Preethi CN

I just published some findings on Claude Sonnet 4's new 1M context window on Amazon Bedrock. TLDR - it's awesome. I did a benchmark with the LongCodeBench QA dataset (https://lnkd.in/ew4fF3VR) and results show consistent, strong reasoning results across different context lengths. The 5x expansion from 200K tokens opens up some real practical use cases - no more chunking for entire codebase analysis or agents with complex long-term memories, and processing hundreds of documents while maintaining full context. Details and benchmarking code are here: https://lnkd.in/eaTmVBD8

27

2 Comments

I love all the learning, networking, and, most importantly, community involvement that we get at San Diego Software Engineering meetup. Last night, Maxim Tarasov shared a ton of useful info both about both the fundamentals and limitations of AI-based code generation, as well as about how he uses AI tools to write software for a leading biotech startup in the San Diego area. A recent article in the Wall Street Journal highlighted the difficulty of finding good candidates in the age of AI-assisted resume writing and submission tools, as well as AI-based screening software. I can’t think of a better way for companies to find local, capable and driven people than to get involved with local communities. I continue to be fascinated and grateful for everyone who participates and contributes, and I hope to see many new faces at our future events. https://lnkd.in/gh55_H7C

53

11 Comments

At Amazon, we’re looking at ways to 𝐢𝐦𝐩𝐫𝐨𝐯𝐞 𝐛𝐮𝐢𝐥𝐝𝐞𝐫 𝐩𝐫𝐨𝐝𝐮𝐜𝐭𝐢𝐯𝐢𝐭𝐲 𝐡𝐨𝐥𝐢𝐬𝐭𝐢𝐜𝐚𝐥𝐥𝐲. This means not just code generation (via tech like Kiro) but where else do builders spend time? Excited that the great work of one of our siblings org, StoreGen, was showcased by Steven Tarcza II, at AWS reInvent this year. (Steve makes a good case for using the term "𝐝𝐞𝐯𝐞𝐥𝐨𝐩𝐞𝐫 𝐚𝐜𝐜𝐞𝐥𝐞𝐫𝐚𝐭𝐢𝐨𝐧" in one of his linkedin posts recently) The vision was to make Amazon developers 3x faster by building AI-native (not just AI-enhanced) solutions that are proactive, intent-driven, and deeply contextual. The team started experimenting in March, with the goal to develop tools by November. They tried lots of experiments. For some it wasn't quite the right time or they couldn't quite figure out the right approach. But a couple paid off big time for our organization. I use them both every day! 𝐒𝐩𝐞𝐜 𝐒𝐭𝐮𝐝𝐢𝐨 - In a company like Amazon, there’s millions of lines of code that were build five, ten, fifteen years ago, with the original owners long gone, and less than ideal pre-existing documentation – and they’re business-critical. How can you quickly understand a codebase when you need to make a change? Code-to-spec generation creates specifications from existing code, and enables semantic search across thousands of packages. I’ve personally found this tool invaluable to understand legacy code bases. A change to one of these systems would have taken me weeks before (mostly reading through the code to understand what it actually did), now it takes me hours. A key feature of SpecStudio is that for all claims it provides links back to the actual line of code. Having correct, up-to-date specs of what a product does opens all kinds of interesting doors: for me, personally, how much testing can we auto-generate from this knowledge? 𝐀𝐈 𝐓𝐞𝐚𝐦𝐦𝐚𝐭𝐞 – Is a proactive team member that monitors chat, documents, tickets, etc. continuously, maintaining a persistent team memory across systems. It autonomously suggests and takes actions: can generate tasks, implement features, and respond to code reviews. Right now it's vended as a slack bot, that lives in our team channels, and enthusiastically makes comments on our conversations. I find this idea particularly interesting. It's not a code generator, it's not a document reader, it's not a ticket-closer. It's a virtual teammate that you can interact with just like you would a real teammate, and partner with. If you're curious, watch Steve's walk at https://lnkd.in/g5z_UkWA

85

4 Comments

See why OCI leads the AI infrastructure market with RoCE-optimized networks, demonstrated scalability to over 32,000 GPUs, and ultralow latency at scale. https://lnkd.in/dnt_ksDb

