saying is: our models are the best, we are close to AGI, invest in us. The slowdown narrative is the pitch, not the concern. But something real is happening underneath the marketing. Companies are pulling back on GenAI spendingStarbucks pulled GenAI projects from their storefront applications. Uber reduced their GenAI budget. McDonald's did the same. The reason is not that GenAI does not work. The reason is that compute and electricity costs are the limiting factor, and the ROI on many use cases does not justify the spend. The output quality problem is also real. Anyone using GenAI heavily for coding has noticed it. The models are producing more output, but the valuable signal is harder to find inside it. The leading theory is that models are being trained on AI-generated content, which produces more AI-generated content, which gets used as training data again. The output is getting noisier because the training data is getting noisier. This is the same pattern that played out with blockchain. A few years ago everything was going to be transformed by blockchain. Property on the blockchain. Art on the blockchain. Then FTX. Then BlockFi. Then silence. The technology did not disappear. The hype did. And what remained were the legitimate use cases that actually needed it. The job market is respondingUS Department of Labor statistics from April and May show job creation far exceeding forecasts. In May, the US market added 170,000 jobs against a forecast of 80,000. Almost double. LinkedIn job postings are up. Recruiters are calling more people. Students are getting interview calls. Sam Altman, who predicted a job apocalypse from AI, has now walked that back. His current position is that jobs will be fine. The CEOs who were most aggressive about AI displacement are quietly revising their timelines. This is good news. Not because AI is going away, but because the hype normalizing means the real signal becomes clearer. What stays and what goesNot every use case needs GenAI. A chatbot that handles 50 FAQs does not need a frontier model. It does not even need RAG in most cases. You can feed those documents into a text-based database and query them directly. A data analytics workflow that runs SQL queries with group by and subqueries does not need GenAI. It needs well-written SQL. Companies are figuring this out and pulling GenAI out of use cases where it was never the right tool. What stays is the legitimate use cases. Complex reasoning tasks. Multi-step agentic workflows. Synthesis across large unstructured document sets. Personalization at scale. These use cases genuinely need GenAI and the investment is justified. What also stays is cost optimization as a discipline. As companies move from experimentation to production, the question shifts from "can we build this with GenAI" to "can we afford to run this with GenAI at scale." The engineers and architects who understand how to optimize inference cost, select the right model for each task, implement semantic caching, and design efficient RAG pipelines will be the ones companies keep and promote. What to do right nowTwo things matter more than anything else in this environment. First, master cloud fundamentals. Whatever GenAI is reduced to, whatever use cases survive the normalization, they will be built on top of existing data and applications. Those data and applications are either on the cloud or getting migrated. That is not changing. The demand for people who understand cloud architecture, system design, and infrastructure trade-offs is not going down. It is going up because the complexity of what gets built on top of cloud is increasing. Second, learn GenAI concepts, not tools. Tools change every quarter. Concepts do not. Understand what RAG actually is and when it is the right solution versus when a simple database query is better. Understand when to use MCP and when MCP overhead makes it the wrong choice. Understand when one agent is sufficient and when a multi-agent architecture is justified. Understand short-term versus long-term memory and when each matters. These decision points are becoming more valuable as the hype fades and real engineering judgment takes over. The people who win in this environment are not the ones who adopted every new tool the fastest. They are the ones who understood the fundamentals well enough to know which tools to use and when. The interview calls are coming. Make sure you are ready when yours does. Keep learning and keep rocking 🚀, Raj P.S. If you have found this newsletter helpful, and want to support me 🙏: Checkout my bestselling courses on AWS, System Design, Kubernetes, DevOps, and more: Max discounted links Checkout my YouTube channel for Cloud Gen AI tutorial and interview prep videos: Here AWS SA Bootcamp with Live Classes, Mock Interviews, Hands-On, Resume Improvement and more: https://www.sabootcamp.com/ |
Free Cloud Interview Guide to crush your next interview. Plus, real-world answers for cloud interviews, and system design from a top AWS Solutions Architect.
Hello Reader, There is a debate that plays out in almost every engineering team adopting cloud infrastructure. Containers or serverless. Kubernetes or Lambda. And almost every time, the person making the argument has already picked a side before the conversation starts. The person adopting serverless says containers are too complex. The person adopting containers says serverless lacks features. Both are partially right and both are missing the point. The real answer depends on five factors,...
Hello Reader, Every CEO in the last two years stood on a stage or got on an earnings call and said some version of the same thing: "AI is going to transform our operations. We are reducing headcount because AI will handle it." Two stories broke this week that every Solutions Architect and cloud professional needs to understand. Starbucks quietly shut down their AI inventory system after nine months. Deleted the blog post announcing it. No press release. Just an internal memo: go back to...
Hello Reader, Most AI agents built today have a fundamental flaw. They forget everything the moment a session ends. You tell the agent your preferences, your constraints, your context. You close the tab. You come back. It has no idea who you are. This is not a bug. It is the default state of every LLM and agent. They are stateless by design. And if you are building agents or going into SA interviews, understanding how memory works at a system design level is now a baseline expectation. Why...