1

Excited to share this new capability from Amazon Connect Tasks that treats AI more like a teammate than just a tool. We have eliminated the context gathering tax by having AI teammates summarize exactly what happened in the previous steps and suggest the next best action to resolve the issue. By letting the AI handle that initial heavy lifting of piecing together the history, the human teammate can jump straight into the actual resolution. 🚀 Saurabh Dey Sri Namburi Dipayan Chattopadhyay Naveen Narayan Marwan Bassyouni #AmazonConnect #AWS #AITeammate #FutureOfWork #CX #BackOfficeEfficiency https://lnkd.in/g8mKfQhn

81

1 Comment

In an interaction with CIO&Leader, Tirthankar Lahiri says enterprise AI must prioritize verifiability over speed, with intelligence embedded at the data layer, governed by strong security, deterministic controls, and human oversight—positioning AI as an assistant, not a decision-maker. https://lnkd.in/garGf4Tx

1

APM for systems is here: The power of New Relic's Transaction model with the cross application visibility of Distributed Tracing now gives you unprecedented insight into how the software systems powering your business critical features are performing -- all on a single pane of glass. https://lnkd.in/gJFfPsRX

40

2 Comments

This year, Amazon has created three new Agentic AI categories in the AWS AI Competency to recognize Partners who have demonstrated deep expertise in developing and implementing enterprise grade Agentic AI solutions. Of the 140,000 Partners in the AWS Partner Network (APN), they have recognized ONLY 60 companies this year and guess what? Automation Anywhere is one of them! What does this mean to an Enterprise? According to AWS data, companies working with AI Competency Partners are 25% faster in implementing Enterprise Solutions. And, data from an IDC report indicates that 23% of organizations expect to achieve full deployment of agentic AI technology in the next 12 months, with 65% of organizations expecting full deployment by 2027. So if you are a company looking to harness the latest advancements in the Agentic AI technology, and implement enterprise grade AI powered solutions, you should talk to us. Our new Agentic Process Automation (APA) Platform is a low-code / no-code way to build AI Agents that understands context, breaks down goals into sequence of tasks through reasoning, remembers past actions and uses the feedback to act autonomously with minimal human intervention. BTW, if you are at AWS re:Invent Conference this week, please stop by our booth at the Expo Hall to build hands-on AI Agents and experience it yourself! Read more here in this AWS Blog: https://lnkd.in/gcDZNHit Supratibh S. Chris Butler Chris Dally Victor Rojo Jay Bala Peter White Adi Kuruganti Patricia Draper Mary Quinn #AutomationAnywhere #AWS #AWSReInvent #AI #AgenticAI #Amazon

57

2 Comments

Amazon Web Services (AWS) just raised the bar for multicloud. AWS Interconnect for multicloud is generally available, the first purpose-built managed service for private, high-speed connectivity between clouds. No third-party fabric, no routing complexity, just a simple, intuitive interface with maximum resiliency, encrypted links, and a 99.99% SLA built in from day one. Flat-rate pricing with no per-GB charges and a simplified global tier structure that replaces the complexity of legacy cross-region data transfer models. Available today with Google Cloud across five regions, with Microsoft Azure arriving later this year. Over 150 enterprise customers validated this during public preview, the strongest signal yet that the market is ready for a fundamentally different approach to multicloud networking. Network complexity no longer dictates where you place workloads or how fast you move. Read more on how to get started: https://lnkd.in/gr7JxwkK #AWS #AWSInterconnect #Multicloud #CloudNetworking #HybridCloud #EnterpriseNetworking #DigitalTransformation

26

We recently announced the GA of Oracle Database@AWS. Customers can now run Oracle Exadata Database Service and Oracle Autonomous Database on dedicated infrastructure on OCI within the AWS U.S. East (N. Virginia) and U.S. West (Oregon) Regions. In addition, to meet growing customer demand, we plan to make the service available in an additional 20 AWS Regions around the world. InfoWorld has the details: https://lnkd.in/gswnvM9z

110

8 Comments

🚨 Amazon ECS finally supports built-in Blue/Green deployments 🚨 https://lnkd.in/gveS_4Cm Good news? Or a sign that AWS is slowly losing its edge? The concept of Blue/Green deployments was coined over 20 years ago and popularised in a book published 15 years ago. AWS launched ECS just 4 years later… and yet it’s taken 11 years for native support to arrive. To be fair, there have been workarounds—Load Balancer swapping, CodeDeploy integrations, third-party tools—but built-in support in 2025? In an era where platform teams are chasing progressive delivery, GitOps, and ephemeral environments, this feels... overdue. What do you think: 👉 Is this a “finally!” moment we should celebrate? 👉 Or is it a sign of lagging innovation in parts of AWS? Keen to hear your thoughts. Especially from those of you running ECS at scale. #AWS #ECS #DevOps #BlueGreen #CICD #CloudComputing #Docker

18

5 Comments

WireMock has indeed been a key enabler for secure, credentials-free CI testing of MultiCloudJ(https://lnkd.in/gaw8jqEa). The recently introduced gRPC extension further allows us to support and validate gRPC clients on Google Cloud using the same model.

6

Heading to Google Cloud Next '26 Looking forward to connecting with leaders shaping cloud, AI, platform, and operational excellence at scale - where execution, governance, and real outcomes matter. https://lnkd.in/giYvzFaZ #CloudTransformation #AILeadership #PlatformEngineering #OperationalExcellence #Leadership #GrowthAlmanac

6

4 Comments

I just read this.. Securing Amazon Bedrock cross-Region inference: Geographic and global | Artificial Intelligence We are excited to share our latest blog post on securing Amazon Bedrock cross-Region inference. This piece delves into the critical aspects of geographic and global security measures that enhance the integrity and performance of machine learning applications across different regions. By understanding and implementing these strategies, organizations can better protect their data while maximizing the capabilities of Amazon Bedrock. Read the full article to explore the key insights and best practices for secure cross-Region inference. You can find the post here: https://lnkd.in/d_9pXR24 #AWS

4

New announcement from AWS re:Invent: the launch of Graviton5, AWS's most advanced custom chip to date. 🚀 Here’s how Graviton5 stands out: • 25% higher performance compared to the previous generation • 192 cores per chip with 5x larger cache, improving performance for complex workloads like real-time gaming, high-performance databases, and more • Up to 15% higher network bandwidth and 20% higher Amazon EBS bandwidth across instance sizes, resulting in faster data transfers and improved performance for distributed applications • Energy efficiency that helps businesses meet sustainability goals, all while maintaining cutting-edge performance Graviton5 has already been a game-changer for companies like Adobe, Epic Games, and Pinterest, offering significant improvements in both cost and performance. Can't wait to see the impact this has on the industry!

15

Today is a big day for Amazon Redshift -- we are launching new Graviton-powered instances and a built-in query engine for data lakes with native support for Apache Iceberg and other open formats. The Graviton-powered RG instances are up to 2.4x faster than the previous generation (RA3) instances for your data lake and data warehouse workloads, at a 30% lower cost per vCPU. Perfect for today's world: large-scale analytics, agentic AI workflows, and cost-sensitive workloads that need answers fast. All available today in 25 AWS Regions! To learn more, read Channy Yun (윤석찬)'s new blog post at https://lnkd.in/g9Te6p33

416

1 Comment

The Google Cloud Next '26 Developer Keynote is a must-watch .. using the newly launched Gemini Enterprise Agent Platform, the keynote demonstrates how to build a real-world agentic application for a marathon in Las Vegas. My favorite segments: * Multi-agent system (Planner, Evaluator, and Simulator) utilizing Agent-to-Agent (A2A) protocol and an Agent Registry for discovery and seamless collaboration * Memory & Context: Agents can be made stateful with session management and a memory bank service. Additionally, developers can use Retrieval Augmented Generation (RAG), integrated with AlloyDB and Apache Spark, to provide agents with domain-specific knowledge. * Observability & Debugging: Using Gemini Cloud Assist, developers can perform natural language-based investigations to identify and fix errors, such as storage bottlenecks or latency issues, in real-time. * Security & Governance: Implement a zero-trust security model featuring Agent Gateways and unique agent identities. All code and project materials demonstrated during the keynote have been open-sourced to allow developers to reproduce the simulation and explore these features hands-on - check it out. https://lnkd.in/gidJCAxM

32

People often ask me about the differences in architecture between Amazon Dynamo (the 2007 SOSP paper), DynamoDB (the AWS serverless NoSQL database), and Aurora DSQL (the AWS serverless SQL databases). I captured the key points on durability, high availability, programming model, and trade-offs in my blog: https://lnkd.in/gkMG_tVD TL;DR: Despite the shared name, Dynamo and DynamoDB have little in common architecturally, and make very different trade-off decisions.

632

8 Comments